When we hit a segv, it's often the case that we might crash again in
the attempt to clean up. Instead we introduce a minimal restore callback
in the backend abstraction, that shuts down as simply as possible. Then
we can call that from the segv handler, and then to aid debugging, we
raise SIGTRAP in the segv handler. This lets us run gdb on weston from
a different vt, and if we tell gdb
(gdb) handle SIGSEGV nostop
gdb won't stop when the segv happens but let weston clean up and switch vt,
and then stop when SIGTRAP is raised.
It's also possible to just let gdb catch the segv, and then use sysrq+k
followed by manual vt switch to get back.
Parse the config file for [output] sections and check for 'name'
and 'mode' keys. The key strings are compared to what is reported
by weston log. The 'mode' key string can be one of the following:
1) WIDTHxHEIGHT - one that is reported by weston log
2) off - Disables the output
3) preferred - Uses the preferred mode
4) current - Uses the mode currently driving the crtc
If a connector is not attached to a crtc, weston fails to bring it up.
Typically kms fbdev drives all crtc and connectors, but if kms hasn't been
initialzed and weston is the first to set a mode of if a monitor is
hotplugged, we just fail to bring it up.
(krh: reformatted and edited a bit from original patch)
We no longer reserve the hw cursor for just the pointer sprite. Any
surface that satisfies the requirements for the hw cursor can be used.
Of course that's more or less always the pointer sprite, but at least now
we don't hard-code the references to compositor->seat anymore.
Determining whether it was possible to use a surface buffer as the
primary framebuffer was always a bit of a hack. Just before rendering,
we look at the second top-most surface to see if it's the right size and
position and then use it if it is. If we have a hw cursor and a drm
plane on top, it's no longer the second top-most and it wouldn't be picked
even if it could work.
With this change, we inspect surfaces in drm_assign_planes and try to find
a candidate for the primary fb there. We track the overlap like for
overlays and if there is a surface that's not overlapped and fills the
output we pick it.
This commit also consolidate the logic to track damage as we move a surface
in and out of the primary plane, which was missing in the drm plane case.
The kms planes support scaling, so try to detect transformations that
are just translations + scaling and program the kms plane accordingly.
In particular, this lets us fullscreen a yuv surfaces with the scale method
and have the compositor use a kms plane for scaling and color conversion.
This lets us create the gbm bo from the wl_buffer instead of the
EGLImage. With planar buffer, the EGLImage corresponds to a plane
and not the entire planar buffer.
More importantly, with some YUV formats (packed YUYV and similar) we use
several EGLImages to sample the different components, but KMS only
want one plane in that case.
Make weston_surface::texture and ::surface an array, while keeping
[0] for RGB surfaces.
Signed-off-by: Gwenole Beauchesne <gwenole.beauchesne@intel.com>
We now pick the driver preferred mode for our initial mode. If no preferred
mode is available we default to the current mode. We also have a command
line option now to keep the current mode if it differs from the preferred.
This commit also drops the built-in mode and insteade uses the current mode
if the connector doesn't report any modes.
When an unused plane is disabled, the destroy listener for a previously
used buffer needs to be removed. This fixes a crash when an overlay
would be reenabled using the same buffer as before, causing the destroy
listener to be inserted twice.
Besides the fact of the frame not being done, assigning planes before
the vblank handler is called will make the state inconsistent, leading
to a crash.
Check if the output passed to drm_output_prepare_overlay_surface() does
actually contain the surface and fail if it doesn't. Also fail if a
surface spans multiple outputs since clearing the damage will cause the
portion that is not in an overlay to not be updated at all.
This way, if initialisation fails (say, udev or Mesa are broken, or we
couldn't find any devices), we'll at least take you back to where you
were, rather than leave you at a totally broken VT you can't escape
from.
Signed-off-by: Daniel Stone <daniel@fooishbar.org>
weston_compositor_init is always called late because most
implementations can't initialise GL until fairly late in the game.
Split it into a base version with the same name, followed by
weston_compositor_init_gl which can be called later on.
This simplifies compositor-wayland, which no longer needs a separate
global handler just for wl_seat.
Signed-off-by: Daniel Stone <daniel@fooishbar.org>
We start tracking which hardware plane a surface is displayed on, which
lets us avoid generating damage when a hardware overlay/cursor is moved
around.
EGLDisplay is helpfully typedeffed as void *, which means that you won't
get conflicting-pointer-type warnings if you accidentally confuse it
with weston_compositor::wl_display. Rename it to make it more clear
which display you're dealing with, and also rename compositor-wayland's
parent.display member to parent.wl_display.
Signed-off-by: Daniel Stone <daniel@fooishbar.org>
wl_input_device has been both renamed and split. wl_seat is now a
virtual object representing a group of logically related input devices
with related focus.
It now only generates one event: to let clients know that it has new
capabilities. It takes requests which hand back objects for the
wl_pointer, wl_keyboard and wl_touch interfaces it exposes which all
provide the old input interface, just under different names.
This commit tracks these changes in weston and the clients, as well as
similar renames (e.g. weston_input_device -> weston_seat). Some other
changes were necessary, e.g. renaming the name for the visible mouse
sprite from 'pointer' to 'cursor' so as to not conflict.
For simplicity, every seat is always exposed with all three interfaces,
although this will change as time goes on.
Signed-off-by: Daniel Stone <daniel@fooishbar.org>
Currently, the drm backend will create and destroy a KMS FB for each
frame. However, the bos for a gbm surface are reused (at least with
mesa) so we can store the fb_id on it and destroy it only on the bo's
destroy callback.
To use the same path for scanning out client buffers, some refactor
was needed. Previously, the bo for the client buffer was destroyed
early so that gbm_surface_release_buffer() would not be called with
it, since at the page flip handler output->scanout_buffer can be
NULL even if the current frame is a client buffer.
This was solved by adding a drm_fb structure that holds a gbm_bo,
an fb_id, and information about the fb coming from a client buffer
or not. A drm_fb is created in such a way that it is destroyed
whenever the bo it references is destroyed. The fields current_*
and next_* in drm_output are changed into only two pointers to
drm_fb's.
Switching display mode may happen when:
1. The fullscreen surface is at top most in fullscreen layer and with
"driver" method. Shell will switch output mode to match the surface
size. If no matched mode found, fall back to "fill" method.
2. The top fullscreen surface is destroyed or unset. Switch back to the
origin mode.
Scott Moreau
Kristian Høgsberg
Wang Quanxian
Gwenole Beauchesne
Ander Conselvan de Oliveira
Daniel Stone
Martin Minarik
Ander Conselvan de Oliveira
Alex Wu