Right now the Vulkan renderer blocks until the frame is complete
after rendering. This is necessary because Vulkan doesn't
interoperate well with implicit sync we use everywhere else.
Use the new kernel API to import a sync_file into a DMA-BUF to
avoid blocking.
We need to wait for the pending command buffer to complete before
re-using stage buffers. Otherwise we'll overwrite the stage buffer
with new contents before the texture is fully uploaded.
We need to wait for any pending command buffer to complete before
we're able to fully destroy a struct wlr_vk_texture: the Vulkan
spec requires the VkDescriptorSet to be kept alive.
So far we've done this in vulkan_end(), after blocking until the
command buffer completes. We'll soon stop blocking, so move this
logic in get_command_buffer(), where we check which commands buffers
have completed in a non-blocking fashion.
vkCmdCopyBufferToImage requires that the buffer offset be a multiple
of the texel block size, which for single plane uncompressed formats
is the same as the number of bytes per pixel. This commit adds an
alignment parameter to vulkan_get_stage_span which ensures that the
provided span (and the sequence of image copy operations derived which
use it) have this alignment.
0xFFFFFFFF milliseconds is 4,294,967,295 ms so about 50 days.
A little bit too close for comfort.
Use int64_t instead of uint64_t to avoid C's implicit conversion
footguns in computations.
This is a first step towards moving texture uploading out of
wlr_compositor.
This commit allows compositors to opt-out of the texture uploading
by passing a NULL wlr_renderer. An immediate user of this is
gamescope, which currently implements a stub wlr_renderer just to
make wlr_compositor happy.
Let's just forward-declare struct wlr_backend instead.
We need to fixup the Vulkan renderer: it needs makedev(), which
got included by chance via <wlr/backend.h> → <wlr/backend/session.h>
→ <libudev.h>.
This can be used to know when wlr_xwayland_server decides to start
a new Xwayland process. At that point the wl_client has already
been created but the Xwayland process hasn't been started yet.
Up until now, wlr_backend_autocreate() created the wlr_session and
then stuffed it into struct wlr_multi_backend so that compositors
can grab it later.
This is an abuse of wlr_multi_backend and the wlr_backend API:
wlr_backend_get_session() and wlr_multi_backend.session only exist
to accomodate the needs of wlr_backend_autocreate(). What's more,
the DRM and libinput backends don't implement
wlr_backend_impl.get_session.
Instead, return the struct wlr_session to the compositor in the
wlr_backend_autocreate() call. wlr_backend_get_session() will be
removed in the next commit.
This is a re-implementation of wl_shm. The motivations for using
this over the one shipped in libwayland are:
- Properly handle SIGBUS when accessing a wl_buffer's underlying
data after the wl_buffer protocol object has been destroyed.
With the current code, we just crash if the client does that
and then shrinks the backing file.
- No need to fight the wl_shm_buffer API anymore. This was awkward
because we weren't notified when clients created a wl_shm buffer,
and this doesn't play well with our wlr_buffer abstraction.
- Access to the underlying FD. This makes it possible to forward
the wl_shm buffer to a parent compositor with the Wayland/X11
backends.
- Better stride checks. We can use our format table to ensure that
the stride is consistent with the bpp and width.
We'll use this function from wlr_shm too.
Add some assertions, use int32_t (since the wire protocol uses that,
and we don't want to use 16-bit integers on exotic systems) and
switch the stride check to be overflow-safe.
Call glGetGraphicsResetStatusKHR in wlr_renderer_begin to figure
out when a GPU reset occurs. Destroy the renderer when this
happens (the OpenGL context is defunct).
The concept of a persistent accumulated surface offset is wrong
from a protocol point-of-view. wl_surface.offset is tied to a
commit, its interpretation depends on the surface role.
For example, with the following sequence:
wl_surface@1.offset(1, 1)
wl_surface@1.commit()
wl_pointer@2.set_cursor(wl_surface@1, 42, 42)
The final cursor hotspot is (42, 42): the commit which happened
before the set_cursor request has no impact on the hotspot
computation.
The wlr_output_cursor logic already uses wlr_surface.current.{dx,dy}.
wlr_scene's drag icon doesn't, update it accordingly.
Allow to get whether has alpha channel of the VkImage, it can help an
optimization to disable blending when the texture doesn't have alpha.
Because the VkFormat isn't enough because it's always set to
VK_FORMAT_B8G8R8A8_SRGB or VK_FORMAT_R8G8B8A8_SRGB.
This allows compositors to indicate which features they support,
and is required to eventually make this API stable.
References: https://github.com/swaywm/sway/issues/7260
32daa43a45 has removed the asymmetry in
the relationship of a wlr_surface and an unmapped wlr_xwayland_surface,
when wlr_surface.role_data wasn't NULL but wlr_xwayland_surface.surface
was. However, this also means that
wlr_xwayland_surface_from_wlr_surface() now returns NULL if the
wlr_surface is unmapped. Fix the documentation to reflect this.
Before re-using a VkCommandBuffer, we need to wait for its
operations to complete. Right now we unconditionally wait for
rendering to complete in vulkan_end(), however we have plans to
fix this [1]. To fully avoid blocking, we need to handle multiple
command buffers in flight at the same time (e.g. for multi-output,
or for rendering followed by texture uploads).
Implement a pool of command buffers. When we need to render, we
pick a command buffer from the pool which has completed its
operations. If we don't find one, try to allocate a new command
buffer. If we don't have slots in the pool anymore, block like we
did before.
[1]: https://gitlab.freedesktop.org/wlroots/wlroots/-/merge_requests/3574
Up until now we were using a VkFence for GPU-to-CPU
synchronization. This has some limitations that become a blocker
when trying to have multiple command buffers in flight at once
(e.g. for multi-output). It's desirable to implement a command
buffer pool [1], but VkFence cannot be used to track command buffer
completion for individual subpasses.
Let's just switch to timeline semaphores [2], which fix this issue,
make synchronization a lot more ergonomic and are a core Vulkan 1.2
feature.
[1]: https://gitlab.freedesktop.org/wlroots/wlroots/-/merge_requests/3802
[2]: https://www.khronos.org/blog/vulkan-timeline-semaphores
Added wlr_vk_renderer_get_* functions to allow get the VkInstance,
VkPhysicalDevice, VkDevice, queue family of a wlr_vk_renderer.
Added wlr_vk_renderer_get_current_image_attribs function to allow get
the VkImage of current renderer buffer to use on compositors.
Added wlr_renderer_is_vk function, it's like the wlr_renderer_is_gles2,
returns true if the wlr_renderer is a wlr_vk_renderer.
Added wlr_vk_image_get_attribs function to get a VkImage and it's
extras information (e.g. a VkImageLayout and VkImageFormat of the
VkImage) from a wlr_texture.
In wlroots we add comments near struct wl_list members to indicate
which type it's linked to. The Vulkan renderer had some comments
with mistakes, and some members without a comment.
We were crashing in the error codepath [1] when
wlr_drm_create_lease() fails.
To fix this, delay the creation of the wlr_drm_lease_v1 until the
request is granted. Previously we were allocating that struct early
without populating the drm_lease field. However that means we ended
up with a half-constructed struct in the error codepath which is
annoying to handle.
[1]: https://github.com/swaywm/sway/issues/7204#issuecomment-1269797356
This field becomes stale too easily: for instance, see 6adca4089c
("backend/drm: don't unconditionally set desired_enabled").
Additionally, drm_connector_alloc_crtc() needs to do some weird
dance, restoring its previous value.
Instead, add a connector arg to realloc_crtcs() to indicate a new
connector we want to enable.
When starting up, the compositor might call wlr_output_set_mode()
with a mode which is already the current one. wlroots will detect
this and make the wlr_output_set_mode() call a no-op. During the
next wlr_output_commit() call, wlroots will perform an atomic
commit without the ALLOW_MODESET flag.
This is an issue, because some drivers need ALLOW_MODESET even if
the mode is the same. For instance, if the FB stride or modifier
changed, some drivers require a modeset.
Add a new flag "allow_artifacts" which is set when the compositor
calls mode-setting functions. Use this flag to figure out whether
we want to perform atomic commits with ALLOW_MODESET.
(The name "allow_artifacts" is picked because ALLOW_MODESET is a
misnomer, see [1].)
[1]: https://patchwork.freedesktop.org/patch/505107/
Closes: https://gitlab.freedesktop.org/wlroots/wlroots/-/issues/3499
Instead of using low-level wl_shm_buffer and wlr_dmabuf_v1_buffer
APIs, use the unified wlr_buffer APIs. That way it doesn't matter
what the exact wlr_buffer implementation is used, any which provides
the necessary capabilities (data_ptr or dmabuf) would work.
Simplifies the logic a bit, and will make the transition to wlr_shm
easier.
CTA-861-H defines a picture aspect ratio which may be attached to
each mode. This affects the way the sink will display the image.
See annexes H.1 and H.2 for examples.
See the spec at [1]. tl;dr EGL has terrible defaults: eglTerminate()
may have side-effects on completely unrelated EGLDisplay objects.
This extension allows us to opt-in to get the sane behavior:
eglTerminate() only free's our own EGLDisplay without affecting
others.
[1]: https://registry.khronos.org/EGL/extensions/KHR/EGL_KHR_display_reference.txt
wlr_buffer.c is difficult to read because it contains a mixed bag
of unrelated things: base buffer type, buffer implementations,
buffer resource factory, and client buffer.
Split each of these into their own file.
dac040f87f mistakenly renamed
xdg_surface_destroy listener, which was listening to *unmap* events, to
xdg_surface_unmap. The actual fix, however, is to listen to destroy
events. This fixes various crashes.
Previously, adaptive sync was just a hint and wouldn't make any
atomic commit fail if the backend didn't support it. The main reason
is wlr_output_test wasn't supported at the time.
Now that we have a way for compositors to test whether a change can
work, let's remove the exception for adaptive sync and convert it to
a regular output state field.
array_realloc will grow the array for the target size like wl_insert_add, but
will also shrink the array if the target size is sufficiently smaller than the
current allocation.
This lets the renderer handle the wlr_buffer directly, just like it
does in texture_from_buffer. This also allows the renderer to batch
the rectangle updates, and update more than the damage region if
desirable (e.g. too many rects), so can be more efficient.
We were firing the new_input signal on backend initialization,
before the compositor had the chance to add a listener for it.
Mimick what's done for wl_keyboard: if the backend hasn't been
started, delay wl_touch initialization.
Closes: https://gitlab.freedesktop.org/wlroots/wlroots/-/issues/3473
When the client doesn't support high-resolution scroll, accumulate
deltas until we can notify a discrete event.
Some mice have a free spinning wheel, making possible to lock the wheel
when the accumulator value is not 0. To avoid synchronization issues
between the mouse wheel and the accumulators, store the last delta and
when the scroll direction changes, reset the accumulator.
On newer versions of libinput, the event LIBINPUT_EVENT_POINTER_AXIS
has been deprecated in favour of LIBINPUT_EVENT_POINTER_SCROLL_WHEEL,
LIBINPUT_EVENT_POINTER_SCROLL_FINGER and
LIBINPUT_EVENT_POINTER_SCROLL_CONTINUOUS.
Where new events are provided by the backend, ignore
LIBINPUT_EVENT_POINTER_AXIS, receive high-resolution scroll events from
libinput and emit the appropiate wlr_pointer signal.
Currently, the "wlr_event_pointer_axis" event stores low-resolution
values in its "delta_discrete" field. Low-resolution values are always
multiples of one, i.e., 1 for one wheel detent, 2 for two wheel
detents, etc.
In order to simplify internal handling of events, always transform in
the backend from the low-resolution value into the high-resolution
value.
The transformation is performed by multiplying by 120. The 120 magic
number is used by the kernel and it is exposed to clients in the
"WLR_POINTER_AXIS_DISCRETE_STEP" constant.
wlr_xdg_surface_from_wlr_surface() for example may return NULL even if
the surface has the xdg surface role if the corresponding xdg surface
has been destroyed.
"max bpc" is a maximum value, the driver is free to choose a
smaller value depending on the bandwidth available.
Some faulty monitors misbehave with higher bpc values. We'll add
a workaround if users get hit by these in practice.
References: https://gitlab.freedesktop.org/wayland/weston/-/issues/612
Whether a texture is opaque or not doesn't depend on the renderer
at all, it just depends on the source buffer. Instead of forcing
all renderers to implement wlr_texture_impl.is_opaque, let's move
this in common code and use the wlr_buffer format to know whether
a texture will be opaque.
This refactors output_ensure_buffer() to not mutate the state passed,
making the previous subtle behavior much more explicit.
Fixes: d483dd2f ("output: add wlr_output_commit_state")
Closes: #3442
CTA-861-G says that "graphics" is used to indicate non-analog (ie,
digital) content. With that bit set, the sink should turn off analog
reconstruction and other related filtering.
This commit ensures that outputs that weren't created by the output
layout helper aren't destroyed on the output layout change.
Consider the following piece of logic:
// struct wlr_output *o1, *o2;
// struct wlr_scene *scene;
// struct wlr_output_layout *layout;
wlr_scene_attach_output_layout(scene, layout);
wlr_output_layout_add_auto(layout, o1);
struct wlr_scene_output *so2 = wlr_scene_output_create(scene, o2);
wlr_output_layout_move(layout, o1, 100, 200);
// so2 is invalid now
Now that the DRM backend no longer depends on GBM, we can make it
optional. The GLES2 renderer still depends on it because of our EGL
device selection.
This is useful for compositors with their own renderers, and for
compositors using the Vulkan renderer.
This allows the make/model/serial to be NULL when unset, and allows
them to be longer than the hardcoded array length.
This is a breaking change: compositors need to handle the new NULL
case, and we stop setting make/model to useless "headless" or
"wayland" strings.
These are trivial wrappers around eglMakeCurrent and
eglGetCurrentContext. Compositors which need to call these
functions will also call other EGL or GL functions anyways. Let's
reduce our API surface a bit by making them private.
This will display red translucent rectangles on the screen regions that
have been damaged. These rectangles will fade out over the span of 250
msecs. If the area is damaged again while the region is fading out,
the timer is reset.
Let's also disable direct scan out when this option is enabled, or else
we won't be able to render the highlight damage regions.
After cancelation we destroy the touch points associated with this
surface as the Wayland spec says:
No further events are sent to the clients from that particular gesture.
Touch cancellation applies to all touch points currently active on this
client's surface. The client is responsible for finalizing the touch
points, future touch points on this surface may re-use the touch point
ID.
Closes: https://gitlab.freedesktop.org/wlroots/wlroots/-/issues/2999
