Before linux_dmabuf v4 support was added, the BO were created using
gbm_bo_create_with_modifiers() which incidentally creates scanout
capable buffers.
We now need to replicate that explicitly when using the fallback path,
with buffers window, otherwise direct scanout will not be possible in
that case.
Signed-off-by: Olivier Fourdan <ofourdan@redhat.com>
Closes: https://gitlab.freedesktop.org/xorg/xserver/-/issues/1535
Suggested-by: Michel Dänzer <mdaenzer@redhat.com>
Reviewed-by: Michel Dänzer <mdaenzer@redhat.com>
Add the implementation for create_pixmap_for_window() in the GBM glamor
backend.
To do so, we just rename the existing xwl_glamor_gbm_create_pixmap() as
internal and add an optional drawable parameter, so that it can be used
either from the regular CreatePixmap code path, or from the new direct
Xwayland glamor's hook.
v2: Fallback to xwl_glamor_get_modifiers() if
xwl_glamor_get_drawable_modifiers() returned 0 modifiers. (Michel)
Signed-off-by: Olivier Fourdan <ofourdan@redhat.com>
Reviewed-by: Michel Dänzer <mdaenzer@redhat.com>
With linux dmabuf v4 support, for direct scanout support, we need more
context that just what CreatePixmap() provides, as we need the actual
drawable to invoke xwl_glamor_get_drawable_modifiers().
Add a specific hook in Xwayland's glamor implementation that we can use
for that purpose.
This is preparation work for the direct scanout fixes.
Signed-off-by: Olivier Fourdan <ofourdan@redhat.com>
Reviewed-by: Michel Dänzer <mdaenzer@redhat.com>
With implicit modifiers, DRM_FORMAT_MOD_INVALID is an allowed modifier,
to indicate that the server can support the format.
When looking for a scanout capable tranche with implicit modifiers, we
ought to check for the availability of a tranche with an invalid
modifier for the given format.
Signed-off-by: Olivier Fourdan <ofourdan@redhat.com>
Reviewed-by: Michel Dänzer <mdaenzer@redhat.com>
The helper function xwl_feedback_is_modifier_supported() walks all the
formats of a feeedback tranche and checks for format/modifier support
availability.
Add scanout support to that so that a caller can easily restrict the
tranches to those which support scanout.
This is preparation work for the implicit scanout support, no functional
change.
Signed-off-by: Olivier Fourdan <ofourdan@redhat.com>
Reviewed-by: Michel Dänzer <mdaenzer@redhat.com>
Separate the callbacks for the default's feedback from the one for
regular windows.
This is preparation work to recreate the window buffer of feedback
updates, no functional change.
Signed-off-by: Olivier Fourdan <ofourdan@redhat.com>
Reviewed-by: Michel Dänzer <mdaenzer@redhat.com>
ZDI-CAN-19866/CVE-2023-1393
If a client explicitly destroys the compositor overlay window (aka COW),
we would leave a dangling pointer to that window in the CompScreen
structure, which will trigger a use-after-free later.
Make sure to clear the CompScreen pointer to the COW when the latter gets
destroyed explicitly by the client.
This vulnerability was discovered by:
Jan-Niklas Sohn working with Trend Micro Zero Day Initiative
Signed-off-by: Olivier Fourdan <ofourdan@redhat.com>
Reviewed-by: Adam Jackson <ajax@redhat.com>
It could happen with the following call path:
frame_callback
xwl_present_frame_callback
xwl_present_msc_bump
xwl_present_execute
xwl_present_flip
xwl_window_create_frame_callback
The nested loop called xwl_present_reset_timer, which may end up calling
xorg_list_del for the entry after the one frame_callback started the
chain for. This resulted in the outer loop never terminating, because
its next element wasn't hooked up to the list anymore.
We avoid this by calling xwl_present_reset_timer as needed in
frame_callback, and bailing from xwl_window_create_frame_callback if it
was called from the former.
We also catch nested calls and FatalError if they ever happen again due
to another bug.
v2:
* Leave xwl_present_reset_timer call in xwl_present_frame_callback,
needed if xwl_present_msc_bump didn't hook up the window to the frame
callback list again.
Closes: https://gitlab.freedesktop.org/xorg/xserver/-/issues/1442
weston-info has been deprecated for quite some time, whereas wayland-info
may not be available yet.
So we use either, depending on what's actually available.
Signed-off-by: Olivier Fourdan <ofourdan@redhat.com>
Reviewed-by: Michel Dänzer <mdaenzer@redhat.com>
`glamor_make_current` is always called before any calls to GL.
Apply some dirty-tracking to whenever we call `glamor_make_current` so
that we can avoid a decent amount of redundant GL work on each
Dispatch cycle.
Gamescope previously was waking up an empty Xwayland server with an
XQueryPointer and I noticed a significant amount of churn doing
redundant GL work.
This has been addressed on the Gamescope side as well, but avoiding any
useless GL context switches and flushes when glamor is doing nothing
is still beneficial for CPU and power usage on portable devices.
Signed-off-by: Joshua Ashton <joshua@froggi.es>
Reviewed-by: Emma Anholt <emma@anholt.net>
Acked-by: Olivier Fourdan <ofourdan@redhat.com
Commit 3c07a01c4 (xwayland: Use xdg-output name for XRandR) changed the
logic to use a fixed sized buffer allocated on the stack to pass to
RROutputCreate() which would then copy it.
Valgrind complains about this:
== Conditional jump or move depends on uninitialised value(s)
== at 0x49954B: MakeAtom (atom.c:87)
== by 0x5108B3: RRMonitorCrtcName (rrmonitor.c:33)
== by 0x510BBB: RRMonitorSetFromServer (rrmonitor.c:92)
== by 0x511882: RRMonitorMakeList (rrmonitor.c:373)
== by 0x512175: ProcRRGetMonitors (rrmonitor.c:634)
== by 0x508091: ProcRRDispatch (randr.c:748)
== by 0x4A860E: Dispatch (dispatch.c:546)
== by 0x4B692F: dix_main (main.c:271)
== by 0x431C90: main (stubmain.c:34)
== Uninitialised value was created by a stack allocation
== at 0x42122C: xwl_output_create (xwayland-output.c:816)
This is actually harmless, but also simple to avoid by just initializing
the content of the array with zeros, so let's just fix that.
Signed-off-by: Olivier Fourdan <ofourdan@redhat.com>
Fixes: commit 3c07a01c4 - xwayland: Use xdg-output name for XRandR
If we allocated with implicit modifiers, then we shouldn't use the
modifier returned by gbm_bo when checking whether the modifier is
supported or not, since it won't be if the compositor only advertises
implicit modifiers, nor should we use the modifier when creating the
Wayland buffer object, as it wasn't explicitly advertised.
Fixes: c6f2598a4 ("xwayland: don't fall back to wl_drm with explicit modifier")
Signed-off-by: Jonas Ådahl <jadahl@gmail.com>
If we're using implicit modifiers, we'll pass NULL and zero modifiers.
Lets just use the legacy API directly instead.
Signed-off-by: Jonas Ådahl <jadahl@gmail.com>
To correctly render a window making use of SHAPE, a compositor
must query the shape rectangles. This may not be a desirable
feature for a Wayland compositor. Allow SHAPE to be turned off at
runtime, so that the compositor can opt-out.
Signed-off-by: Simon Ser <contact@emersion.fr>
The linux_dmabuf_v1 protocol doesn't guarantee any DRM node type:
the compositor may send a primary node or a render node. Use
drmDevice so that device comparisons are node-type-insensitive.
Signed-off-by: Simon Ser <contact@emersion.fr>
Closes: https://gitlab.freedesktop.org/xorg/xserver/-/issues/1447
With wl_pointer.axis_v120 support, the wl_seat supported version has
been bumped to 8, but Xwayland is still using MAP_SHARED which is
prohibited, wl_seat version 7 and above enforces the use of MAP_PRIVATE
for keymaps.
Use MAP_PRIVATE for the keymaps mmap().
Signed-off-by: Olivier Fourdan <ofourdan@redhat.com>
Closes: https://gitlab.freedesktop.org/xorg/xserver/-/issues/1512
Fixes: 3a02f56b4 - hook up wl_pointer.axis_v120 events
It's incorrect to strip an explicit modifier. Daniels' docs [1]
states:
> when importing a buffer, the user may supply `DRM_FORMAT_MOD_INVALID` as the
> buffer modifier (or not supply a modifier) to indicate that the modifier is
> unknown for whatever reason; this is only acceptable when the buffer has
> not been allocated with an explicit modifier
[1]: https://lore.kernel.org/dri-devel/20210905122742.86029-1-daniels@collabora.com/
Signed-off-by: Simon Ser <contact@emersion.fr>
When the modifier is not supported by the compositor, and the
DMA-BUF contains multiple planes, xwl_pixmap->buffer is NULL.
Avoid crashing when calling wl_buffer_add_listener().
Signed-off-by: Simon Ser <contact@emersion.fr>
With the addition of linux_dmabuf v4, the code adds dev_t in various
places but did not include <sys/types.h>.
While that works on glibc, it may fail to build on other libc
implementations such as musl libc.
Make sure to explicitly include <sys/types.h> where we use dev_t.
Signed-off-by: Olivier Fourdan <ofourdan@redhat.com>
Closes: https://gitlab.freedesktop.org/xorg/xserver/-/issues/1445
Fixes: bddfe190de - Implement linux_dmabuf_feedback event handlers
Adding new flip reasons after the TearFree ones would break the assumption
that `reason >= PRESENT_FLIP_REASON_DRIVER_TEARFREE` implies either of the
TearFree reasons. Document this in the PresentFlipReason enum in order to
save someone a very bad headache in the future.
Signed-off-by: Sultan Alsawaf <sultan@kerneltoast.com>
Reviewed-by: Martin Roukala <martin.roukala@mupuf.org>
The timing of PresentCompleteNotify events is inaccurate when a driver uses
TearFree because there's no way to know exactly when a presentation will
appear on the display without receiving a notification directly from the
driver indicating that the TearFree flip containing a presentation's pixmap
is actually visible on the display.
To fix the inaccurate PresentCompleteNotify timing, make use of the new
assumption that drivers which export TearFree permit a NULL pixmap to be
passed to their flip callback in order to make a presentation track the
exact TearFree flip responsible for rendering it onto the display.
Signed-off-by: Sultan Alsawaf <sultan@kerneltoast.com>
Acked-by: Martin Roukala <martin.roukala@mupuf.org>
It's possible for present_execute_copy to enqueue a vblank even when
TearFree is used, specifically when the present_queue_vblank in
present_scmd_pixmap fails and the subsequent vblank enqueue in
present_execute_copy somehow doesn't. This could happen if the DRM event
queue is exhausted when present_queue_vblank is called, but is no longer
exhausted by the time present_execute_copy is reached.
This exceedingly unlikely chain of events can lead to a vblank getting
enqueued a second time by the TearFree machinery in present_execute, which
is not good.
Although this scenario is very unlikely, prevent it by first checking that
the vblank wasn't enqueued by present_execute_copy before attempting to
enqueue it for TearFree.
Signed-off-by: Sultan Alsawaf <sultan@kerneltoast.com>
Acked-by: Martin Roukala <martin.roukala@mupuf.org>
When using TearFree, DRI clients have no way of accurately knowing when
their copied pixmaps appear on the display without utilizing the kernel
driver's notification indicating that the TearFree flip containing their
pixmap is complete. This is because the target CRTC's MSC can change while
the predicted completion MSC is calculated and even while the page flip
IOCTL is sent to the kernel due to scheduling delays and/or unfortunate
timing. Even worse, a page flip isn't actually guaranteed to be finished
after one vblank; it may be several MSCs until a flip actually finishes
depending on delays and load in hardware.
As a result, DRI clients may be off by one or more MSCs when they naively
expect their pixmaps to be visible at MSC+1 with TearFree enabled. This,
for example, makes it impossible for DRI clients to achieve precise A/V
synchronization when TearFree is enabled.
This change therefore adds a way for DRI clients to receive a notification
straight from the TearFree flip-done handler to know exactly when their
pixmaps appear on the display. This is done by checking for a NULL pixmap
pointer to modesetting's DRI flip routine, which indicates that the DRI
client has copied its pixmap and wants TearFree to send a notification when
the copied pixmap appears on the display as part of a TearFree flip. The
existing PageFlip scaffolding is reused to achieve this with minimal churn.
The Present extension will be updated in an upcoming change to utilize this
new mechanism for DRI clients' presentations.
Signed-off-by: Sultan Alsawaf <sultan@kerneltoast.com>
Acked-by: Martin Roukala <martin.roukala@mupuf.org>
It is possible for vblank events to run out of order with respect to one
another because the event which was queued to the kernel has the privilege
of running before all other events are handled. This allows kernel-queued
events to run before other, older events which should've run first.
Although this isn't a huge problem now, it will become more problematic
after the next change which ties DRI client notifications to TearFree page
flips. This increases the likelihood of DRI clients erroneously receiving
presentation-completion notifications out of order; i.e., a client could
receive a notification for a newer pixmap it submitted *before* receiving a
notification for an older pixmap.
Ensure vblank events always run in sequential order by removing the bias
towards kernel-queued events, and therefore forcing them to run at their
sequential position in the queue like other events.
Signed-off-by: Sultan Alsawaf <sultan@kerneltoast.com>
Reviewed-by: Martin Roukala <martin.roukala@mupuf.org>
There is more than one place with the confusing TearFree state check for a
CRTC. Instead of open-coding the TearFree check everywhere, introduce a
helper, ms_tearfree_is_active_on_crtc, to cover the TearFree state checks.
Suggested-by: Martin Roukala <martin.roukala@mupuf.org>
Signed-off-by: Sultan Alsawaf <sultan@kerneltoast.com>
Reviewed-by: Martin Roukala <martin.roukala@mupuf.org>
Check that the VT is owned and that the CRTC is on before exporting info to
Present stating that TearFree is available. Also, since `trf->buf[0].px` is
checked, the `ms->drmmode.tearfree_enable` check is redundant and can
therefore be removed.
Signed-off-by: Sultan Alsawaf <sultan@kerneltoast.com>
Reviewed-by: Martin Roukala <martin.roukala@mupuf.org>
The event allocation for ms_do_pageflip is leaked on error because callers
of ms_do_pageflip have no way of knowing whether or not a page flip
succeeded for any CRTCs. If a page flip succeeded for at least one CRTC,
then it's not safe for the caller to free the event allocation, and the
allocation won't be leaked. The event allocation is only leaked when not a
single CRTC's page flip succeeded.
Since all callers of ms_do_pageflip allocate the event pointer, and all of
them intentionally leak the event allocation when ms_do_pageflip returns an
error, just free the event pointer inside ms_do_pageflip when a page flip
doesn't succeed for any CRTC.
Signed-off-by: Sultan Alsawaf <sultan@kerneltoast.com>
Reviewed-by: Martin Roukala <martin.roukala@mupuf.org>
The CRTC pointer will soon be needed in the TearFree flip handlers, so pass
it in instead of passing in drmmode_tearfree_ptr.
No functional change.
Signed-off-by: Sultan Alsawaf <sultan@kerneltoast.com>
Reviewed-by: Martin Roukala <martin.roukala@mupuf.org>
Rather than passing the reference CRTC's vblank pipe to ms_do_pageflip,
just pass the pointer to the reference CRTC directly instead. This is
clearer and more useful than the vblank pipe, since the vblank pipe is only
used to identify whether or not a given CRTC is the reference CRTC.
No functional change.
Signed-off-by: Sultan Alsawaf <sultan@kerneltoast.com>
Reviewed-by: Martin Roukala <martin.roukala@mupuf.org>
This #ifdef is redundant since ms_do_pageflip is already enclosed within a
larger GLAMOR_HAS_GBM #ifdef.
No functional change.
Signed-off-by: Sultan Alsawaf <sultan@kerneltoast.com>
Reviewed-by: Martin Roukala <martin.roukala@mupuf.org>
This moves lnx_platform.c to the shared directory and adds it to
the BSD build. This is needed for PRIME render offload.
Acked-by: Olivier Fourdan <ofourdan@redhat.com>
This is to make sure the hardware gets the device states regardless
whether the internal state has changed or not, to overcome situations
that device LEDs are out of sync e.g. switching between VTs.
Signed-off-by: Yao Wei (魏銘廷) <yao.wei@canonical.com>
The increment sign wasn't taking into account when checking if the next
value is past our current value. The result was that for negative
increments, we kept looping indefinitely, locking up the server.
Easiest to reproduce with the evdev driver which has a negative
increment on the y axis.
Fixes 0a22502c34
dix: switch scroll button emulation to multiples of increment
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
The current algorithm triggers a bug in Xwayland when two devices have
different granularity of scrolling. In Xwayland, the scroll increment is
1 and all physical devices scroll through the same (x)wayland pointer
device.
This may cause events to get lost when changing devices:
- mouse scrolls by full increment, current value is 1.0
last scroll button was sent for valuator value 0.0,
delta is 1.0 and we emulate a button event.
- touchpad scrolls by partial increment, current value is 1.3
last scroll button was sent for valuator value 1.0, delta is 0.3
and no button event is emulated
- mouse scrolls by full increment, current value is -0.7,
last scroll button was sent for valuator value 1.0, delta is -0.7
and no button event is emulated
Thus the wheel event appears to get lost. Xwayland cannot reliably
detect this case because we don't see the physical devices.
We can work around this by instead emulating buttons whenever we cross
a multiple of increment. However, this has a drawback:
high-resolution scroll devices can now trigger a button event storm by
jittering across the multiple of increment. e.g. in the example above
the touchpad moving from 1.3 to 1.0 would cause a click, despite this
being a third of an increment.
Fixes#1339
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Acked-by: Olivier Fourdan <ofourdan@redhat.com>
storeLastValuators() takes the index in the mask for the x and y axis.
Completely pointless because any device that doesn't have x/y on 0 and
1, respectively, is going to break in fun ways anyway. And we only have
two callers two this function, both of which hardcode 0 and 1.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
This enables fixing a deadlock case on the client side, where the client
ends up blocked waiting for a Present event that will never come because
the window was destroyed. The new PresentWindowDestroyed flag allows the
client to avoid blocking indefinitely.
Signed-off-by: Adam Jackson <ajax@redhat.com>
See-also: https://gitlab.freedesktop.org/mesa/mesa/-/issues/116
See-also: https://gitlab.freedesktop.org/mesa/mesa/-/issues/6685
Reviewed-by: Michel Dänzer <mdaenzer@redhat.com>
This allows developers to setup Xwayland as a subproject of a
Wayland compositor, and have it correctly pick up the features
advertised in the dependency variables.
Signed-off-by: Simon Ser <contact@emersion.fr>
Remove the xwayland.pc.in file:
- This avoids writing down each pkg-config variable twice: once in
the Meson files to set the configuration data, once in the .pc.in
file to print it.
- We'll be able to re-use the same variables for use as a
subproject.
Signed-off-by: Simon Ser <contact@emersion.fr>
If wl_output provides us with an output name, use that as well.
If we have both xdg_output.name and wl_output.name (from version >= 4),
prefer the latter.
Signed-off-by: Olivier Fourdan <ofourdan@redhat.com>#
See-also: https://gitlab.freedesktop.org/wayland/wayland-protocols/-/merge_requests/189
Reviewed-by: Michel Dänzer <mdaenzer@redhat.com>
With the wl_output protocol, the actual bind to the interface is done in
xwl_output_create().
Pass the version number from the registry so we can bind to the minimum
version.
Signed-off-by: Olivier Fourdan <ofourdan@redhat.com>
Reviewed-by: Michel Dänzer <mdaenzer@redhat.com>
Currently, Xwayland assigns sequential output names for XRandR. When an
output is hotplugged, a new name is assigned sequentially (XWAYLAND0,
XWAYLAND1, etc.). This is a problem because if a monitor is unplugged
and plugged again, it will get a new name each time.
Luckily, xdg-output provides us with a name for the outputs.
Even though the protocol states that the name is not a reflection of the
underlying DRM connector name, it is to remain consistent across
sessions with the same hardware and software configuration.
So we could use the xdg-output name for the XRandR reported name for the
output.
Doing so is a bit tricky though, because the output name is set at
creation and is not supposed to change. The xdg-output event that
provides us with the name will come at a later time.
So we just allocate a default fixed size for the output name at creation
and just replace the default output name with the xdg-output name when
that is known.
Also, historically, some X11 clients were expecting output names in
Xwayland to be named XWAYLAND<x> and used that to check whether they
were running on Xwayland. Those clients should now use the Xwayland X11
extension which is designed specifically for that purpose.
Signed-off-by: Olivier Fourdan <ofourdan@redhat.com>
Closes: https://gitlab.freedesktop.org/xorg/xserver/-/issues/1353
See-also: https://gitlab.freedesktop.org/xorg/xserver/-/merge_requests/954
Reviewed-by: Michel Dänzer <mdaenzer@redhat.com>
Since commit 204f10c2, we notify XRandR clients that the randr
configuration has changes as soon as an new output is created.
Yet, this might be premature, considering that at that point, we are
still to receive the wl_output and xdg-output events that will most
likely change the setup.
So instead of calling RRTellChanged() from xwl_output_create(), wait
until we get to call apply_output_change(), which occurs after the done
events from both xdg-output and wl_output are received.
Signed-off-by: Olivier Fourdan <ofourdan@redhat.com>
Reviewed-by: Michel Dänzer <mdaenzer@redhat.com>
Olivier Fourdan
Benno Schulenberg
Michel Dänzer
Joshua Ashton
Alan Coopersmith
Jonas Ådahl
Simon Ser
Austin Shafer
Sultan Alsawaf
Yao Wei
Peter Hutterer
Adam Jackson