This is a modified version of a patch we've been carry-ing in Fedora and
RHEL for years now. This patch automatically adds secondary GPUs to the
master as output sink / offload source making e.g. the use of
slave-outputs just work, with requiring the user to manually run
"xrandr --setprovideroutputsource" before he can hookup an external
monitor to his hybrid graphics laptop.
There is one problem with this patch, which is why it was not upstreamed
before. What to do when a secondary GPU gets detected really is a policy
decission (e.g. one may want to autobind PCI GPUs but not USB ones) and
as such should be under control of the Desktop Environment.
Unconditionally adding autobinding support to the xserver will result
in races between the DE dealing with the hotplug of a secondary GPU
and the server itself dealing with it.
However we've waited for years for any Desktop Environments to actually
start doing some sort of autoconfiguration of secondary GPUs and there
is still not a single DE dealing with this, so I believe that it is
time to upstream this now.
To avoid potential future problems if any DEs get support for doing
secondary GPU configuration themselves, the new autobind functionality
is made optional. Since no DEs currently support doing this themselves it
is enabled by default. When DEs grow support for doing this themselves
they can disable the servers autobinding through the servers cmdline or a
xorg.conf snippet.
Signed-off-by: Dave Airlie <airlied@gmail.com>
[hdegoede@redhat.com: Make configurable, fix with nvidia, submit upstream]
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
---
Changes in v2:
-Make the default enabled instead of installing a xorg.conf
snippet which enables it unconditionally
Changes in v3:
-Handle GPUScreen autoconfig in randr/rrprovider.c, looking at
rrScrPriv->provider, rather then in hw/xfree86/modes/xf86Crtc.c
looking at xf86CrtcConfig->provider. This fixes the autoconfig not
working with the nvidia binary driver
The miPointerSpriteFunc swcursor code expects there to only be a single
framebuffer and when the cursor moves it will undo the damage of the
previous draw, potentially overwriting what ever is there in a new
framebuffer installed after a flip.
This leads to all kind of artifacts, so we need to disable pageflipping
when a swcursor is used.
The code for this has shamelessly been copied from the xf86-video-amdgpu
code.
Fixes: https://gitlab.freedesktop.org/xorg/xserver/issues/828
Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Fix the following compiler warning:
drmmode_display.c: In function ‘drmmode_create_bo’:
drmmode_display.c:1019:9: warning: ISO C90 forbids mixed declarations and code [
1019 | uint32_t num_modifiers;
| ^~~~~~~~
Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
When the pixmapPrivateKeyRec was moved from a global to being embedded
inside the drmmode_rec these 2 where missed, clean them up.
Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
The modesetting driver (which now often is used with Intel GPUs),
relies on DRI2ScreenInit() to setup the DRI and VDPAU driver names.
Before this commit it would always assign the same name to the 2 names,
but the VDPAU driver for i965 GPUs should be va_gl.
This commit adds a special case for the i965 case, replacing the
VDPAU driver name with "va_gl" if the GPU is using the i965 driver
for DRI.
Note this commit adds a FIXME comment for a related memory leak, that leak
was already present and fixing it falls outside of the scope of this commit.
BugLink: https://bugzilla.redhat.com/show_bug.cgi?id=1413733
Cc: kwizart@gmail.com
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
This script generates a header that has a comment containing the build path for
no real reason. As this source can end up deployed on targets in debug packages
this means there is both potentially sensitive information leakage about the
build environment, and a source of change for reproducible builds.
Stop trying to link to a shared library we no longer build
Fixes: commit c1703cdf3b - "xfree86: Link fb statically"
Signed-off-by: Alan Coopersmith <alan.coopersmith@oracle.com>
"bool" conflicts with C++ (meh) and stdbool.h (ngh alright fine). This
is a driver-visible change and will likely break the build for mach64,
but it can be fixed by simply using xf86ReturnOptValBool like every
other driver.
Signed-off-by: Adam Jackson <ajax@redhat.com>
<sys/io.h> on ARM hasn't worked for a long, long time, so it was removed
it from glibc upstream.
Remove the include to avoid a compilation failure on ARM with glibc.
Signed-off-by: Olivier Fourdan <ofourdan@redhat.com>
Closes: https://gitlab.freedesktop.org/xorg/xserver/issues/840
Don't link against fb, it's the driver's responsibility to load that
first. Underlinking like this is unpleasant but this matches what
autotools does.
Fixes: xorg/xserver#540
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>
Promote the generated file containing the date & time build was
configured to top-level.
Rename it from xf86Build.h to buildDateTIme.h.
Use it as well in XQuartz, stringize BUILD_DATE when needed.
If SYSTEMD_LOGIND is not defined, systemd_logind_take_fd is defined as a
macro evaluating to -1 by systemd-logind.h, leaving paused
uninitialized.
../hw/xfree86/common/xf86Xinput.c: In function ‘xf86NewInputDevice’:
../hw/xfree86/common/xf86Xinput.c:919:16: warning: ‘paused’ may be used uninitialized in this function [-Wmaybe-uninitialized]
../hw/xfree86/common/xf86Xinput.c:877:10: note: ‘paused’ was declared here
../hw/xfree86/os-support/stub/stub_init.c: In function ‘xf86OSInputThreadInit’:
../hw/xfree86/os-support/stub/stub_init.c:29:1: warning: old-style function definition [-Wold-style-definition]
Drivers may need to loop over the allocated screens during PreInit, for example
to consolidate xorg.conf options that apply to a GPU device as a whole.
Currently, this works for protocol screens becuase x86Screens is exported, but
does not work for GPU screens.
Export xf86GPUScreens and xf86NumGPUScreens for consistency with xf86Screens and
xf86NumScreens.
Signed-off-by: Aaron Plattner <aplattner@nvidia.com>
If the user sets Option "Enable" "TRUE" for a monitor, the X
server will connect the connector a crtc but tell the user it
is disconnected.
However the user in this case is mutter, when it gets it's view
of the output configuration it sees the output is disconnected
and never sets it up again, which seems like the right thing to do.
If we let the user enable a monitor, lets just set it as always
connected.
Reviewed-by: Olivier Fourdan <ofourdan@redhat.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
systemd-logind since version 234 (released 2017-07-12) supports being
restarted without losing state [1]. From the systemd NEWS file [2]:
* systemd-logind may now be restarted without losing state. It stores
the file descriptors for devices it manages in the system manager
using the FDSTORE= mechanism. Please note that further changes in
other components may be required to make use of this (for example
Xorg has code to listen for stops of systemd-logind and terminate
itself when logind is stopped or restarted, in order to avoid using
stale file descriptors for graphical devices, which is now
counterproductive and must be reverted in order for restarts of
systemd-logind to be safe. See
https://cgit.freedesktop.org/xorg/xserver/commit/?id=dc48bd653c7e101.)
This reverts commit dc48bd653c.
Closes: #531
[1] https://github.com/systemd/systemd/pull/5600
[2] 9f09a95a7e
Normally, the X server infers the initial screen size based on any
connected outputs. However, if no outputs are connected, the X server
picks a default screen size of 1024 x 768. This option overrides the
default screen size to use when no outputs are connected. In contrast
to the "Virtual" Display SubSection entry, which applies unconditionally,
"NoOutputInitialSize" is only used if no outputs are detected when the
X server starts.
Parse this option in the new exported helper function
xf86AssignNoOutputInitialSize(), so that other XFree86 loadable drivers
can use it, even if they don't use xf86InitialConfiguration().
Signed-off-by: Andy Ritger <aritger@nvidia.com>
Reviewed-by: Keith Packard <keithp@keithp.com>
Since the Solaris kernel tracks IOPL per thread, and doesn't inherit
raised IOPL levels when creating a new thread, we need to turn it on
in the input thread for input drivers like vmmouse that need register
access to work correctly.
Signed-off-by: Alan Coopersmith <alan.coopersmith@oracle.com>
Keeping track of kernel state in user space doesn't buy us anything,
and introduces bugs, as we were keeping global state but the Solaris
kernel tracks IOPL per thread.
Signed-off-by: Alan Coopersmith <alan.coopersmith@oracle.com>
The VGA arbiter controls the PCI bus' routing of legacy VGA resources,
specifically the video memory aperture at 0xa0000-0xb0000 (640k should
be etc.) and a handful of I/O ports. Since 128k is far too small for a
real framebuffer these days, every driver instead maps a linear version
of VRAM through the PCI BAR. And no DRI2 drivers ever need I/O port
access, because all operations they might be used for (legacy VGA CRTC
setup, mostly) happen on the kernel side.
In other words, this just works, and we can stop breaking it.
Signed-off-by: Adam Jackson <ajax@redhat.com>
Reviewed-by: Dave Airlie <airlied@redhat.com>
A user of Adélie Linux reported that modesetting wasn't working properly on
their Intel i7-9700K-integrated UHD 630 GPU. Xorg.0.log showed:
[ 131.902] (EE) modeset(0): [DRI2] No driver mapping found for PCI device 0x8086 / 0x3e98
[ 131.902] (EE) modeset(0): Failed to initialize the DRI2 extension.
Indeed, that PCI ID is missing from i965_pci_ids. Adding it fixed the issue
and allowed the system to work with i965_dri under modesetting.
All of the null checks here are redundant, you can't get to those paths
unless RANDR's already been initialized. Delete them, and remove the
pointer too.
Signed-off-by: Adam Jackson <ajax@redhat.com>
Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>
If the driver calls xf86HandleColormaps, CMapChangeGamma updates the HW
gamma LUT of all CRTCs via xf86RandR12LoadPalette. However,
xf86RandR12ChangeGamma was then clobbering the gamma LUT of the RandR
1.2 compatibility output's CRTC with the gamma curves computed from the
screen's global gamma values.
Fix this by bailing if xf86RandR12LoadPalette is installed.
Fixes: 02ff0a5d7e "xf86RandR12: Fix XF86VidModeSetGamma triggering a
BadImplementation error"
Noticed when porting this logic to xf86-video-nouveau, and valgrind
complained about conditional jump based on uninitialized data.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Pekka Paalanen <pekka.paalanen@collabora.com>
Gitlab very kindly exposes the details of the git commit message (among
much else) in the environment. Unfortunately, piglit tries to handle the
environment in non-UTF8-safe ways, which means if the top-of-tree commit
mentions non-ASCII characters (say, in the author's name) then all the
tests fail and so does the pipeline.
Fortunately none of those variables are things our piglit invocation
needs. Since I've failed to rebuild the docker image as yet, just clear
the likely variables from the environment before running piglit.
This-makes-me: ☹
Believe it or not, somehow we've never done this in legacy mode! We
currently simply change the DPMS property on the CRTC's output's
respective DRM connector, but this means that we're just setting the
CRTC as inactive-not disabled. From the perspective of the kernel, this
means that any shared resources used by the CRTC are still in use.
This can cause problems for drivers that are not yet fully atomic,
despite using the atomic helpers internally. For instance: if CRTC-1 and
CRTC-2 are still enabled and use shared resources within the kernel (an
MST topology, for example), and then userspace tries to go enable CRTC-3
on the same topology this might suddenly fail if CRTC-3 needs the shared
resources CRTC-1 and CRTC-2 are using. While I don't know of any
situations in the mainline kernel that actually trigger this, future
plans for reworking the atomic check of MST drivers are absolutely
going to make this into a real issue (they already are in my WIP
branches for the kernel).
So: actually do the right thing here and disable CRTCs when they're not
going to be used anymore, even in legacy mode.
Signed-off-by: Lyude Paul <lyude@redhat.com>
Some Broadcom set-top-box boards have PCI busses, but the GPU is still
probed through DT. We would dereference a null busid here in that
case.
Signed-off-by: Eric Anholt <eric@anholt.net>
Lifted from vfb. xfree86 had almost the same thing but unparameterized,
port it to the vfb style.
Signed-off-by: Adam Jackson <ajax@redhat.com>
Reviewed-by: Alan Coopersmith <alan.coopersmith@oracle.com>
Could cause privilege elevation and/or arbitrary files overwrite, when
the X server is running with elevated privileges (ie when Xorg is
installed with the setuid bit set and started by a non-root user).
CVE-2018-14665
Issue reported by Narendra Shinde and Red Hat.
Signed-off-by: Matthieu Herrb <matthieu@herrb.eu>
Reviewed-by: Alan Coopersmith <alan.coopersmith@oracle.com>
Reviewed-by: Peter Hutterer <peter.hutterer@who-t.net>
Reviewed-by: Adam Jackson <ajax@redhat.com>
At the point where xf86BusProbe runs we haven't yet taken our own VT,
which means we can't perform drm "master" operations on the device. This
is tragic, because we need master to fish the bus id string out of the
kernel, which we can only do after drmSetInterfaceVersion, which for
some reason stores that string on the device not the file handle and
thus needs master access.
Fortunately we know the format of the busid string, and it happens to
almost be the same as the ID_PATH variable from udev. Use that instead
and stop calling drmSetInterfaceVersion.
Reviewed-by: Peter Hutterer <peter.hutterer@who-t.net>
Signed-off-by: Adam Jackson <ajax@redhat.com>
If the X server is terminated while its VT is not active, it should
not change the current VT.
v2: Query current state in xf86CloseConsole using VT_GETSTATE instead of
keeping track in xf86VTEnter/xf86VTLeave/etc.
This hasn't done anything besides return TRUE in a long long time.
Reviewed-by: Eric Anholt <eric@anholt.net>
Signed-off-by: Adam Jackson <ajax@redhat.com>
These are so close to identical that most DDXes implement one in terms
of the other. All the relevant cases can be distinguished by the error
code, so merge the functions together to make things simpler.
Reviewed-by: Eric Anholt <eric@anholt.net>
Signed-off-by: Adam Jackson <ajax@redhat.com>