If we end up trying to do a mode set on a DRM state that has already
changed behind our back without us yet having seen the hotplug event we
may fail with `EINVAL`. Since the renderer layer doesn't handle mode set
failure, it'll still try to page flip later on, which will then also
fail. When failing, it'll try to look up the cached mode set in order to
retry the mode set later on, as is needed to handle other error
conditions. However, if the mode set prior to the page flip failed, we
won't cache the mode set, and the page flip error handling code will get
confused.
Instead of asserting that a page flip always has a valid cached mode set
ready to look up, handle it being missing more gracefully by failing to
mode set. It is expected that things will correct themself as there
should be a hotplug event waiting around the the corner, to reconfigure
the monitor configuration setting new modes.
Fixes: https://gitlab.gnome.org/GNOME/mutter/issues/917https://gitlab.gnome.org/GNOME/mutter/merge_requests/1007
(cherry picked from commit ce3409b2b7)
This avoids using bogus geometric values from an unmapped actor to
determine whether an actor is on a logical monitor or not. This would
happen when committing to a subsurface of a yet to be mapped toplevel.
https://gitlab.gnome.org/GNOME/mutter/merge_requests/961
(cherry picked from commit 6d15231f10)
As with most other state that ends up being pushed to the actor and the
associated shaped texture, also push the texture and the corresponding
metadata from the actor surface. This fixes an issue when a toplevel
surface was reset, where before the subsurface content was not properly
re-initialized, as content state synchronization only happened on
commit, not when asked to synchronize.
https://gitlab.gnome.org/GNOME/mutter/merge_requests/961
(cherry picked from commit f0df07cba3)
A actor surface may be reset by an xdg_toplevel if a NULL buffer is
attached. This should reset the actor state of the toplevel to an empty
state, while unmapping the previous actor. Subsurfaces, however, should
stay intact, including their relationship to the toplevel. They should
also not be yanked away from the actor of the actor surface prior to it
resetting, so that a window-destroy animation can include the subsurface
actor.
This fixes a potential crash when a subsurface tries to commit to its
wl_surface after the destroy animation of the toplevel has finished, as
the actor would at that point have been destroyed and cleared from the
actor surface struct, causing a segmentation fault.
https://gitlab.gnome.org/GNOME/mutter/merge_requests/961
(cherry picked from commit fe7bece31e)
Similar to wl_list_foreach(), add
META_WAYLAND_SURFACE_FOREACH_SUBSURFACE() that iterates over all the
subsurfaces of a surface, without the caller needing to care about
implementation details, such as leaf nodes vs non-leaf nodes.
https://gitlab.gnome.org/GNOME/mutter/merge_requests/961
(cherry picked from commit fcfe90aa9f)
While it's not very relevant now, as we would rarely create it anyway
since the buffer nor texture never changes for a surface, it will be in
the future, as the actor state (including its content,
MetaShapedTexture) will be synchronized by the MetaWaylandActorSurface
at a later point in time, and not by MetaWaylandSurface, at state
application time.
https://gitlab.gnome.org/GNOME/mutter/merge_requests/961
(cherry picked from commit 76ee026caa)
Using the same scale for the window as the
logical monitor only works correctly when having
the experimental 'scale-monitor-framebuffer'
feature enabled.
Without this experimental feature, the stream
will contain a black screen, where the actual
window only takes a small part of it.
Therefore, use a scale of 1 for the non-
experimental case.
Patch is based on commit 3fa6a92cc5.
https://gitlab.gnome.org/GNOME/mutter/merge_requests/976
(cherry picked from commit e89cea8e5a)
Presumably this function is supposed to be like
meta_kms_impl_simple_handle_page_flip_callback() but the condition in the
if-statement is inverted. Fix the inversion to make these two functions look
alike.
This is part 2 of 2 fixing a complete desktop freeze when drmModePageFlip()
fails with EINVAL and the fallback to drmModeSetCrtc() succeeds but the success
is not registered correctly as completed "flip". The freeze occurs under
wait_for_pending_flips() which calls down into meta_kms_impl_device_dispatch()
which ends up poll()'ing the DRM fd even though drmModeSetCrtc() will not
produce a DRM event, hence the poll() never returns. The freeze was observed
when hotplugging a DisplayLink dock for the first time on Ubuntu 19.10.
This patch makes meta_set_fallback_feedback_idle() actually end up calling into
notify_view_crtc_presented() which decrements
secondary_gpu_state->pending_flips so that wait_for_pending_flips() can finish.
CC stable: gnome-3-34
https://gitlab.gnome.org/GNOME/mutter/merge_requests/953
(cherry picked from commit 832a522cce)
mode_set_fallback() schedules a call to mode_set_fallback_feedback_idle(), but
it is possible for Mutter to repaint before the idle callbacks are dispatched.
If that happens, mode_set_fallback_feedback_idle() does not get called before
Mutter enters wait_for_pending_flips(), leading to a deadlock.
Add the needed interfaces so that meta_kms_device_dispatch_sync() can flush all
the implementation idle callbacks before it checks if any "events" are
available. This prevents the deadlock by ensuring
mode_set_fallback_feedback_idle() does get called before potentially waiting
for actual DRM events.
Presumably this call would not be needed if the implementation was running in
its own thread, since it would eventually dispatch its idle callbacks before
going to sleep polling on the DRM fd. This call might even be unnecessary
overhead in that case, synchronizing with the implementation thread needlessly.
But the thread does not exist yet, so this is needed for now.
This is part 1 of 2 fixing a complete desktop freeze when drmModePageFlip()
fails with EINVAL and the fallback to drmModeSetCrtc() succeeds but the success
is not registered correctly as completed "flip". The freeze occurs under
wait_for_pending_flips() which calls down into meta_kms_impl_device_dispatch()
which ends up poll()'ing the DRM fd even though drmModeSetCrtc() will not
produce a DRM event, hence the poll() never returns. The freeze was observed
when hotplugging a DisplayLink dock for the first time on Ubuntu 19.10.
CC stable: gnome-3-34
https://gitlab.gnome.org/GNOME/mutter/merge_requests/953
(cherry picked from commit 79491df2b8)
When a window that should be stacked above another one is placed in a lower
layer than the other window, we currently allow promoting it to the higher
layer when it has a "transient type". We should do the same when the window
is an actual transient of the other window.
This is particularly relevant for wayland windows, where types play a
much smaller role: Transient windows like non-modal dialogs (and since
commit 666bef7a, popup windows as well) currently end up underneath their
always-on-top parent.
https://gitlab.gnome.org/GNOME/mutter/issues/587
(cherry picked from commit 1c89fce30e)
This was wrongly introduced in 75cffd0ec4. As the comment above explains, we
only want to queue redraws in response to surface/buffer damage.
This triggered a full redraw when using DMA buffers on Wayland as we currently
create a new texture on every buffer_attach(), breaking partial invalidation.
Fixes https://gitlab.gnome.org/GNOME/mutter/issues/947
(cherry picked from commit 0247d35e5a)
Since commit a2a8f0cda we force the focus surface of the
meta-wayland-pointer to NULL while the pointer is invisible. This
introduced an issue with the a11y magnifier of the shell, which uses
`meta_cursor_tracker_set_pointer_visible` to hide the real cursor and
show its own magnified cursor at the correct position: Because the
meta-wayland-pointer is still used to communicate with Wayland clients,
the UI of the windows will not respond to mouse movement anymore as
soon as the real cursor is hidden.
Fix this issue for now by adding an additional method to the
cursor-tracker which allows disabling the behavior commit a2a8f0cda
introduced.
https://gitlab.gnome.org/GNOME/mutter/merge_requests/832
There might be some inconsistent event for which we don't have a known
source device.
In the current state we don't handle them and we could crash when getting
the current device tool.
So, add an utility function that retrieves the source device for an event
that warns if no device is found, and use this for Motion, Key and Button
events.
In case we don't have a valid source in such case, just return early instead
of trying to generate invalid clutter events.
https://gitlab.gnome.org/GNOME/mutter/merge_requests/823
(cherry picked from commit 427670cc63)
On wl_data_source destruction we used to indirectly unset the DnD selection
owner via the wl_resource destructor triggering the destruction of the
MetaWaylandDataSource, which would be caught through the weak ref set by
the MetaWaylandDragGrab.
This works as long as the grab is held, however we have a window between
the button being released and the drop site replying with
wl_data_offer.finish that the MetaWaylandDataSource is alive, but its
destruction wouldn't result in the call chain above to unsetting the DnD
source.
In other selection sources, we let the MetaWaylandDataDevice hold the
"ownership" of the MetaWaylandDataSource, and its weak ref functions unset
the respective MetaSelection owners. Do the same here, so the
MetaWaylandDataSource destruction is listened for all its lifetime.
Closes: https://gitlab.gnome.org/GNOME/mutter/issues/591
(cherry picked from commit e5af790acb)
This is wrong for both clipboard and DnD, as the selection source
will still be able to focus another surface, and churn another
wl_offer.
We should just detach the data offer from the data source in this
case, and let the source live on. However, we should still check
that there is a source and an offer to finish DnD, do that when
handling the drop operation instead.
https://gitlab.gnome.org/GNOME/mutter/issues/591
(cherry picked from commit 48639ac5da)
Those were used to signal clipboard ownership around, but that got
replaced by MetaSelection and friends. These signals are no longer
listened on, so can be safely removed.
https://gitlab.gnome.org/GNOME/mutter/issues/591
(cherry picked from commit 46b3811e22)
Properly take the panel_orientation_transform into account in
update_monitor_crtc_cursor. This fixes us sometimes drawing the cursor
on two monitors at the same time as we did not properly swap the crtc
width/height when a panel_orientation_transform is active.
https://gitlab.gnome.org/GNOME/mutter/merge_requests/927
(cherry picked from commit 8907a29912)
Add missing clutter_x11_[un]trap_x_errors around the XIGetProperty call
in meta-input-settings-x11.c's get_property helper function.
This fixes mutter crashing with the following error if the XInput device
goes away at an unconvenient time:
X Error of failed request: XI_BadDevice (invalid Device parameter)
Major opcode of failed request: 131 (XInputExtension)
Minor opcode of failed request: 59 ()
Device id in failed request: 0x200011
Serial number of failed request: 454
Current serial number in output stream: 454
https://gitlab.gnome.org/GNOME/mutter/merge_requests/928
We currently assume that the actor_animate() helper function returns
a timeline. However Clutter may skip implicit animations and simple
set properties directly, for example when the actor is hidden.
The returned timeline will be NULL in that case, and we abort when
using it as instance parameter to g_signal_connect().
Fix this by only setting up a completed handler when we are actually
animating, and complete the effect directly otherwise.
https://gitlab.gnome.org/GNOME/mutter/merge_requests/925
The actor is already in surface coordinate space, so we should not scale
with the buffer scale to transform surface coordinates to stage
coordinates.
This bug causes input method using wayland text-input protocol to
receive wrong cursor location. Reproduced in ibus (when candidate
window is open) with scaling factor other than 1.
This commit also fixes pointer confinement.
https://gitlab.gnome.org/GNOME/mutter/merge_requests/915
Java applications might use override-redirect windows as parent windows for
top-level windows, although this is not following the standard [1].
In such case, the first non-override-redirect child window that is created
was marked as being on_all_workspaces since the call to
should_be_on_all_workspaces() returns TRUE for its parent, and this even
though the on_all_workspaces_requested bit is unset.
When a further child of this window was added, it was set as not having a
workspace and not being on_all_workspaces, since the call to
should_be_on_all_workspaces() for its parent would return FALSE (unless if
it is in a different monitor, and the multiple-monitors workspaces are
disabled).
Since per commit 09bab98b we don't recompute the workspace if the
on_all_workspaces bit is unset, we could end up in a case where a window can
be nor in all the workspaces or in a specific workspace.
So let's just ignore the transient_for bit for a window if that points to an
override-redirect, using the x11 root window instead.
Add a stacking test to verify this scenario (was failing before of this
commit).
Fixes https://gitlab.gnome.org/GNOME/mutter/issues/885https://gitlab.gnome.org/GNOME/mutter/merge_requests/895
[1] https://standards.freedesktop.org/wm-spec/wm-spec-latest.html#idm140200472512128
(cherry picked from commit bacbbbd628)
Once we set the transient_for, we look for parent MetaWindow, so instead
of overwriting this value for loops check, just use another function
and avoid to look for the xwindow again when setting the MetaWindow parent.
https://gitlab.gnome.org/GNOME/mutter/merge_requests/895
(cherry picked from commit c85fb107c0)
Override-redirect windows have no workspace by default, and can't be parent
of a top-level window, so we must check that the parent window is not an
O-R one when setting the workspace state.
https://gitlab.gnome.org/GNOME/mutter/merge_requests/895
(cherry picked from commit 2644e54c51)
We're expected by MetaWaylandSurface to always pick the frame callbacks
out from the pending state when committing (applying) so that no frame
callbacks are unaccounted for. We failed to do this if our actor for
some reason (e.g. associated window was unmanaged) was destroyed. To
handle this situation better, store away the frame callbacks until we
some later point in time need to pass them on forward.
https://gitlab.gnome.org/GNOME/mutter/merge_requests/893
(cherry picked from commit 0e5a5df5fe)
When using DesktopIcons extension and clicking in an icon, gnome-shell
starts an infinite loop caused by the first focus change that may trigger
on X11 a focus in/out event that leads to stage activation/deactivation
which never ends.
This happens because as part of meta_x11_display_set_input_focus_xwindow()
to focus the X11 stage window, we unset the display focus, but this also
causes to request the X11 display to unset the focus since we convolute by
calling meta_x11_display_set_input_focus() with no window, that leads to
focusing the no_focus_window and then a focus-in / focus-out dance that the
shell amplifies in order to give back the focus to the stage.
In order to fix this, mimic what meta_display_set_input_focus() does, but
without updating the X11 display, and so without implicitly calling
meta_x11_display_set_input_focus(), stopping the said convolution and
properly focusing the requested xwindow.
Also ensure that we're not doing this when using an older timestamp, since
this check isn't performed anymore.
Fixes https://gitlab.gnome.org/GNOME/mutter/issues/896
Fixes https://gitlab.gnome.org/GNOME/mutter/issues/899https://gitlab.gnome.org/GNOME/mutter/merge_requests/909
(cherry picked from commit efe5bed5b4)
Otherwise we'll end up trying to access the out of date state later.
Fixes the following test failure backtrace:
#0 _g_log_abort ()
#1 g_logv ()
#2 g_log ()
#3 meta_monitor_manager_get_logical_monitor_from_number ()
#4 meta_window_get_work_area_for_monitor ()
#5 meta_window_get_tile_area ()
#6 constrain_maximization ()
#7 do_all_constraints ()
#8 meta_window_constrain ()
#9 meta_window_move_resize_internal ()
#10 meta_window_tile ()
https://gitlab.gnome.org/GNOME/mutter/merge_requests/912
Create the intermediate shadow framebuffer for use exclusively when a
shadowfb is required.
Keep the previous offscreen framebuffer is as an intermediate
framebuffer for transformations only.
This way, we can apply transformations between in-memory framebuffers
prior to blit the result to screen, and achieve acceptable performance
even with software rendering on discrete GPU.
https://gitlab.gnome.org/GNOME/mutter/merge_requests/917
(cherry picked from commit 551641c74822ca2e3c685e49603836ebf5397df2)
This allows xdg_popup.grab() to work with styli. Without this check
we would bail out and emit xdg_popup.popup_done, leaving stylus users
unable to interact with popup menus, comboboxes, etc...
Closes: https://gitlab.gnome.org/GNOME/mutter/issues/886
When a touch sequence was rejected, the emulated pointer events would be
replayed with old timestamps. This caused issues with grabs as they
would be ignored due to being too old. This was mitigated by making sure
device event timestamps never travelled back in time by tampering with
any event that had a timestamp seemingly in the past.
This failed when the most recent timestamp that had been received were
much older than the timestamp of the new event. This could for example
happen when a session was left not interacted with for 40+ days or so;
when interacted with again, as any new timestamp would according to
XSERVER_TIME_IS_BEFORE() still be in the past compared to the "most
recent" one. The effect is that we'd always use the `latest_evtime` for
all new device events without ever updating it.
The end result of this was that passive grabs would become active when
interacted with, but would then newer be released, as the timestamps to
XIAllowEvents() would out of date, resulting in the desktop effectively
freezing, as the Shell would have an active pointer grab.
To avoid the situation where we get stuck with an old `latest_evtime`
timestamp, limit the tampering with device event timestamp to 1) only
pointer events, and 2) only during the replay sequence. The second part
is implemented by sending an asynchronous message via the X server after
rejecting a touch sequence, only potentially tampering with the device
event timestamps until the reply. This should avoid the stuck timestamp
as in those situations, we'll always have a relatively up to date
`latest_evtime` meaning XSERVER_TIME_IS_BEFORE() will not get confused.
https://gitlab.gnome.org/GNOME/mutter/merge_requests/886
In a similar vein to commit 8fd55fef85. This notably failed when setting
the focus on the stage (eg. to redirect key events to Clutter actors).
Deeper in MetaDisplay focus updating machinery, it would check
meta_stage_is_focused() which would still return FALSE at the time it's
called.
This would not typically have side effects, but our "App does not respond"
dialogs see the focus change under their feet, so they try to bring
themselves to focus again. This results in a feedback loop.
Changing the order results in later checks on the X11 POV of the focus
being correct, so focus is not mistakenly stolen from the close dialog,
and it actually succeeds in keeping the key focus.
Fixes: https://gitlab.gnome.org/GNOME/gnome-shell/issues/1607https://gitlab.gnome.org/GNOME/mutter/merge_requests/876
(cherry picked from commit 71c3f4af31)
Syncronized subsurfaces that call into `merge_pending_state` might
otherwise not create new destroy handlers, ending up with a invalid
handler ids, throwing errors and leaking.
https://gitlab.gnome.org/GNOME/mutter/merge_requests/868
(cherry picked from commit 98892391d7)
