Dispmanx is the prorietary display API on the Raspberry Pi, which
provides hardware compositing. Every visible surface is assigned a
Dispmanx element, and the hardware or firmware will do all compositing
onto screen. The API supports translation, scaling, flips, discrete
rotations in 90-degree steps, alpha channel on the surfaces, and
full-surface alpha on top.
Previously, Dispmanx capabilities were used via the weston_plane
mechanism, where surfaces were assigned to planes when possible, and
otherwise transparently falling back to GLESv2 compositing. Because we
have no way to use the same memory buffer as a GL texture and a Dispmanx
resource, we had to prepare for both. In the worst case, that means one GL
texture, and two (double-buffered case) Dispmanx resources, all the size
of a whole surface, for all surfaces. This was eating memory fast. To
make things worse (and less slow), the wl_shm buffer was kept around,
since it was copied to either a texture or a resource as needed. This
caused all clients to need two buffers. In a Dispmanx-only renderer, we
can drop the GL texture, and we can release wl_shm buffer immediately
after the first copy, so clients become effectively single-buffered. So
from the worst case of 5 buffers per surface, we go down to 3 or just
2 (single-buffered Dispmanx element, one wl_shm buffer in the client)
buffers per surface.
As this will replace the GL renderer on rpi, we cannot fall back to the
GLESv2 compositing anymore. We lose arbitrary surface rotation, but we
lose also the GL fallback, which caused glitches.
This patch depends on new RaspberryPi firmware. Older firmware may not
render ARGB surfaces correctly, solid color surfaces maybe cause a
performance hit, and the output may completely fail in case the firmware
does not fall back internal off-line compositing properly as needed.
This new rpi-renderer support surface translation and scaling, but not
rotation or transpose (not even in 90-deg steps). In theory, 90-deg step
surface rotation is possible to support. Output transformations are
supported, but flipped variants do not seem to work right.
As a detail, menus and other surfaces that are simply translated with
respect to another surface caused falling back to the GL renderer. The
rpi-renderer handles them directly.
This patch only adds the new renderer, but does not hook it up into use.
Signed-off-by: Pekka Paalanen <pekka.paalanen@collabora.co.uk>
Both GL and pixman renderer (pixman probably only because GL did?)
return the screen capture image as y-flipped, therefore Weston y-flips
it again. However, the future rpi-renderer can produce only right-way-up
(non-flipped) screen captures, and does not need an y-flip.
Add a capability flag for y-flip, which the rpi-renderer will not set,
to get screen captures the right way up.
The wcap recording code needs yet another temporary buffer for the
non-flipped case, since the WCAP format is flipped, and the code
normally overwrites the input image as it compresses it. This becomes
difficult, if the compressor is supposed to flip while processing.
Signed-off-by: Pekka Paalanen <pekka.paalanen@collabora.co.uk>
The upcoming rpi-renderer cannot handle arbitrary rotations. Introduce
Weston capability bits, and add a bit for arbitrary rotation. GL and
Pixman renderers support it.
Shell or any other module must not produce surface transformations with
rotation, if the capability bit is not set. Do not register the surface
rotation binding in desktop shell, if arbitary rotation is not
supported.
Signed-off-by: Pekka Paalanen <pekka.paalanen@collabora.co.uk>
The array of arguments supplied to execv must be NULL terminated. If
unitialized values are used as pointers the exec call may fail with a
EFAULT error ("Bad address").
https://bugs.freedesktop.org/show_bug.cgi?id=64874
If you specify e.g. scale=2 in an output section in weston.ini
we scale all modes by that factor.
We also correctly scale cursor positioning, but ATM there is no
scaling of the cursor sprite itself.
If you specify e.g. scale=2 in weston.ini an output section for the
X11 backend we automatically upscale all normal surfaces by this
amount. Additionally we respect a buffer_scale set on the buffer to
mean that the buffer is already in a scaled form.
This works with both the gl and the pixman renderer. The non-X
backends compile and work, but don't support changing the output
scale (they do downscale as needed due to buffer_scale though).
This also sends the new "scale" and "done" events on wl_output,
making clients aware of the scale.
Rather than storing the shadow_image in the untransformed space
and rotating on copy to hw_buffer we store both on the transformed
space. This means a copy between them is a straight copy, and that
apps supplying correctly transformed surface buffers need not
change them.
We also correctly handle all output transform including the previously
unhandled flipped ones, as well as client supplied buffer_transforms (which
were previously ignored).
We also simplify the actual rendering by just converting any damage
region to output coordinates and set it on a clip and composite
the whole buffer, letting pixman do the rectangle handling. This
means we always do all the transforms, including the surface positioning
as a pixman_image transform. This simplifies the code and sets us up
for handling scaling at a later stage.
The transform looks complicated, but in practice it ends up being
an integer translation almost always, so it will hit the pixman
fastpaths.
We pick the window scale/tranform based on what the output uses, which means
we can avoid rotations in the compositor, and get sharper rendering
in scaled outputs.
We used to just store the buffer size here which is not right if the
surface has a buffer_transform or a buffer_scale. To fix this we pass
the transform and scale into the toysurface prepare and swap calls and
move both the surface to buffer and the buffer to surface size
conversion there.
Without this interactive resize on the top or left sides of a transformed
or scaled surface will not work correctly.
Apparently some compilers complain about set but not used variables
'available' and 'bufs', but I don't get the warning. Still, separate the
debugging code from shm_surface_buffer_release(), so that we only
compute 'bufs' when it is printed. This should fix the warnings.
The debugging code now prints the shm_surface buffer state before and
after, instead of just after.
Signed-off-by: Pekka Paalanen <pekka.paalanen@collabora.co.uk>
The shell_grab_start function sets up a destroy notification on the
shell surface such that when the shell surface is destroyed the pointer
on the grab to the shell surface is set to NULL.
We must therefore check whether the shell surface is NULL and end the
grab if it is.
https://bugs.freedesktop.org/show_bug.cgi?id=64689
Mention, that sub-surfaces are not clipped to the parent.
Be more accurate on surface commit vs. apply state.
Mention the initial stacking order.
Signed-off-by: Pekka Paalanen <pekka.paalanen@collabora.co.uk>
This patch fixes a crash with the surface_pong when one of the
seats doesn't have a pointer. This was the case with the RDP compositor
that use a fake seat with no mouse or keyboard.
This patch brings back the user environment from the shell.
In the future, weston-launch could create the Wayland socket earlier, in
which case the user's shell could be used to run Wayland-specific tools
in the new Weston session.
Signed-off-by: Quentin Glidic <sardemff7+git@sardemff7.net>
The only way to create a sub-surface loop by recursive nesting is to
make the main surface (which does not have a role) a sub-surface of any
of its sub-surfaces. All other cases should already be cought.
This change checks for that exact case, and sends a protocol error.
Signed-off-by: Pekka Paalanen <pekka.paalanen@collabora.co.uk>
It should not be possible to create a loop by nesting sub-surfaces.
Currently Weston fails this test.
Signed-off-by: Pekka Paalanen <pekka.paalanen@collabora.co.uk>
wl_subsurface.set_desync should apply the cached wl_surface state.
Otherwise, the sub-surface may be stuck: a commit on the parent surface,
if desynchronized, will not commit the sub-surface because it is
desynchronized, too. A commit on the sub-surface may not happen, if it
is waiting for the frame callback from the previous commit.
Signed-off-by: Pekka Paalanen <ppaalanen@gmail.com>
Add a test for varying the object destruction order in a complex
sub-surface tree.
This test attemps to fuzz the destruction of a sub-surface tree to make
sure the server does not crash on any wl_surface or wl_subsurface
destruction sequence.
Signed-off-by: Pekka Paalanen <pekka.paalanen@collabora.co.uk>
surface-global-test and surface-test did not get updated to
the new module_init(...) signature when it changed in
a50e6e4c50. Thus, they
failed to compile. Simply running 'make check' shows the
problem. This patch fixes it.
fixes https://bugs.freedesktop.org/show_bug.cgi?id=64691
Signed-off-by: U. Artie Eoff <ullysses.a.eoff@intel.com>
The subsurface patches changed the center_on_output() behavior a bit.
Instead of using the buffer width and height, it now looks at surface
geometry. In lock_surface_configure() we haven't set up the geometry
when we call center_on_output() so the lock surface would never show
up.
Just the same as it is done in shell.c, if the input method process exits
for any reason, we relaunch it automatically, as it is not possible to
launch a standalone application outside of the weston process.
In v2:
- Proper error message when giving up.
Signed-off-by: Eduardo Lima (Etrunko) <eduardo.lima@intel.com>
This set of changes adds support for searching for a given config file
in the directories listed in $XDG_CONFIG_DIRS if it wasn't found in
$XDG_CONFIG_HOME or ~/.config. This allows packages to install custom
config files in /etc/xdg/weston, for example, thus allowing them to
avoid dealing with home directories.
To avoid a TOCTOU race the config file is actually open()ed during the
search. Its file descriptor is returned and stored in the compositor
for later use when performing subsequent config file parses.
Signed-off-by: Ossama Othman <ossama.othman@intel.com>
The function drm_output_start_repaint_loop() unconditionally issues a
page flip, even if the crtc for that output has not been enabled yet.
That causes the page flip to fail, and drm_output_repaint() is never
called.
Solve this by bypassing the initial page flip if the output needs a
mode set.
This has the caveat of affecting latency predictability for that first
frame and when a "driver" mode fullscreen surface causes a mode set.
However, on both cases the mode set would take an unpredictable amount
of time anyway.
https://bugs.freedesktop.org/show_bug.cgi?id=63812https://bugs.freedesktop.org/show_bug.cgi?id=64183
This allows users to change the assigned display profile in GNOME (using
gnome-control-center) or KDE (using colord-kde) and also allows the profiling
tools to correctly inhibit the calibration state whilst measuring the native
screen response.
The subsurface-server-protocol.h header should not be included
by any headers that are part of the SDK since it is not exported.
Otherwise, SDK consumers will break during compilation.
Move this include from compositor.h to compositor.c.
Fixes https://bugs.freedesktop.org/show_bug.cgi?id=64537
Signed-off-by: U. Artie Eoff <ullysses.a.eoff@intel.com>
In case a toytoolkit application manages to schedule resizes constantly,
throttle them to the main surface display.
When resizing, all surfaces are updated synchronously, so it also makes
sense to synchronize on the main surface's frame callback particularly.
Rendering any faster will not make sense.
Signed-off-by: Pekka Paalanen <ppaalanen@gmail.com>
Mesa's eglSwapBuffers() waits for the frame event from the previous
swapBuffers, before it returns. Apparently eglSwapInterval(), which
should be able to disable the wait, is unimplemented for now.
When a sub-surface contains an EGL widget, and the commit mode is
synchronized, the frame events will not be delivered to EGL until the
parent surface gets committed. Therefore rendering the EGL widget twice
would lead to a deadlock.
When the window is being resized, we need to force a repaint of the EGL
widget, too, to make the whole window consistent. For that, we need to
make sure the frame event from the previous eglSwapBuffers() actually
arrives.
This patch adds an extra wl_surface.commit(parent), when the window is
being resized, which should guarantee, that the previous eglSwapBuffers
gets its event.
To properly handle an EGL widget in a sub-surface, running in its own
thread, the EGL widget's automatic updates should be paused before
sending the extra wl_surface.commit(parent). A natural place for the
pause would be in the widget's resize hook. However, wl_surface.commit
cannot be called right after resize hooks, because it would commit new,
incomplete surface state. Therefore this patch is not enough for
threaded toytoolkit applications. Luckily those do not exist yet.
When eglSwapInterval() gets implemented, this patch should be reverted.
Signed-off-by: Pekka Paalanen <ppaalanen@gmail.com>
Add a demo program with:
- a main surface (green)
- a Cairo-image sub-surface (red)
- a raw GLESv2 widget (triangle)
Sub-surface input region is set empty to avoid problems in toytoolkit.
If Cairo links to libGL, then we will end up with also libGLESv2 linked
to subsurfaces program, and both libs getting really used, which leads
to disaster.
Do not build subsurfaces demo, if Cairo links to libGL and cairo-egl is
usable.
The GL rendering loop is not tied to the toytoolkit or the widget, but
runs directly from its own frame callback. Therefore it runs
independent of the rest of the application. This also relies on one of
two things:
- eglSwapInterval(0) is implemented, and therefore eglSwapBuffers never
blocks indefinitely, or
- toytoolkit has a workaround, that guarantees that eglSwapBuffers will
return soon, when we force a repaint on resize.
Otherwise the demo will deadlock.
The code is separated into three sections:
1. The library component, using only EGL, GLESv2, and libwayland-client
APIs, and not aware of any toolkit details of the parent application.
This runs independently until the parent application tells otherwise.
2. The glue code: a toytoolkit application widget, who has its own
rendering machinery.
3. The application written in toytoolkit.
This patch also adds new toytoolkit interfaces:
- widget_get_wl_surface()
- widget_get_last_time()
- widget_input_region_add()
Toytoolkit applications have not had a possibility to change the input
region. The frame widget (decorations) set the input region on its own
when used, otherwise the default input region of everything has been
used. If a window does not have a frame widget, it can now use
widget_input_region_add() to set a custom input region.
These are not window methods, because a widget may lie on a different
wl_surface (sub-surface) than the window.
Changes in v3:
- replace set_commit_mode with set_sync and set_desync
Signed-off-by: Pekka Paalanen <ppaalanen@gmail.com>
Add redraw_needed flag to all surfaces, in addition to having one in
window. The window redraw_needed flag is changed to force a redraw of
the whole window, regardless of frame events.
widget_schedule_redraw() now schedules the redraw only for the surface,
where the widget is on. window_schedule_redraw() is equivalent to
scheduling a redraw for all (sub-)surfaces of the window.
We still use only one deferred task for all redraws.
surface_redraw() will skip the redraw, if the window does not force a
redraw and the surface does not need a redraw. It will also skip the
redraw, if the frame callback from the previous redraw has not triggered
yet. When the frame callback later arrives, the redraw task will be
scheduled, if the surface still needs a redraw.
If the window forces a redraw, the redraw is executed even if there is a
pending frame callback. This is for resizing: resizing should trigger a
window repaint, as it really wants to update all surfaces in one go, to
apply possible sub-surface size and position changes. Resizing is the
only thing that makes a window force a redraw.
With this change, subsurfaces demo can avoid repainting the cairo
sub-surface while still animating the GL sub-surface.
Signed-off-by: Pekka Paalanen <ppaalanen@gmail.com>
The new application API window_add_subsurface() will create a plain
widget that is on a new sub-surface.
The sub-surface position is taken from the surface's root widget
allocation. This way widget allocations are always in the main surface
(i.e. window) coordinates. However, Cairo drawing coordinates will now
be different to widget coordinates for sub-surfaces. Cairo coordinates
are fixed by applying a translation in widget_cairo_create(), so that
widget drawing code can simply use the widget allocation as before.
Sub-surfaces are hooked up into resize, window flush, redraw, and
find_widget. Window maintains a list of sub-surfaces in top-first order.
Add a client settable default commit mode, and toggle the mode when
resizing to guarantee in-sync updates of a window and its sub-surfaces.
Changes in v3:
- replaced set_commit_mode with set_sync and set_desync
Signed-off-by: Pekka Paalanen <ppaalanen@gmail.com>
Increase the maximum number of shm "leaves" to three, and rewrite the
leaf release and pick algorithms. The new algorithms hopefully improve
on buffer re-use while freeing unused buffers.
The goal of the new release algorithm is to always leave one free leaf
with storage allocated, so that the next redraw could start straight on
it.
The new leaf picking algorithm will prefer a free leaf that already has
some storage allocated, instead of just picking the first free leaf that
may need to allocate a new buffer.
Triple-buffering is especially for sub-surfaces, where the compositor
may have one wl_buffer busy on screen, and another wl_buffer busy in the
sub-surface cached state due to the synchronized commit mode. To be
able to forcibly repaint at that situation for e.g. resize, we need a
third buffer.
Signed-off-by: Pekka Paalanen <ppaalanen@gmail.com>
Pekka Paalanen
Ander Conselvan de Oliveira
Alexander Larsson
Pekka Paalanen
Kristian Høgsberg
Mun Gwan-gyeong
Rob Bradford
Hardening
Quentin Glidic
U. Artie Eoff
Dima Ryazanov
Richard Hughes
Eduardo Lima (Etrunko)
Ossama Othman