This fixes modesetting driver build failure which can be triggered with
the following configure options:
$ ./configure --disable-dri --disable-dri2 --disable-dri3
--disable-config-udev --enable-xorg
Bugzilla: https://bugs.gentoo.org/689768
Signed-off-by: Alexander Tsoy <alexander@tsoy.me>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
(cherry picked from commit 334f110714)
When compiling with link time optimization, GCC thinks it's discovered
undefined behavior:
events.c: In function 'XineramaConfineCursorToWindow':
events.c:609:13: warning: iteration 2147483647 invokes undefined behavior [-Waggressive-loop-optimizations]
events.c:609:11: note: within this loop
events.c:605:49: warning: array subscript -1 is below array bounds of 'struct _Window *[16]' [-Warray-bounds]
events.c:606:31: warning: array subscript -1 is below array bounds of 'struct _Screen *[16]' [-Warray-bounds]
events.c:610:39: warning: array subscript -2 is below array bounds of 'struct _Screen *[16]' [-Warray-bounds]
events.c:617:38: warning: array subscript -2 is below array bounds of 'struct _Window *[16]' [-Warray-bounds]
events.c:619:35: warning: array subscript -2 is below array bounds of 'struct _Screen *[16]' [-Warray-bounds]
This results from
i = PanoramiXNumScreens - 1;
RegionCopy(&pSprite->Reg1, &pSprite->windows[i]->borderSize);
off_x = screenInfo.screens[i]->x;
off_y = screenInfo.screens[i]->y;
where GCC believes that PanoramiXNumScreens might be 0. Unfortunately
GCC is just smart enough to be an annoyance because this case is not
actually possible: XineramaConfineCursorToWindow() is only called when
noPanoramiXExtension is false, and if noPanoramiXExtension is false then
PanoramiXNumScreens must be >1 (see PanoramiXExtensionInit()).
So, add an assert(!noPanoramiXExtension), which to my surprise provides
GCC with information even in release builds and lets GCC understand that
the code is not doing anything that is undefined behavior.
I chose this solution instead of the proposed assert(i >= 0) because the
same pattern occurs in CheckVirtualMotion() but is inside an
'if (!noPanoramiXExtension)' and does not generate any warnings.
Fixes: xorg/xserver#590
Signed-off-by: Matt Turner <mattst88@gmail.com>
(cherry picked from commit 61aa40aeb3)
It doesn't require shared memory dir and thus allows
to avoid cases when this dir is detected incorrectly,
as in https://bugreports.qt.io/browse/QTBUG-71440
Signed-off-by: Alexander Volkov <a.volkov@rusbitech.ru>
(cherry picked from commit f6753c117e)
Prodding the builder's filesystem for tmp dirs doesn't necessarily
tell you anything about what the actual host's filesystem is going to
look like, so we should just try the dirs at runtime.
Signed-off-by: Eric Anholt <eric@anholt.net>
(cherry picked from commit 19f6cb570b)
Since group_info and width are used for the key actions allocations,
when modifying them we need to take care of reallocation key actions if
needed.
(cherry picked from commit fabc421962)
With a 32-bit build, putting the initialized field at the end of the
struct bumped the struct size from 20 bytes to 24, changing the layout
of other structs embedding struct _SyncObject. While this would be
acceptable on master, it caused crashes with 1.20.
Making the initialized field a char and putting it in the hole before
the beingDestroyed field restores the 32-bit ABI as well.
Fixes https://gitlab.freedesktop.org/xorg/xserver/issues/892
Reviewed-by: Aaron Plattner <aplattner@nvidia.com>
Reviewed-by: Alex Goins <agoins@nvidia.com>
(cherry picked from commit 0d8cc7bb18)
This fixes 'non-desktop' displays staying powered on after their lease
has been revoked.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=111620
Cc: Keith Packard <keithp@keithp.com>
Signed-off-by: Andres Rodriguez <andresx7@gmail.com>
If Xwayland gets to realize a window meant for composition before the
compositor redirected windows (i.e. redirect mode is not RedirectDrawManual
yet), the window would stay "invisible" as we wouldn't create a
wl_surface/wl_shell_surface for it at any later point.
This scenario may happen if the wayland compositor sets up a X11 socket
upfront, but waits to raise Xwayland until there are X11 clients. In this
case the first data on the socket is the client's, the compositor can hardly
beat that in order to redirect subwindows before the client realizes a
Window.
In order to jump across this hurdle, allow the late creation of a matching
(shell) surface for the WindowPtr on SetWindowPixmapProc, so it is ensured
to be created after the compositor set up redirection.
Signed-off-by: Carlos Garnacho <carlosg@gnome.org>
Reviewed-by: Michel Dänzer <mdaenzer@redhat.com>
Reviewed-by: Olivier Fourdan <ofourdan@redhat.com>
(cherry picked from commit 78cc8b6f96)
This is just called from xwl_window_realize() ATM, but will be useful in
future commits.
Signed-off-by: Carlos Garnacho <carlosg@gnome.org>
(cherry picked from commit c2e8ae9640)
This will be dissociated in future commits to handle the cases
where windows are being realized before there is a compositor
handling redirection.
In that case, we still want the DamagePtr to be registered upfront
on RealizeWindowProc before a corresponding xwl_window might be
created. Most notably, it cannot be lazily created on
SetWindowPixmapProc as damage accounting gets broken.
Signed-off-by: Carlos Garnacho <carlosg@gnome.org>
(cherry picked from commit 4e50440ae2)
vnd has already verified that the context tag is valid before this gets
called, and we only set the context tag private data to non-null for
indirect clients. Mesa happens to be buggy and doesn't send MakeCurrent
requests nearly as much as it should for direct contexts, but if you fix
that, then unbinding a direct context would fail here with
GLXBadContextTag.
Sadly Mesa will still need to carry a workaround here for broken
servers, but we should still fix the server.
(cherry picked from commit 95dcc81cb1)
The initialized field was added in:
commit 82f01ad786
Author: Alex Goins <agoins@nvidia.com>
Date: Wed Apr 10 13:48:02 2019 -0500
xsync: Add resource inside of SyncCreate, export SyncCreate
But it added this field not at the end of SyncObject. It may not have
been _usefully_ possible to create those from another extension prior to
that commit, but that's still an ABI-incompatible change.
(cherry picked from commit 194ba38728)
Commit 56c0a71fdd incremented the
GLXSERVER_VENDOR_ABI_MINOR_VERSION define, but this define was not actually
being used to set glxServer.minorVersion.
Update the initializer for glxServer to use the correct version numbers.
Signed-off-by: Aaron Plattner <aplattner@nvidia.com>
(cherry picked from commit b4231d6902)
Add a new function, GlxServerExports::setClientScreenVendor, which will change
the vendor that handles GLX requests for a screen, but only for requests from
a specific client.
v2: Increment the GLXVND minor version number.
v3: Note the GLXVND version requirement for setClientScreenVendor.
Signed-off-by: Aaron Plattner <aplattner@nvidia.com>
Reviewed-by: Aaron Plattner <aplattner@nvidia.com>
Reviewed-by: Adam Jackson <ajax@redhat.com>
(cherry picked from commit 56c0a71fdd)
When GlxGetXIDMap looks up an unknown XID, it will now look up a vendor based
on the screen number for the XID and the client that sent the current request.
In GlxGetXIDMap, if the XID is for a regular X window, then it won't be in the
(XID -> vendor) mapping, so we have to look up a vendor by screen number.
With this change, GlxGetXIDMap will use the (screen -> vendor) map for
whichever client sent the current request, instead of using the global
(screen -> vendor) map.
Since GlxGetXIDMap doesn't take a ClientPtr argument, GlxDispatchRequest will
store the client for the current request in a global variable. That way, the
ABI for GLXVND doesn't need to change.
v2: Fix an error check in GlxDispatchRequest.
Signed-off-by: Aaron Plattner <aplattner@nvidia.com>
Reviewed-by: Aaron Plattner <aplattner@nvidia.com>
Reviewed-by: Adam Jackson <ajax@redhat.com>
(cherry picked from commit 8b67ec7cc6)
Each client now has its own (screen, vendor) mapping.
Currently, it's just a copy of the global mapping, but later changes will allow
it to change.
Signed-off-by: Aaron Plattner <aplattner@nvidia.com>
Reviewed-by: Aaron Plattner <aplattner@nvidia.com>
Reviewed-by: Adam Jackson <ajax@redhat.com>
(cherry picked from commit 37a36a6b5b)
As shown by DRI3 adding the SyncCreateFenceFromFD() function, extensions may
want to create a fence, then initialize it in their own way. This currently
can't be done without adding a function directly to Xext/sync.c due to the fact
that the RTFence resource type is private and there is no external interface to
add to it.
To facilitate other X extensions creating fences and initializing them, this
change exports SyncCreate() and adds the resource directly within it. Callers no
longer need to call AddResource() after SyncCreate(), they only need to
initialize the SyncObject.
To prevent FreeFence() and FreeCounter() from segfaulting if the call to
AddResource() fails before the sync object is initialized, this adds a new
'initialized' parameter to SyncObject that, when FALSE, causes FreeFence() and
FreeCounter() to skip de-initialization and simply free the object.
Initialization after adding the resource shouldn't otherwise be a problem due to
the single-threaded nature of X.
Signed-off-by: Alex Goins <agoins@nvidia.com>
Reviewed-by: James Jones <jajones@nvidia.com>
Signed-off-by: Aaron Plattner <aplattner@nvidia.com>
(cherry picked from commit 7f962c70b6)
Both `gbm_bo_create()` and `gbm_bo_create_with_modifiers()` can fail and
return `NULL`.
If that occurs, `xwl_glamor_gbm_create_pixmap()` will not create a
pixmap for the (NULL) GBM bo, but would still try to free the bo which
leads to a crash in mesa:
[...]
#7 <signal handler called>
#8 in gbm_bo_destroy (bo=0x0) at ../src/gbm/main/gbm.c:439
#9 in xwl_glamor_gbm_create_pixmap () at xwayland-glamor-gbm.c:245
#10 in ProcCreatePixmap () at dispatch.c:1440
#11 in Dispatch () at dispatch.c:478
#12 in dix_main () at main.c:276
To avoid the crash, only free the GBM bo if not `NULL`.
Signed-off-by: Olivier Fourdan <ofourdan@redhat.com>
Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>
Bugzilla: https://bugzilla.redhat.com/1729925
(cherry picked from commit d9ec525059)
There's not really a good way to query this from the wayland server, so
just set the maximum to the X11 protocol limits. While we're at it,
lower the minimum screen size to something implausibly small too, just
in case.
Fixes: xorg/xserver#850
Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>
(cherry picked from commit d0850241c6)
Running Xwayland non-rootless and resizing the output would lead to a
crash while trying to update the larger areas of the root window.
Make sure we resize the backing pixmap according to the new output size
to avoid the crash.
Signed-off-by: Olivier Fourdan <ofourdan@redhat.com>
Closes: https://gitlab.freedesktop.org/xorg/xserver/issues/834
Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>
(cherry picked from commit ce9455b5ee)
Copied from Mesa with no modifications.
This update brings in a significant number of new platform ID's.
Syncs with mesa up to commit e334a595e ("intel/icl: Add new ICL
PCI-IDs").
Signed-off-by: Matt Roper <matthew.d.roper@intel.com>
(cherry picked from commit a8d9ebeb43)
When Retrieving touch delivery data we need to check if we have an active
grab on such device, and in that case use it to delivery events.
If we don't do this, when rejecting the touch events in DeactivatePointerGrab,
we will end-up in creating an implicit grab that will change the device
deviceGrab's state, causing a recursion during TouchEndTouch.
Fixes#7https://bugs.freedesktop.org/show_bug.cgi?id=96536
(cherry picked from commit 35e5a76cc1)
This reverts commit ef91da2757.
I thought this couldn't work under any circumstances, but I was wrong,
and drivers are already checking for cases that really cannot work.
Fixes issue #839.
(cherry picked from commit 2f0dd1816d)
Hiding the tablet tool cursor results in it being hidden forever after.
This is due to the stale frame callback that will neither be disposed
or replaced. This can be reproduced in krita (X11) as the pointer
cursor is hidden while over the canvas.
Clearing the frame callback ensures the correct behavior in future
xwl_tablet_tool_set_cursor() calls (i.e. a new cursor surface being
displayed, and a new frame callback created), and is 1:1
with xwl_seat_set_cursor() for pointers.
Signed-off-by: Carlos Garnacho <carlosg@gnome.org>
(cherry picked from commit dea4a74621)
If a pixmap is not exportable, `glamor_gbm_bo_from_pixmap()` would fail
and the modesettings driver would consequently fail to do its page flip,
which both prevents Present from working and also fill up the logs with
error messages such as:
(EE) modeset(0): Failed to get GBM bo for flip to new front.
(EE) modeset(0): present flip failed
Refactor the code so that `glamor_gbm_bo_from_pixmap()` takes care of
making the pixmap exportable.
Signed-off-by: Olivier Fourdan <ofourdan@redhat.com>
Signed-off-by: Yuxuan Shui yshui@hadean.com
See-also: https://gitlab.freedesktop.org/xorg/xserver/merge_requests/131
Closes: https://gitlab.freedesktop.org/xorg/xserver/issues/68
Fixes: 86b2d8740a "glamor: Reallocate pixmap storage without modifiers
if necessary"
(cherry picked from commit 26fe29f4fa)
The current code in `xwl_glamor_gbm_create_pixmap_for_bo()` may fail in
several cases that are not checked for:
- `eglCreateImageKHR()` may have failed to create the image,
- `glEGLImageTargetTexture2DOES()` may fail and set an error,
- `glamor_set_pixmap_texture()` may fail for very large pixmaps
because the corresponding FBO could not be created.
Trying to upload content to a pixmap with no texture will crash Mesa,
glamor and Xwayland, e.g.:
XXX fail to create fbo.
(EE)
(EE) Backtrace:
(EE) 0: Xwayland (OsSigHandler+0x29)
(EE) 1: libpthread.so.0 (funlockfile+0x50)
(EE) 2: libc.so.6 (__memmove_avx_unaligned_erms+0x215)
(EE) 3: dri/i965_dri.so (_mesa_format_convert+0xab3)
(EE) 4: dri/i965_dri.so (_mesa_texstore+0x205)
(EE) 5: dri/i965_dri.so (store_texsubimage+0x28c)
(EE) 6: dri/i965_dri.so (intel_upload_tex+0x13b)
(EE) 7: dri/i965_dri.so (texture_sub_image+0x134)
(EE) 8: dri/i965_dri.so (texsubimage_err+0x150)
(EE) 9: dri/i965_dri.so (_mesa_TexSubImage2D+0x48)
(EE) 10: Xwayland (glamor_upload_boxes+0x246)
(EE) 11: Xwayland (glamor_copy+0x4d1)
(EE) 12: Xwayland (miCopyRegion+0x96)
(EE) 13: Xwayland (miDoCopy+0x43c)
(EE) 14: Xwayland (glamor_copy_area+0x24)
(EE) 15: Xwayland (damageCopyArea+0xba)
(EE) 16: Xwayland (compCopyWindow+0x31c)
(EE) 17: Xwayland (damageCopyWindow+0xd3)
(EE) 18: Xwayland (miResizeWindow+0x7b7)
(EE) 19: Xwayland (compResizeWindow+0x3a)
(EE) 20: Xwayland (ConfigureWindow+0xa96)
(EE) 21: Xwayland (ProcConfigureWindow+0x7d)
(EE) 22: Xwayland (Dispatch+0x320)
(EE) 23: Xwayland (dix_main+0x366)
(EE) 24: libc.so.6 (__libc_start_main+0xf3)
(EE) 25: Xwayland (_start+0x2e)
(EE)
Fatal server error:
(EE) Caught signal 11 (Segmentation fault). Server aborting
(EE)
Check for the possible cases of failure above and fallback to the
regular glamor pixmap creation when an error is detected.
Signed-off-by: Olivier Fourdan <ofourdan@redhat.com>
Closes: https://gitlab.freedesktop.org/xorg/xserver/issues/661
(cherry picked from commit fc6380a11b)
On pointer enter notification, Xwayland checks for an existing pointer
warp with a `NULL` sprite.
In turn, `xwl_pointer_warp_emulator_maybe_lock()` checks for an existing
grab and the destination window using `XYToWindow()` which does not
check for the actual sprite not being `NULL`.
So, in some cases, when the pointer enters the surface and there is an
existing X11 grab which is not an ownerEvents grab, Xwayland would crash
trying to dereference the `NULL` sprite pointer:
#0 __GI_raise ()
#1 __GI_abort () at abort.c:79
#2 OsAbort () at utils.c:1351
#3 AbortServer () at log.c:879
#4 FatalError () at log.c:1017
#5 OsSigHandler () at osinit.c:156
#6 OsSigHandler () at osinit.c:110
#7 <signal handler called>
#8 XYToWindow (pSprite=0x0, x=0, y=0) at events.c:2880
#9 xwl_pointer_warp_emulator_maybe_lock () at xwayland-input.c:2673
#10 pointer_handle_enter () at xwayland-input.c:434
Avoid the crash by simply checking for the sprite being not `NULL` in
`xwl_pointer_warp_emulator_maybe_lock()`
Signed-off-by: Olivier Fourdan <ofourdan@redhat.com>
Bugzilla: https://bugzilla.redhat.com/1708119
(cherry picked from commit 0a07446318)
Terms:
dev->last.valuator[] is the last value given to us by the driver
dev->valuator.axisVal[] is the last value sent to the client
dev->last.scroll[] is the abs value of the scroll axis as given by the driver,
used for button emulation calculation (and the remainder)
This function updates the device's last.valuator state based on the current
master axis state. This way, relative motion continues fluidly when switching
between devices. Before mouse 2 comes into effect, it's valuator state is
updated to wherever the pointer currently is so the relative event applies on
top of that.
This can only work for x/y axes, all other axes aren't guaranteed to have the
same meaning and/or may not be present:
- xtest device: no valuator 2
- mouse: valuator 2 is horizontal scroll axis
- tablet: valuator 2 is pressure
Scaling the current value from the pressure range into the range for
horizontal scrolling makes no sense. And it causes scroll jumps:
- scroll down, last.valuator == axisVal == 20
- xdotool click 1, the XTest device doesn't have that valuator
- scroll up
- updateSlaveDeviceCoords reset last.valuator to 0 (axisVal == 20)
- DeviceClassesChangedEvent includes value 20 for the axis
- event is processed, last.value changes from 0 to -1
- axisVal is updated to -1, causing a jump of -21
The same applies when we switch from tablet to mouse wheel if the pressure
value is 0 on proximity out (basically guaranteed). So let's drop this code
altogether and only leave the scaling for the relative x/y motion.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
(cherry picked from commit d7b1753d44)
If `_glamor_create_tex()` fails to allocate the FBO because of
GL_OUT_OF_MEMORY error, the `pixmap_priv->fbo` is NULL.
However, `glamor_get_pixmap_texture()` doesn't actually check whether
the `pixmap_priv->fbo` is NULL and will segfault with a NULL pointer
dereference trying to access the `pixmap_priv->fbo->tex`.
Signed-off-by: Olivier Fourdan <ofourdan@redhat.com>
Closes: https://gitlab.freedesktop.org/xorg/xserver/issues/647
(Cherry picked from commit 74479a99ec)
wl_drm's protocol "device" event provides the path to the DRM device,
which may not be a render node, thus causing Xwayland to fall back to
DRM authentication which may fail if the user has switched to another
VT while Xwayland is starting.
Search for a render node corresponding to the given DRM device and try
to use it instead, as render nodes do not need DRM authentication and
Xwayland can make use of them if it can find one.
Closes: https://bugs.freedesktop.org/108038
Signed-off-by: Olivier Fourdan <ofourdan@redhat.com>
(Cherry picked from commit 361894497c)
If they don't, flipping will result in corrupted display.
Test case:
* Run Xorg at 1920x1080 with no window manager
* glxgears -geometry 2048x1080
(Cherry picked from commit ef91da2757)
xwl_present_cleanup frees the struct xwl_present_window memory,
so if there's a pending callback, we have to destroy it to prevent
use-after-free in xwl_present_sync_callback.
Should fix issue #645.
Reviewed-by: Olivier Fourdan <ofourdan@redhat.com>
(Cherry picked from commit 2e18eec6f0)
This reverts commit 8694395fcf.
Some scenarios have come to light where this failed to ensure the docker
image exists:
* If the master branch of a forked repository is used for an MR which
doesn't modify .gitlab-ci.yml, the docker-image job may not run.
* If the docker-image job of the first pipeline in a forked repository
is cancelled or fails for any reason, and .gitlab-ci.yml isn't
modified for the next pipeline run.
(Cherry picked from commit 0ec9a13c2a)
If the server resets, most client workqueues are cleaned up as the
clients are killed.
The one exception is the server's client, which is exempt from
the killing spree.
If that client has a queued work procedure active, it won't get
cleared on reset.
This commit ensures it gets cleared too.
(cherry picked from commit 8738ce85df)
Fixes: xorg/xserver#670
We don't set the PIGLIT_DIR and XTS_DIR environment variables for make
distcheck for now, otherwise it complains about files left behind by
piglit.
(cherry picked from commit 432fad04e7)
Instead of testing window->redirectDraw.
With Xwayland, the toplevel window is always redirected, so this would
unnecessarily preclude flipping there in some cases, e.g. with wlroots
based Wayland compositors or with fullscreen X11 windows in weston.
Fixes issue #631.
(Cherry picked from commit a093a88531)
Otherwise there's normally no need to run it. It will also run when a
new branch is created, which ensures that the docker image always exists
(e.g. in a newly forked repository).
Inspired by https://gitlab.freedesktop.org/mesa/mesa/merge_requests/143
(Cherry picked from commit 8694395fcf)
Although piglit could now handle non-ASCII characters in the
environment, meson was still failing without this (even though it's
using Python 3).
Reviewed-by: Eric Anholt <eric@anholt.net>
(Cherry picked from commit bc6998b728)
The latter use Python 2 and break with any non-ASCII characters in the
environment, the former uses Python 3 and works fine in that case.
Reviewed-by: Eric Anholt <eric@anholt.net>
(Cherry picked from commit 4aaaf69229)
This removes the dependency on an externally generated docker image, and
should make it easier to update the docker image or make other changes
related to it.
This is based on Debian testing, because I'm most familiar with Debian.
But it should be easy to base it on another distro.
v2:
* Use kaniko instead of docker-in-docker for image generation, so it can
also work in unprivileged runners.
* Drop piglit.conf & tetexec.cfg overrides, just make sure the files in
the image work.
(Cherry picked from commit f56d8e2282)
