It was passing O_CLOEXEC as permission bits instead of as a flag.
Signed-off-by: Michel Dänzer <michel.daenzer@amd.com>
Reviewed-by: Keith Packard <keithp@keithp.com>
As we are not freeing elements while iterating the list of timers, we
can forgo using the safe variant, and reduce the number of pointer
dances required for the insertion sort.
Reviewed-by: Adam Jackson <ajax@redhat.com>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Currently, we use xorg_list_add(new, head->prev) which is functionaly
equivalent to xorg_list_append(), but with more pointer chasing, so
reduce the strain on the reader and compiler by using the simpler
append().
Reviewed-by: Adam Jackson <ajax@redhat.com>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Hi,
I upgraded Xwayland and the assorted libraries from git masters today,
and noticed that glamor wouldn't work anymore on i.MX6/etnaviv. The
error was:
No provider of glVertexAttribDivisor found. Requires one of:
Desktop OpenGL 3.3
OpenGL ES 3.0
GL extension "GL_ANGLE_instanced_arrays"
GL extension "GL_ARB_instanced_arrays"
GL extension "GL_EXT_instanced_arrays"
GL extension "GL_NV_instanced_arrays"
The problem is that etnaviv offers GLSL 140 on GL 2.1 and glamor
rendering assumes that glVertexAttribDivisor() is always available on
GLSL>=130, which is not the case here. Forcing GLSL 120 makes glamor
work fine again on this platform. After chatting with ajax in
#xorg-devel, the following solution was proposed.
This is my first time of submitting a patch, so please excuse me and
advise if I'm doing it wrong ;)
Cheers
Lukas (mntmn)
Reviewed-by: Eric Anholt <eric@anholt.net>
Fixes DRI2 client driver name mapping for newer AMD GPUs with the
modesetting driver, allowing the DRI2 extension to initialize.
Fixes using GL with the modesetting driver for me.
Seems we were way behind on this one, time to look into something
more scalable?
Signed-off-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Instead, substitute the same values as autotools does
Reviewed-by: Adam Jackson <ajax@redhat.com>
Signed-off-by: Jon Turney <jon.turney@dronecode.org.uk>
meson.build has code to set the module_dir variable to
${libdir}/xorg/modules if the module_dir option string is empty.
However, this has several problems:
1. The variable is only used for an unused @moduledir@ substitution in
the man page. The rule for xorg-server.pc uses option('module_dir')
directly instead.
2. The 'module_dir' option has a default value of 'xorg/modules' so the
above rule doesn't do anything by default.
3. The xorg-server.pc rule uses ${exec_prefix}/option('module_dir'), so
the effect of #2 is that the default moduledir is different between
autoconf and meson. E.g. if ${prefix} is /X, then you get
autoconf: moduledir=/X/lib/xorg/modules
meson: moduledir=/X/xorg/modules
Fix this by using the module_dir variable when generating xorg-server.pc, and by
using join_paths() to assign module_dir unconditionally.
v2: Keep the 'xorg/modules' default path, but use join_paths() unconditionally (Thierry Reding)
Signed-off-by: Aaron Plattner <aplattner@nvidia.com>
Reviewed-by: Peter Hutterer <peter.hutterer@who-t.net>
Modifiers support needs gbm as a dependency. Without setting the dependency
included headers are not found reliably and the build might fail if the
headers are not placed in the default system include paths.
Signed-off-by: Roman Gilg <subdiff@gmail.com>
Reviewed-by: Adam Jackson <ajax@redhat.com>
Because this is an automakefile, things inside @@ get expanded, which
means your sed ends up saying s|/var/log|/var/log| and your manual pages
still have @logdir@ in them. Fix this by hiding the @s inside a trivial
character range, which keeps the pattern preserved all the way into the
Makefile.
Signed-off-by: Adam Jackson <ajax@redhat.com>
Reviewed-by: Alan Coopersmith <alan.coopersmith@oracle.com>
The code would fall through to the PresentIdleNotify case, and nothing
good would come of it.
Signed-off-by: Adam Jackson <ajax@redhat.com>
Reviewed-by: Alan Coopersmith <alan.coopersmith@oracle.com>
We probably don't want a fake crtc to be visible to clients, and we
definitely don't want to generate events every time we create such a
fake (which would happen as a side effect from RRCrtcCreate hitting
RRTellChanged). As it happens we're not actually using that crtc for
anything because xwayland doesn't store any state on the crtc object,
so it suffices to use the real crtc for the screen.
Signed-off-by: Adam Jackson <ajax@redhat.com>
Tested-by: Roman Gilg <subdiff@gmail.com>
Reviewed-by: Roman Gilg <subdiff@gmail.com>
Again, this was causing 32-bit build warnings due to mixing CARD* and
stdint.
Signed-off-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Adam Jackson <ajax@redhat.com>
We were mixing stdint and CARD* types, causing compiler warnings on
32-bit. Just switch over to stdint, which is what we'd like the server
to be using long term, anyway.
Reviewed-by: Adam Jackson <ajax@redhat.com>
We were mixing stdint and CARD* types, causing compiler warnings on
32-bit. Just switch over to stdint, which is what we'd like the server
to be using long term, anyway.
Signed-off-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Adam Jackson <ajax@redhat.com>
Instead of reusing xwl_window introduce a persistent window struct for every
window, that asks for Present flips.
This struct saves all relevant data and is only freed on window destroy.
Signed-off-by: Roman Gilg <subdiff@gmail.com>
Tested-by: Olivier Fourdan <ofourdan@redhat.com>
The old 32-Bit wraparound handling didn't actually work, due to some
integer casting bug, and the mapping was ill equipped to deal with input
from the new true 64-bit GetCrtcSequence/QueueCrtcSequence api's
introduced in Linux 4.15.
For 32-Bit truncated input from pageflip events and old vblank events
and old drmWaitVblank ioctl, implement new wraparound handling, which
also allows to deal with wraparound in the other direction, e.g., if a
32-Bit truncated sequence value is passed in, whose true 64-Bit
in-kernel hw value is within 2^30 counts of the previous processed
value, but whose 32-bit truncated sequence value happens to lie just
above or below a 2^32 boundary, iow. one of the two values 'sequence'
vs. 'msc_prev' lies above a 2^32 border, the other one below it.
The method is directly translated from Mesa's proven implementation of
the INTEL_swap_events extension, where a true underlying 64-Bit wide
swapbuffers count (SBC) needs to get reconstructed from a 32-Bit LSB
truncated SBC transported over the X11 protocol wire. Same conditions
apply, ie. successive true 64-Bit SBC values are close to each other,
but don't always get received in strictly monotonically increasing
order. See Mesa commit cc5ddd584d17abd422ae4d8e83805969485740d9 ("glx:
Handle out-of-sequence swap completion events correctly. (v2)") for
explanation.
Additionally add a separate path for true 64-bit msc input originating
from Linux 4.15+ drmCrtcGetSequence/QueueSequence ioctl's and
corresponding 64-bit vblank events. True 64-bit msc's don't need
remapping and must be passed through.
As a reliability bonus, they are also used here to update the tracking
values msc_prev and ms_high with perfect 64-Bit ground truth as baseline
for mapping msc from pageflip completion events, because pageflip events
are always 32-bit wide, even when the new kernel api's are used. Because
each pageflip(-event) is always preceeded close in time (and vblank
count) by a drmCrtcQueueSequence queued event or drmCrtcGetSequence
query as part of DRI2 or DRI3+Present swap scheduling, we can be certain
that each pageflip event will get its truncated 32-bit msc remapped
reliably to the true 64-bit msc of flip completion whenever the sequence
api is available, ie. on Linux 4.15 or later.
Note: In principle at least the 32-bit mapping path could also be
backported to earlier server branches, as this seems to be broken for at
least server 1.16 to 1.19.
Signed-off-by: Mario Kleiner <mario.kleiner.de@gmail.com>
Reviewed-by: Adam Jackson <ajax@redhat.com>
Cc: Keith Packard <keithp@keithp.com>
Cc: Michel Dänzer <michel.daenzer@amd.com>
The function is ported from intel-ddx uxa backend around 2013, where its
stated purpose was to apply a vblank_offset to msc values to correct for
problems with those kernel provided msc values. Some (somewhat magic and
puzzling to myself) heuristic tried to guess if provided values were
unreasonable and tried to adapt the corrective vblank_offset to account
for that.
Except: It wasn't applied to kernel provided msc values, but the values
delivered by clients via DRI2 or Present, so valid client targetmsc
values, e.g., requesting a vblank event > 1000 vblanks in the future,
triggered the offset correction in arbitrarily wrong ways, leading to
wrong msc values being returned and thereby vblank events queued to the
kernel for the wrong time. This causes glXSwapBuffersMscOML and
glXWaitForMscOML to swap / return immediately whenever a swap/wait in >
1000 vblanks is requested.
The original code was also written to only deal with 32 bit mscs, but
server 1.20 modesetting ddx can now use new Linux 4.15+ kernel vblank
api to process true 64 bit msc's, which may confuse the heuristic even
more due to 32 bit integer truncation/wrapping.
This code caused various problems in the intel-ddx in the past since
year 2013, and was removed there in 2015 by Chris Wilson in commit
42ebe2ef9646be5c4586868cf332b4cd79bb4618:
" uxa: Remove the filtering of bogus Present MSC values
If the intention was to filter the return values from the kernel, the
filtering would have been applied to the kernel values and not to the
incoming values from Present. This filtering introduces crazy integer
promotion and truncation bugs all because Present feeds garbage into its
vblank requests.
"
Indeed, i found a Mesa bug yesterday which can cause Mesa's
PresentPixmap request to spuriously feed garbage targetMSC's into the
driver under some conditions. However, while other video drivers seem to
cope relatively well with that, modesetting ddx causes KDE-5's
plasmashell to lock up badly quite frequently, and my suspicion is that
the code removed in this commit is one major source of the extra
fragility.
Also my own tests fail for any swap scheduled more than 1000 vblanks
into the future, which is not uncommon for some scientific applications.
Iow. modesetting's swap scheduling seems to be more robust without this
function afaics.
Signed-off-by: Mario Kleiner <mario.kleiner.de@gmail.com>
Reviewed-by: Adam Jackson <ajax@redhat.com>
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Keith Packard <keithp@keithp.com>
Tested-by: Mike Lothian <mike@fireburn.co.uk>
The autoconf build hard-codes XCONFIGFILE to just 'xorg.conf':
XF86CONFIGFILE="xorg.conf"
AC_DEFINE_DIR(XCONFIGFILE, XF86CONFIGFILE, [Name of configuration file])
Later, the X server passes that into DoSubstitution() which expands the path:
DoSubstitution(template="/etc/X11/%X", ..., XConfigFile="xorg.conf")
This returns "/etc/X11/xorg.conf".
The Meson build, on the other hand, sets XCONFIGFILE to
join_paths(get_option('sysconfdir'), 'xorg.conf'). If sysconfdir is /etc, this
results in '/etc/xorg.conf', resulting in DoSubstitution returning
'/etc/X11/etc/xorg.conf'.
Fix this by just hard-coding XCONFIGFILE to 'xorg.conf'.
Signed-off-by: Aaron Plattner <aplattner@nvidia.com>
Reviewed-by: Thierry Reding <treding@nvidia.com>
'libdir' defaults to 'lib', so running X -showDefaultLibPath just prints 'lib'
instead of '/usr/lib' or '/usr/local/lib'. Use joint_paths() to get the correct
full path.
Signed-off-by: Aaron Plattner <aplattner@nvidia.com>
Reviewed-by: Thierry Reding <treding@nvidia.com>
Copied from Mesa with no modifications.
Gives us Cofeelake platform names updates and sync on Kaby Lake,
Ice Lake PCI IDs.
Acked-by: Kenneth Graunke <kenneth@whitecape.org>
Signed-off-by: Anuj Phogat <anuj.phogat@gmail.com>
Looks quite a bit like the travis path, doesn't it? Still, nice to not
rely on an external service if we don't have to.
Signed-off-by: Adam Jackson <ajax@redhat.com>
Instead of getting the current msc value from the window, which might be
different to old one directly take the last saved msc value saved in
the window_priv struct.
Signed-off-by: Roman Gilg <subdiff@gmail.com>
Reviewed-by: Adam Jackson <ajax@redhat.com>
commit 56c90e29f0 [1.10.99.901]
Author: Adam Jackson <ajax@redhat.com>
Date: Mon Nov 15 14:29:14 2010 -0500
randr: Add RRConstrainCursorHarder
introduced a regression as it ignored the effect of panning and
transforms upon the crtc bounds. The result was that the cursor would be
constrained to the visible area even though the panning arena was much
bigger, or the cursor was constrained to a region that did not even
match the visible area when the output was transformed or reflected.
This supercedes the hack introduced by
commit 1bf81af4a6 [1.12.99.904]
Author: Rui Matos <tiagomatos@gmail.com>
Date: Mon Jul 30 14:32:12 2012 -0400
xf86RandR12: Don't call ConstrainCursorHarder() if panning is enabled
which disabled the cursor constraints if a panning mode was active, but
did not fix the regression with arbitrary output transforms.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=39949
Reviewed-by: Adam Jackson <ajax@redhat.com>
Cc: Rui Matos <tiagomatos@gmail.com>
Clean up only if the request points to the presenting window or its top
parent window.
Since in this case all events are removed unconditionally, always stop
the timer.
Reviewed-by: Adam Jackson <ajax@redhat.com>
Signed-off-by: Roman Gilg <subdiff@gmail.com>
There's no real point - if we don't have EGL then the extension check is
also going to fail - and the entrypoint is new in 1.5.0, which we don't
need to require yet.
Signed-off-by: Adam Jackson <ajax@redhat.com>
Reviewed-by: Mario Kleiner <mario.kleiner.de@gmail.com>
GBM objects were never destroyed after looking for format and
modifier compatibility when deciding whether flipping or copying
a presented pixmap.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=106106
Signed-off-by: Louis-Francis Ratté-Boulianne <lfrb@collabora.com>
If dmabuf_capable is false, because the server "dmabuf_capable"
debug flag isn't set, treat it as successfull query with zero
returned formats, instead of failure.
This allows the servers cache_formats_and_modifiers() function
to cache the fact that formats are not supported during the
current server generation, instead of pointless retesting at
every invocation.
Signed-off-by: Mario Kleiner <mario.kleiner.de@gmail.com>
Reviewed-by: Adam Jackson <ajax@redhat.com>
Checking for dri3_screen_info_rec.version >= 2 is insufficient,
as some shipping drivers, e.g., intel-ddx, nouveau-ddx, set the
version to DRI3_SCREEN_INFO_VERSION, ie. to whatever version the
installed servers headers define. On server 1.20 that would
be version 2, but the drivers still don't support the v1.2
hooks. Make sure all hooks are defined before reporting v1.2.
Also make clamping of reported version to minimum of client
or server robust against possible future clients with possible
majorVersion >= 2.0.
Signed-off-by: Mario Kleiner <mario.kleiner.de@gmail.com>
Reviewed-by: Adam Jackson <ajax@redhat.com>
Cc: Daniel Stone <daniels@collabora.com>
The old info->fd_from_pixmap() driver hook, which is
preferentially used in dri3_fd_from_pixmap(), can return
error codes other than -1, e.g., -EINVAL (-22) on nouveau-ddx.
Not handling the error causes a broken/corrupted X-Connection
resulting from a failed request.
This fixes failure of sddm-greeter to start up under nouveau-ddx
with DRI3 enabled and DRI3 protocol version properly clamped
to 1.0 by the server (see followup patch).
Fixes: 75bba3aedc ("dri3: Use single-FD screen call for single-FD request")
Signed-off-by: Mario Kleiner <mario.kleiner.de@gmail.com>
Reviewed-by: Adam Jackson <ajax@redhat.com>
Cc: Daniel Stone <daniels@collabora.com>
Both xf86-video-intel and xf86-video-nouveau cause OpenGL clients to
fail when used with DRI3 on server 1.20 with Mesa 18.1.
Reason is that the servers DRI3 version is now unconditionally reported
as DRI3 1.2 to 1.2 capable clients. This causes clients using Mesa 18.1
to use the new DRI 3.2 requests DRI3GetSupportedModifiers,
DRI3PixmapFromBuffers, etc. Drivers other than modesetting-ddx do not
support the needed hooks like info->pixmap_from_fds or
info->get_formats, info->get_modifiers. Unfortunately we can't simply
report the servers DRI3 version as 1.0 in this case, as the reported
version can not be specific to a X-Screen, and different screens may
have drivers with different capabilities.
Luckily the server has fallbacks to ->pixmap_from_fd, ->fd_from_pixmap,
and simply reporting an empty set of supported modifiers for the
DRI3GetSupportedModifiers request if the ddx doesn't support DRI 3.2.
Clients like Mesa 18.1's dri3 loader respond to the empty set of
reported modifiers by falling back to a dri driver selected buffer
format (image->createImageWithModifiers responds to a NULL modifier_list
by acting like ->createImage()). This works, but Mesa 18.1 will still
try to use the DRI3PixmapFromBuffers request to create the corresponding
pixmap, just passing in a modifier that corresponds to whatever tiling
the dri driver selected by default. To prevent this request - and
thereby the client - from failing with a BadImplementation error, remove
the check for modifier == DRM_MOD_FORMAT_INVALID in the pixmap_from_fd
fallback path of dri3_pixmap_from_fds() and trust that if we hit the
fallback path then the client will have passed a buffer with some driver
specific default tiling that can be handled by pixmap_from_fd.
Another approach would be for Mesa's dri3 loader to keep track how a
buffer was created (with explicit modifiers or not), and then call
DRI3PixmapFromBuffers or DRI3PixmapFromBuffer, but then any future DRI3
client implementation would need to be fixed, so the server side is
probably the better place for this.
Tested on Intel Ivybridge and NVidia Pascal.
Fixes: 6e7c40f62d ("dri3: Add multi-planar/modifier buffer requests")
Signed-off-by: Mario Kleiner <mario.kleiner.de@gmail.com>
Cc: Daniel Stone <daniels@collabora.com>
Cc: Louis-Francis Ratté-Boulianne <lfrb@collabora.com>
Tested-by: Mike Lothian <mike@fireburn.co.uk>
Reviewed-by: Adam Jackson <ajax@redhat.com>
Only call the get_supported_modifiers vfunc if the DRI3 screen struct is
sufficiently new.
Signed-off-by: Daniel Stone <daniels@collabora.com>
Reviewed-by: Adam Jackson <ajax@redhat.com>
This adds initial support for displaying Xwayland applications through
the use of EGLStreams and nvidia's custom wayland protocol by adding
another egl_backend driver. This also adds some additional egl_backend
hooks that are required to make things work properly.
EGLStreams work a lot differently then the traditional way of handling
buffers with wayland. Unfortunately, there are also a LOT of various
pitfalls baked into it's design that need to be explained.
This has a very large and unfortunate implication: direct rendering is,
for the time being at least, impossible to do through EGLStreams. The
main reason being that the EGLStream spec mandates that we lose the
entire color buffer contents with each eglSwapBuffers(), which goes
against X's requirement of not losing data with pixmaps. no way to use
an allocated EGLSurface as the storage for glamor rendering like we do
with GBM, we have to rely on blitting each pixmap to it's respective
EGLSurface producer each frame. In order to pull this off, we add two
different additional egl_backend hooks that GBM opts out of
implementing:
- egl_backend.allow_commits for holding off displaying any EGLStream
backed pixmaps until the point where it's stream is completely
initialized and ready for use
- egl_backend.post_damage for blitting the content of the EGLStream
surface producer before Xwayland actually damages and commits the
wl_surface to the screen.
The other big pitfall here is that using nvidia's wayland-eglstreams
helper library is also not possible for the most part. All of it's API
for creating and destroying streams rely on being able to perform a
roundtrip in order to bring each stream to completion since the wayland
compositor must perform it's job of connecting a consumer to each
EGLstream. Because Xwayland has to potentially handle both responding to
the wayland compositor and it's own X clients, the situation of the
wayland compositor being one of our X clients must be considered. If we
perform a roundtrip with the Wayland compositor, it's possible that the
wayland compositor might currently be connected to us as an X client and
thus hang while both Xwayland and the wayland compositor await responses
from eachother. To avoid this, we work directly with the wayland
protocol and use wl_display_sync() events along with release() events to
set up and destroy EGLStreams asynchronously alongside handling X
clients.
Additionally, since setting up EGLStreams is not an atomic operation we
have to take into consideration the fact that an EGLStream can
potentially be created in response to a window resize, then immediately
deleted due to another pending window resize in the same X client's
pending reqests before Xwayland hits the part of it's event loop where
we read from the wayland compositor. To make this even more painful, we
also have to take into consideration that since EGLStreams are not
atomic that it's possible we could delete wayland resources for an
EGLStream before the compositor even finishes using them and thus run
into errors. So, we use quite a bit of tracking logic to keep EGLStream
objects alive until we know the compositor isn't using them (even if
this means the stream outlives the pixmap it backed).
While the default backend for glamor remains GBM, this patch exists for
users who have had to deal with the reprecussion of their GPU
manufacturers ignoring the advice of upstream and the standardization of
GBM across most major GPU manufacturers. It is not intended to be a
final solution to the GBM debate, but merely a baindaid so our users
don't have to suffer from the consequences of companies avoiding working
upstream. New drivers are strongly encouraged not to use this as a
backend, and use GBM like everyone else. We even spit this out as an
error from Xwayland when using the eglstream backend.
Signed-off-by: Lyude Paul <lyude@redhat.com>
Acked-by: Daniel Stone <daniels@collabora.com>
Reviewed-by: Adam Jackson <ajax@redhat.com>
Just a small autogenerated header that will soon contain more then just
one macro.
Signed-off-by: Lyude Paul <lyude@redhat.com>
Reviewed-by: Daniel Stone <daniels@collabora.com>
Reviewed-by: Adam Jackson <ajax@redhat.com>
This takes all of the gbm related code in wayland-glamor.c and moves it
into it's own EGL backend for Xwayland, xwayland-glamor-gbm.c.
Additionally, we add the egl_backend struct into xwl_screen in order to
provide hooks for alternative EGL backends such as nvidia's EGLStreams.
Signed-off-by: Lyude Paul <lyude@redhat.com>
Reviewed-by: Lyude Paul <lyude@redhat.com>
Reviewed-by: Daniel Stone <daniels@collabora.com>
Reviewed-by: Adam Jackson <ajax@redhat.com>
32 is not a valid depth, and the default is now 24 not 8.
Signed-off-by: Adam Jackson <ajax@redhat.com>
Reviewed-by: Alan Coopersmith <alan.coopersmith@oracle.com>
fb is happy to do TrueColor to 8bpp drawables, but mesa is not. Depth 12
is the biggest pseudocolor anyone ever really did, and 15 is the least
truecolor.
Without this Xvfb at depth 8 would "have" GLX, but no vendors
would actually back any of the screens. libGL will attempt to call
GLXQueryServerString to figure out the GLX version, and vnd will throw
an error because there's no vendor to dispatch that to, and then clients
crash.
Signed-off-by: Adam Jackson <ajax@redhat.com>
GLX registers an extension before we know if there are any screens that
can actually do it. It's inconvenient to shrink the extension list, so
instead allow the extension to simply zero out its base opcode to
indicate that it needed to panic and disable itself.
Signed-off-by: Adam Jackson <ajax@redhat.com>
