connect_drm_connector() may be called long after create_drm_connector().
During that time the DRM mode might have changed. Avoid working with
stale information.
match_obj() might return a configuration where the CRTC for an
enabled connector is switched to another one.
We don't support this correctly: the wlr_output common code would
need to query again the supported formats, re-allocate the
swapchain, etc.
What's more, the kernel doesn't even support this [1]: it
requires planes to be disabled to change their CRTC, it rejects
commits directly switching the CRTC used by a plane.
[1]: https://cgit.freedesktop.org/drm/drm-misc/tree/drivers/gpu/drm/drm_atomic.c?h=6e90293618ed476d6b11f82ce724efbb9e9a071b#n697
If the commit fails, then our local state becomes out-of-sync with
the kernel's. Additionally, when disabling a connector without going
through dealloc_crtc(), conn->crtc would still be set.
Fix this by updating conn->crtc in drm_connector_commit_state().
The raw enum value wasn't informative enough. It's not trivial to
tell whether 0 means connected or disconnected.
Drop the status from the state after realloc, since the exact same
information is printed right above.
Instead of having a pending_fb field on the struct wlr_drm_plane,
move it to struct wlr_drm_connector_state. That way, there's no
risk having a stale pending FB around: the state doesn't survive
across tests and commits.
The cursor is a special case because it's disconnected from the
atomic state: the wlr_backend_impl.set_cursor hook sets the cursor
for the next commit. Move the field to
wlr_drm_connector.cursor_pending_fb.
We'll move the pending primary FB into the connector state in the
next commit, dropping wlr_drm_plane.pending_fb in the process.
Introduce a dedicated field for the cursor, which has to be managed
in a special way due to our set_cursor API.
We were calling drm_connector_supports_vrr() before
drm_connector_alloc_crtc(). Thus, when an output is currently off,
the VRR test would always fail, because it checks that the
vrr_enabled CRTC prop exists.
destroy_wl_buffer() is called from backend_destroy(). We need to
ensure the wlr_buffer is unlocked when we're waiting for a
wl_buffer.release event from the parent compositor.
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.
The following situation can be dangerous:
- Output DP-1 is plugged in, compositor enables it.
- User VT switches away.
- User unplugs DP-1.
- User VT switches back.
- scan_drm_connectors() figures out the output is now disconnected,
uninitializes the struct wlr_output.
- The loop restoring previous output state in handle_session_active()
accesses the struct wlr_output to figure out what to restore.
By chance, we zero out the struct wlr_output after uninitializing it,
so enabled and current_mode will always be zero. But let's make sure
we handle this case explicitly, to remind future readers that it exists
and make the code less fragile.
The old drm_connector_set_mode() function did that by calling
drm_crtc_page_flip(). We lost this in the refactoring.
Fixes: f216e97983 ("backend/drm: drop drm_connector_set_mode()")
Once we are DRM master, the CRTC cannot be changed behind our back
except during a VT switch.
After a VT switch, we try to restore whatever KMS state we had last
programmed. Reloading the current CRTC from KMS breaks this and
can result in a modeset without a FB.
Closes: https://gitlab.freedesktop.org/wlroots/wlroots/-/issues/3432
The drm_connector_commit_state() call in handle_session_active()
was not resulting in any atomic commit, because it didn't match any
of the if branches: active = true, no new buffer was committed,
and adaptive sync/gamma LUT were unchanged. Thus the commit was a
no-op.
Later on, when the compositor performs regular page-flips, the
kernel would return EINVAL indicating that a modeset was needed.
Rework the logic to use a non-blocking page-flip commit if a buffer
was committed, and use a blocking commit if the connector is on or
is being disabled. The only case where we should skip the atomic
commit is when disabling (active = false) an already-disabled
connector (conn->crtc == NULL).
Note, 6936e163b5 ("backend/drm: short-circuit no-op commits")
has introduced early returns for other situations where we don't
need to perform an atomic commit (e.g. updating scale or transform
of an output).
Fixes: f216e97983 ("backend/drm: drop drm_connector_set_mode()")
Closes: https://gitlab.freedesktop.org/wlroots/wlroots/-/issues/3432
Extract the logic to fetch the current mode to a separate function
to make it more readable. Stop dying in an assert when
get_drm_prop_blob() fails. Always make it so the drmModeModeInfo
pointer is allocated so that we can free() it unconditionally.
We already have disconnect_drm_connector() to handle the
CONNECTED → DISCONNECTED transition. Let's add
connect_drm_connector() to handle DISCONNECTED → CONNECTED. This
makes scan_drm_connectors() shorter and easier to follow.
No functional change, literally just moving code around.
We were using the legacy API (with a detour through drmModeEncoder)
to find out the current CRTC for a connector. Use the atomic API
when available.
Also extract the whole logic into a separate function for better
readability, and better handle errors.
Instead of special-casing modesets, we can just cut the wrapper
and directly call drm_crtc_page_flip(). drm_connector_test() should
already have the checks previously done in drm_connector_set_mode(),
all we need to do is update enabled/mode after a successful atomic
commit.
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.
We were unconditonally setting desired_enabled = true for all
connected connectors. This makes realloc_crtcs() always keep a CRTC
active for these, even if the user doesn't want to enable them.
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
Stack trace:
#0 0x00007f17081f5b99 in wl_list_insert (list=list@entry=0x2d8, elm=elm@entry=0x7ffe7f7e85d0)
at ../wayland-1.21.0/src/wayland-util.c:48
#1 0x00007f17081f5f2e in wl_signal_emit_mutable (signal=signal@entry=0x2d8, data=data@entry=0x7ffe7f7e8660)
at ../wayland-1.21.0/src/wayland-server.c:2167
#2 0x00007f170815a971 in handle_switch_toggle (wlr_switch=0x2a0, event=0x55d5ba13dc00)
at ../backend/libinput/switch.c:50
#3 handle_libinput_event (event=0x55d5ba13dc00, backend=0x55d5b975d740) at ../backend/libinput/events.c:234
#4 handle_libinput_readable (fd=<optimized out>, mask=<optimized out>, _backend=<optimized out>)
at ../backend/libinput/backend.c:58
#5 handle_libinput_readable (fd=fd@entry=34, mask=mask@entry=1, _backend=_backend@entry=0x55d5b975d740)
at ../backend/libinput/backend.c:48
#6 0x00007f170815c110 in backend_start (wlr_backend=0x55d5b975d740) at ../backend/libinput/backend.c:109
#7 0x00007f1708160996 in multi_backend_start (wlr_backend=0x55d5b97583d0) at ../backend/multi/backend.c:32
All we can do to influence adaptive sync on the X11 backend is set the
_VARIABLE_REFRESH window property like mesa automatically does. We don't
have any control beyond that, so we set the state to enabled on creating
the output and never allow changing it (just like the Wayland backend).
Adaptive sync is effectively always enabled when using the Wayland
backend. This is not something we have control over, so we set the
state to enabled on creating the output and never allow changing it.
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.
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
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.
Starting with Linux Kernel v5.0 two new axes are available for
mice that support high-resolution wheel scrolling: REL_WHEEL_HI_RES and
REL_HWHEEL_HI_RES.
Both axes send data in fractions of 120 where each multiple of 120
amounts to one logical scroll event. Fractions of 120 indicate a wheel
movement less than one detent.
Three new events are now available on libinput:
LIBINPUT_EVENT_POINTER_SCROLL_WHEEL,
LIBINPUT_EVENT_POINTER_SCROLL_FINGER, and
LIBINPUT_EVENT_POINTER_SCROLL_CONTINUOUS.
These events replace the LIBINPUT_EVENT_POINTER_AXIS event, so new
clients should simply ignore that event.
Also, two new APIs are available to access the high-resolution data:
libinput_event_pointer_get_scroll_value() and
libinput_event_pointer_get_scroll_value_v120().
Add a project argument (LIBINPUT_HAS_SCROLL_VALUE120) to allow
building against old versions of libinput or, where high-resolution
scroll is available, support it.
"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
The EDID 1.4 spec says that the serial number is optional:
> If this field is not used, then enter “00h, 00h, 00h, 00h”.
Leave the wlr_output.serial field NULL in that case, and hide it
from the output description.
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.
Some output commits (changing e.g. the output scale or transform)
don't require any change in the KMS state. Instead of going through
a KMS commit, return early. Blocking KMS commits can be expensive.
The wlr_output API requires compositors to wait for wlr_output.frame
before submitting a new buffer. However, compositors can perform a
commit which doesn't involve a buffer anytime.
If the commit is a modeset, we set DRM_MODE_ATOMIC_ALLOW_MODESET and
block until the commit is done. If it isn't, we currently always
perform a non-blocking commit. This is an issue because a previous
page-flip might still be in flight kernel-side, returning EBUSY.
Fix this by using blocking commits when a buffer isn't submitted by
the compositor.
Closes: https://github.com/swaywm/sway/issues/6962
References: https://gitlab.freedesktop.org/wlroots/wlroots/-/issues/2239
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.
Prior to [1], if an entry in a DRM format set was different than a
single LINEAR modifier, implicit modifiers were always allowed. This
has changed and now implicit modifiers are only allowed if INVALID
is in the list of modifiers.
So now we can safely enable explicit modifiers for cross-GPU imports,
without risking receiving buffers with an implicit modifier. This
should improve perf a bit on setups where two GPUs from the same vendor
are used.
This fixes the first bullet point from [2].
[1]: https://gitlab.freedesktop.org/wlroots/wlroots/-/merge_requests/3231
[2]: https://gitlab.freedesktop.org/wlroots/wlroots/-/issues/3331
Maintaining our internal table up-to-date is tedious: one needs to
manually go through the PnP ID registry [1] and check whether we're
missing any entry.
udev_hwdb already has an API to fetch a manufacturer name from its
PnP ID. Use that instead.
[1]: https://uefi.org/pnp_id_list
Instead of waking up each 16ms to emit a frame event, arm the timer
when the output is committed. This allows the headless backend to
idle when nothing changes on screen, and behaves similarly to the
other backends.
All the code logic related to the pointer has been moved to its own file.
The seat is responsible for the lifetime of its wlr_wl_pointer(s), and assigning
them to the relevant wlr_wl_output. The wlr_wl_pointer becomes a simple helper
to manager the wlr_pointer associated to the seat's wl_pointer and its lifetime.
The destroy callback in wlr_touch_impl has been removed. The function
`wlr_touch_finish` has been introduced to clean up the resources owned by a
wlr_touch.
`wlr_input_device_destroy` no longer destroys the wlr_touch, attempting to
destroy a wlr_touch will result in a no-op.
The field `name` has been added to the wlr_touch_impl to be able to identify
a given wlr_touch device.
The destroy callback in wlr_tablet_tool_impl has been removed. The function
`wlr_tablet_tool_finish` has been introduced to clean up the resources owned by
a wlr_tablet_tool.
`wlr_input_device_destroy` no longer destroys the wlr_tablet_tool, attempting to
destroy a wlr_tablet_tool will result in a no-op.
The field `name` has been added to the wlr_tablet_tool_impl to be able to
identify a given wlr_tablet_tool device.
The destroy callback in wlr_tablet_pad_impl has been removed. The function
`wlr_tablet_pad_finish` has been introduced to clean up the resources owned by a
wlr_tablet_pad.
`wlr_input_device_destroy` no longer destroys the wlr_tablet_pad, attempting to
destroy a wlr_tablet_pad will result in a no-op.
The field `name` has been added to the wlr_tablet_pad_impl to be able to identify
a given wlr_tablet_pad device.
The destroy callback in wlr_switch_impl has been removed. The function
`wlr_switch_finish` has been introduced to clean up the resources owned by a
wlr_switch.
`wlr_input_device_destroy` no longer destroys the wlr_switch, attempting to
destroy a wlr_switch will result in a no-op.
The field `name` has been added to the wlr_switch_impl to be able to identify
a given wlr_switch device.
The destroy callback in wlr_pointer_impl has been removed. The function
`wlr_pointer_finish` has been introduced to clean up the resources owned by a
wlr_pointer.
`wlr_input_device_destroy` no longer destroys the wlr_pointer, attempting to
destroy a wlr_pointer will result in a no-op.
The field `name` has been added to the wlr_pointer_impl to be able to identify
a given wlr_pointer device.
The destroy member in wlr_keyboard_impl has been removed. The function
`wlr_keyboard_finish` has been introduce to clean up the resources owned by a
wlr_keyboard.
`wlr_input_device_destroy` no longer destroys the wlr_keyboard, attempting to
destroy a wlr_keyboard will result in a no-op.
The field `name` has been added to the wlr_keyboard_impl to be able to identify
a given wlr_keyboard device.
The libinput backend is now optional. However, this means that a
user building wlroots without the correct libinput dependencies
will end up with a compositor which doesn't respond to input events.
wlr_backend_autocreate is supposed to return a sensible setup, so in
this case let's just error out and explain what happened. Users can
suppress the check by setting WLR_LIBINPUT_NO_DEVICES=1 (already used
to suppress the zero input device case inside the libinput backend).
Compositors which really want to create a bare DRM backend can easily
create it manually instead of using wlr_backend_autocreate.
This patch makes it so we bind to zwp_linux_dmabuf_v1 version 4 and
we use it to grab the main device. v4 sends supported formats via a
table so we need to handle this as well.
v4 allows wlroots to remove the requirement for Mesa's internal
wl_drm interface.
wlroots picks names for all outputs, but it might be desirable for
compositor to override it.
For instance, Sway will use a headless output as a fallback in
case no outputs are connected. Sway wants to clearly label the
fallback output as such and label "real" headless outputs starting
from HEADLESS-1.
When using `meson --buildtype=release`, `-Wextra -Werror` is passed.
This includes `-Werror=maybe-uninitialized`, which complains about
the instances fixed in this commit.
drmModeAddFB2 doesn't support explicit modifiers. Only accept INVALID
which indicates an implicit modifier and LINEAR which may indicate
that GBM_BO_USE_LINEAR has been used.
The headless backend no longer needs a parent renderer: it no longer
needs to return it in wlr_backend_impl.get_renderer, nor does it
need to return its DRM FD in wlr_backend_impl.get_drm_fd. Drop this
function altogether since it now behaves exactly like
wlr_headless_backend_create.
Sometimes the headless backend is used standalone with the Pixman
renderer, sometimes it's used together with another backend which
has already picked a DRM FD. In both of these cases it doesn't make
sense to pick a DRM FD.
Broadly speaking the headless backend doesn't really care which DRM
device is used for the buffers it receives. So it doesn't really
make sense to tie it to a particular DRM device.
Let the backend users (e.g. wlr_renderer_autocreate) open an arbitrary
DRM FD as needed instead.
This field's ownership is unclear: it's in wlr_input_device, but
it's not managed by the common code, it's up to each individual
backend to use it and clean it up.
Since this is a backend implementation detail, move it to the
backend-specific structs.
There's no guarantee that the parent Wayland compositor uses
CLOCK_MONOTONIC for reporting presentation timestamps, they could
be using e.g. CLOCK_MONOTONIC_RAW or another system-specific clock.
Forward the value via wlr_backend_impl.get_presentation_clock.
References: https://gitlab.freedesktop.org/wlroots/wlroots/-/merge_requests/3254#note_1143061
Most (and possibly all) compositors using wlroots only ever render
fully opaque content. To provide better performance, this change
switches the default format used by wlr_output buffers from
ARGB8888 to the opaque XRGB8888.
Compositors like mutter, kwin, and weston already default to
XRGB8888, so this change is unlikely to expose any new bugs in
underlying drivers and hardware.
This does not affect the hardware cursor's buffer format, which is
still ARGB8888 by default.
As part of this change, the X11 backend (which does not support
changing format at runtime) now picks a true color, 24 bit depth
visual (i.e. XRGB8888) instead of a 32 bit depth (ARGB8888) one.
They are never used in practice, which makes all of our flag
handling effectively dead code. Also, APIs such as KMS don't
provide a good way to deal with the flags. Let's just fail the
DMA-BUF import when clients provide flags.
The BO handle table exists to avoid double-closing a BO handle,
which aren't reference-counted by the kernel. But if we can
guarantee that there is only ever a single ref for each BO handle,
then we don't need the BO handle table anymore.
This is possible if we create the handle right before the ADDFB2
IOCTL, and close the handle right after. The handles are very
short-lived and we don't need to track their lifetime.
Because of multi-planar FBs, we need to be a bit careful: some
FB planes might share the same handle. But with a small check, it's
easy to avoid double-closing the same handle (which wouldn't be a
big deal anyways).
There's one gotcha though: drmModeSetCursor2 takes a BO handle as
input. Saving the handles until drmModeSetCursor2 time would require
us to track BO handle lifetimes, so we wouldn't be able to get rid
of the BO handle table. As a workaround, use drmModeGetFB to turn the
FB ID back to a BO handle, call drmModeSetCursor2 and then immediately
close the BO handle. The overhead should be minimal since these IOCTLs
are pretty cheap.
Closes: https://github.com/swaywm/wlroots/issues/3164
Since libinput is an optional dependency the libinput backend is possibly undeclared.
wlr_backend_destroy(backend) below will clean up the child libinput backend if any.
This commit adds two error-handling cases to the function
attempt_dmr_backend. Specifically:
- In the case where the number of found GPUs is zero, we now
print a log message indicating this and return a NULL pointer
- In the case where we could not successfully create a backend
on any GPU, we now log a message indicating this and return
a NULL pointer
This allows us to provide more descriptive error messages,
as well as avoid a SEGFAULT (the function
`ensure_primary_backend_renderer_and_allocator` dereferences the pointer
given, which could be NULL until this patch) when these cases arise.
Simon Ser
illiliti
Alexander Orzechowski
Simon Zeni
John Lindgren
vanfanel
Isaac Freund
José Expósito
vaxerski
Kirill Primak
Samuel Čavoj
David96
Chris Chamberlain
Quantum
Moon Sungjoon
Manuel Stoeckl
Roman Gilg
Ronan Pigott
Haelwenn (lanodan) Monnier
MarkusVolk
Anthony Super