Using values from the color representation protocol. Only supported for
YCbCr formats in KMS drivers for now.
Signed-off-by: Robert Mader <robert.mader@collabora.com>
If we're in a steady state, doing nothing but flipping buffers, we can
try to avoid going through our full routine of brute-forcing an
acceptable plane state, by instead just reusing the old state and
changing only the FB it refers to.
Signed-off-by: Daniel Stone <daniels@collabora.com>
Signed-off-by: Derek Foreman <derek.foreman@collabora.com>
Keeping track of what mode we prepared the state in will be useful
for printing debug information later, when we have a way to reuse
old state.
Signed-off-by: Derek Foreman <derek.foreman@collabora.com>
We're going to add a way to reuse state, but we don't currently have a way
to represent invalid state - such as before we've ever commit any state
at all.
Signed-off-by: Derek Foreman <derek.foreman@collabora.com>
These have the same form, and we're about to add a few more with the same
form again, so make a helper function.
This sweeps up the one weird one that printed [view] while failing [state].
Signed-off-by: Derek Foreman <derek.foreman@collabora.com>
This will be handy later when trying to determine when it's ok to reuse
output state, as changing this would invalidate previous state.
Signed-off-by: Derek Foreman <derek.foreman@collabora.com>
Although Vulkan's renderbuffers are dmabufs, the difference between a
client dmabuf and a backend-allocated dmabuf is a very meaningful one.
Add a separate buffer type for BUFFER_DMABUF_BACKEND instead of
overloading BUFFER_DMABUF for this.
This tangentially allows the Vulkan renderer to do direct scanout of
client buffers in mixed mode.
Signed-off-by: Daniel Stone <daniels@collabora.com>
The DRM documentation states:
> Unless explicitly specified (via CRTC property or otherwise), the active
> area of a CRTC will be black by default. This means portions of the active
> area which are not covered by a plane will be black, and alpha blending of
> any planes with the CRTC background will blend with black at the lowest zpos.
See https://dri.freedesktop.org/docs/drm/gpu/drm-kms.html#plane-abstraction
This means the view for the primary plane does not need to cover the
whole output and black areas of the scene-graph can be left out.
Doing so has various benefits - most importantly it:
1. allows us to use the plane-only path in more situations and with one
less plane, reducing memory bandwidth usage.
2. opens the path to offload arbitrary background colors in the future.
Iterate over the all visible paint nodes, remove solid-opaque-black
views so they are not considered for plane assignment and aggregate a
region that is later used to assign the primary plane.
Signed-off-by: Robert Mader <robert.mader@collabora.com>
So far scene-graph building and plane assignment happened within a single
loop. In the future we want to be able to optimize the scene-graph
before assigning, thus split up the loop into two steps.
No behavioral changes intended.
Signed-off-by: Robert Mader <robert.mader@collabora.com>
Instead of doing duplicated calculations. Apart from being faster and
less code, it should help ensuring correctness of the given regions.
Signed-off-by: Robert Mader <robert.mader@collabora.com>
In order to streamline our checks and to make it easier to ensure
that we replace paint nodes with solid placeholders correctly.
Fix the failure reason in one case while on it.
Signed-off-by: Robert Mader <robert.mader@collabora.com>
This test is critical, so only do it once, and throw some asserts around
to make sure we don't mess it up.
Signed-off-by: Derek Foreman <derek.foreman@collabora.com>
Video underlay "holes" are fully transparent but must be rendered
fully opaque. However, they appear to be fully transparent to
the current gl-renderer test which is missing a need_hole check.
Add a paint node attribute for this so it can be more easily
checked in assign_planes and renderers.
Signed-off-by: Derek Foreman <derek.foreman@collabora.com>
Some client use subsurfaces with fully transparent single-pixel buffers
for various reasons, such as making it easier to order trees of
subsurfaces. As they are invisible we can simply ignore them in the
scene graph.
This allows us to use direct-scanout/plane-only in more circumstances,
as ensured in the test.
Signed-off-by: Robert Mader <robert.mader@collabora.com>
Output color effects are applied to the whole output scenegraph. It
depends on color-management being disabled, as the color effects are
applied in sRGB content.
For now we added only a few accessibility options: color inversion,
deuteranopia, protanopia and tritanopia CVD correction.
Note that surfaces presented on outputs that contains a color effect
can't be used for direct scanout (i.e. bypass composition and offloading
to KMS overlay planes). The color effect is applied in our GL-renderer.
Signed-off-by: Leandro Ribeiro <leandro.ribeiro@collabora.com>
Now that FAILURE_REASONS_FORCE_RENDERER isn't set for all not-scanout
capable failures, add some other reasons to the test in
dma_feedback_maybe_update.
Signed-off-by: Derek Foreman <derek.foreman@collabora.com>
In a2a6030902 I introduced a crash when no
cursor plane exists. Fix this by pulling the cursor plane tests into their
own function and doing an early return when we have a conclusive no-cursor
failure state.
Fixes a2a6030902
Signed-off-by: Derek Foreman <derek.foreman@collabora.com>
While this is generally drm specific stuff, it's leaked into paint nodes
a little bit already.
Pushing it into the core gives us the ability to print failure reasons
in the scene graph debug text (in a future commit), which can be very
informative.
Signed-off-by: Derek Foreman <derek.foreman@collabora.com>
We have a big bitfield of failure reasons for when we can't place a paint
node on a plane - but between the introduction of this bitfield and now
we added many new reasons, so it's an expecation that the reason
FAILURE_REASON_FORCE_RENDERER only means that the compositor is configured
to force the use of the renderer via the WESTON_FORCE_RENDERER environment
variable, or debug key bind, or similar.
Some of the code retained the older generic use of FORCE_RENDERER to mean
any number of things.
Add some new reasons and use other existing reasons to disambiguate the
FORCE_RENDERER reason.
Since we already print the (somewhat misleading) failure reason strings
late in the repaint pass, we can drop the (currently accurate) debug
prints we fired off along the way, and just print them accurately at the
end.
We also move the failure reason reset point to the start of repaint,
as it is convenient to have it be valid after plane assignment so
we can print it in scene graph debug text in a later commit.
Signed-off-by: Derek Foreman <derek.foreman@collabora.com>
The caller to drm_plane_state_coords_for_paint_node() has already tested
for valid transforms, so drm_plane_state_coords_for_paint_node() is never
called with a transform that needs to be tested.
Replace the test with an assert() - and this also lets us stop returning
a bool, and drop the debug prints that could never trigger.
Signed-off-by: Derek Foreman <derek.foreman@collabora.com>
On weston_output_repaint(), we have:
output_assign_planes(output);
...
output->repaint(output);
output_assign_planes() avoids calling output->assign_planes() when
output->disable_planes > 0. This brings a few complications to backends,
because they can't rely on the repaint loop starting from a certain
function: sometimes it starts from assign_planes(), others from
repaint().
Let's be more consistent: always call assign_planes() and let backends
handle output->disable_planes.
Signed-off-by: Leandro Ribeiro <leandro.ribeiro@collabora.com>
If we call drm_output_find_plane_for_view() with
DRM_OUTPUT_PROPOSE_STATE_RENDERER_AND_CURSOR for a view that is not a
cursor view, it should fail to get a plane.
In such case, set the failure reason to FAILURE_REASONS_FORCE_RENDERER.
Signed-off-by: Leandro Ribeiro <leandro.ribeiro@collabora.com>
In the next commit we introduce a real "renderer-only" mode. So let's
rename DRM_OUTPUT_PROPOSE_STATE_RENDERER_ONLY to
DRM_OUTPUT_PROPOSE_STATE_RENDERER_AND_CURSOR, as it also assign views
to cursor planes.
Signed-off-by: Leandro Ribeiro <leandro.ribeiro@collabora.com>
drm_output_propose_state() is called during from drm_assign_planes(). At this
point, the output is enabled, but it may be disabled in the current state, that
is copied.
So explicitly enable the output by setting dpms to WESTON_DPMS_ON.
The same thing is already done in drm_output_repaint() but that's to late for
drm_output_propose_state(). Move that to the fallback path when planes are
disabled. It's no longer needed in general and changing the state between the
atomic test and commit is not a good idea anyways.
Without this, the atomic commit may fail, even though the corresponding test was
successful.
Signed-off-by: Michael Olbrich <m.olbrich@pengutronix.de>
The sampler type, which was initially meant for the GL renderer, is
currently used to derive the color model of pixel formats. This commit
gets rid of this legacy field and replaces it with an explicit one
indicating the color model.
Note that it also fixes the issue that formats wouldn't be considered
YUV if ENABLE_EGL isn't defined because the sampler type field in that
particular case would always be 0.
Signed-off-by: Loïc Molinari <loic.molinari@collabora.com>
Upstream KMS drivers currently default to either ITU-R BT.601 YCbCr or
ITU-R BT.709 YCbCr. Ensure consistent behavior by setting a default.
We use BT.601 as that is what our fallback shader uses at the moment.
Signed-off-by: Robert Mader <robert.mader@collabora.com>
Up to now, we've been logging only the failures that we get from
drm_fb_get_from_paint_node(). So let's start logging the other failures
we get in drm_output_find_plane_for_view() as well.
Signed-off-by: Leandro Ribeiro <leandro.ribeiro@collabora.com>
This should help people to debug. Instead of printing an hex value,
print each enum value present in the bitmask as a string (comma
separated).
Signed-off-by: Leandro Ribeiro <leandro.ribeiro@collabora.com>
Client won't be able to do direct-scanout with this surface, and the
format/modifier doesn't matter in this case. So let's remove the scanout
tranche from dma-buf feedback in such case.
Signed-off-by: Leandro Ribeiro <leandro.ribeiro@collabora.com>
Currently we have the following situation: we add a scanout tranche
because if the client re-allocates with another format/modifier, the
chances of being placed in a DRM/KMS plane is higher.
But then we run out of overlay planes. So we remove the scanout tranche,
because the format/modifier available in the renderer tranche are
optimal for rendering.
Now Weston detects again that the format/modifier is what may be
avoiding the view being place in a plane, re-adding the scanout tranche.
And we have an endless loop.
To avoid this, let's accumulate the reasons why placing the view in a
place failed. So if we detect that we don't have planes available, no
matter the format/modifier, we won't add the scanout tranche.
Signed-off-by: Leandro Ribeiro <leandro.ribeiro@collabora.com>
There are a few points in the code where we are wrongly using
FAILURE_REASONS_ADD_FB_FAILED, probably because we didn't have so many
"failure reasons" previously. This update such cases to use enum's that
make sense.
Signed-off-by: Leandro Ribeiro <leandro.ribeiro@collabora.com>
Do not skip all the planes if a single one of them do not support
fences. The other may do.
Signed-off-by: Leandro Ribeiro <leandro.ribeiro@collabora.com>
Now that secondary planes can be both underlay and overlay, this flag's
meaning also changed. Update it for correctness.
Signed-off-by: Leo Li <sunpeng.li@amd.com>
Previously, whether a KMS plane is underlay-able is determined by
whether it's zpos_max is < the primary plane's zpos_min. In other words,
a plane will only be considered underlay-able if its entire valid zpos
range is under the primary plane's lowest zpos.
This is too restrictive - it's possible for planes to have a valid zpos
range that spans below and above the primary's zpos range.
Therefore, allow planes to be used as underlays if their zpos_min is <
the primary plane's zpos_min.
In addition, force rendering on a view if it contains alpha, and is
occluded by a rendered view. If such a view is overlaid, it would render
with incorrect zorder. If it's underlaid, it would render with incorrect
alpha-blending due to hole-punching. Therefore, it must be rendered.
Force rendering prevents the view from going into
`drm_output_find_plane_for_view()`, which serves as an optimization, but
is also observed to prevent dmabuf feedback (derived from
`try_view_on_plane_failure_reasons`) from ping-ponging between two
values, causing some apps (like weston-simple-egl) to constantly
reallocate its buffers.
Because a plane can now - if supported - be used as an underlay, an
overlay, or both, add a `enum drm_plane_subtype` to differentiate
between them. Then, print it's subtype and underlay/overlay assignment
once a decision is made.
v2:
* Squash w/ patch to force rendering on alpha view occluded by rendered
view
* Bring back plane subtype enum to be more expressive about plane
capabilities
* Correct need_hole != false when a view's assignment changes from
underlay to overlay
Signed-off-by: Leo Li <sunpeng.li@amd.com>
We need to reset the underlay check inside the view evaluation loop,
otherwise once we need an underlay we'll treat every following view as
needing an underlay.
Fixes: 1065d23406 ("backend-drm: Improve plane assignment for underlay platform")
Signed-off-by: Derek Foreman <derek.foreman@collabora.com>
Fundamentally, the flags are a property of each paint node, rather than
each view as such. Move them over there so it gets a little less painful
to work with.
Signed-off-by: Daniel Stone <daniels@collabora.com>
1. remove the restriction on underlay planes when finding plane.
Because the view on the underlay plane can be displayed by drawing
a through hole on primary plane, so we can try underlay planes.
2. Add step to check if the view is assigned on underlay plane, When
it is successfully placed on a HW plane. Because we need to set the
underlay view pnode->need_hole to true so that gl-renderer
will draw a hole for it when repainting.
3. Avoid assigning views to underlay HW planes when the backend format
is opaque and avoid assigning views with alpha to underlay HW planes.
4. when overlay plane is not enough, try to find underlay plane on
platform with both overlay and underlay plane.
Signed-off-by: Chao Guo <chao.guo@nxp.com>
Signed-off-by: Marius Vlad <marius.vlad@collabora.com>
On underlay platforms, the HW planes can be placed below the primary
plane, so some views that intersect with the renderer region can try
to be placed on the underlay planes. In order to assign these views
to underlay planes, the improvement is as follows:
1. Add current_lowest_zpos_overlay. Record the current lowest zpos
of the overlay planes.
2. Add current_lowest_zpos_underlay. Record the current lowest zpos
of the underlay plane. It is initialized to scanout_plane::zpos.
3. Add need_underlay to indicate whether to find underlay plane for
view.
4. The views that intersect with the renderer region and underlay
views should be assigned to underlay planes.
Signed-off-by: Chao Guo <chao.guo@nxp.com>
Doing this might cause unnecessary DRM importing, which might result in
an error when the DRM's memory address mapping is nearly full.
Return before attempting to create drm_fb to avoid that, since it will
fail in the later check anyway.
Signed-off-by: Jeffy Chen <jeffy.chen@rock-chips.com>
If a view is non-opaque - such as an overlay over a video - we shouldn't
force it to be on the primary plane, as that's where the underlying
content should be placed, such as the video view.
dc0de9ee already mentioned: "This check should be changed in future to
only filter for opaque views, but that's for another time."
Adding "Fixes" at this is arguably a bug fix:
Fixes: dc0de9ee (backend-drm: Move overlay vs. primary plane check earlier)
Fixes: 2538aacc (backend-drm: Construct a zpos candidate list of planes)
Signed-off-by: Robert Mader <robert.mader@collabora.com>
We can't use the surface damage to determine when to upload new cursor
images because when heads overlap the first repainted head will accumulate
that damage as plane damage.
We can't easily use plane damage either because the plane isn't really
assigned until after an atomic test, which requires the cursor fb to be
current.
Untangle this mess a little by always testing with the first cursor fb,
which is identical to the second in all ways, then replace with the correct
fb in repaint.
Signed-off-by: Derek Foreman <derek.foreman@collabora.com>
