Pick smaller types where possible, including bitfielding some Bools and
small enums, then shuffle the result to be hole-free. 192 -> 128 bytes
on LP64, 144 -> 96 bytes on ILP32.
Signed-off-by: Adam Jackson <ajax@redhat.com>
This is a really awkward interface, since we're calling it well before
the driver knows what device it's going to drive. Drivers with both KMS
and UMS support therefore don't know whether to say they need I/O port
access or not, and have to assume they do.
With this change we now call it only to query whether port access might
be needed; we don't use that to determine whether to call a driver's
probe function or not, instead we call them unconditionally. If the
driver doesn't check whether port access was enabled, they might crash
ungracefully. To accomodate this, we move xorgHWAccess to be explicitly
intentionally exported (sigh xf86Priv.h) so that drivers can check that
before they attempt port access.
v2: Move initial xf86EnableIO() nearer the logic that determines whether
to call it, suggested by Simon Farnsworth.
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Reviewed-by: Simon Farnsworth <simon.farnsworth@onelan.co.uk>
Signed-off-by: Adam Jackson <ajax@redhat.com>
We load the driver list, then enable I/O, then call driver probe based
on whether I/O enable succeeded. That's bad, because the loaded
security policy might forbid port access. We happen to treat that as
fatal for some reason, which means even drivers that don't need I/O
access (like kms and fbdev) don't get the chance to run. Facepalm.
How about we just make that non-fatal instead, that sounds like a much
better plan.
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Reviewed-by: Simon Farnsworth <simon.farnsworth@onelan.co.uk>
Signed-off-by: Adam Jackson <ajax@redhat.com>
Pull platform methods into their own sections for legibility, and
rewrite the ifdefs to be more concise.
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Reviewed-by: Simon Farnsworth <simon.farnsworth@onelan.co.uk>
Signed-off-by: Adam Jackson <ajax@redhat.com>
If we are not seat 0 the following apply:
don't probe any bus other than platform
don't probe any drivers other than platform
assume the first platform device we match on the bus is the primary GPU.
This just adds checks in the correct places to ensure this, and
with this X can now start on a secondary seat for an output device.
v2: fix Seat0 macros
Reviewed-by: Keith Packard <keithp@keithp.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
for input devices we handle change like remove/add, but for
drm devices we get change events when we hotplug outputs,
so lets just ignore change at this level, and let the drivers
handle it. We may in the future want to route driver udev
from here instead.
Reported-by: Timo Aaltonen <timo.aaltonen@canonical.com>
Reviewed-by: Keith Packard <keithp@keithp.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
This solves a race if we are trying to dynamically power off
secondary GPUs. Its not the greatest fix ever but it probably
as good as we can do for now.
The GPU probing causes the devices to be powered up, then when
we scan the PCI bus we get the correct information from the kernel,
rather than a bunch of 0xff due to the device being powered off.
drop gratuitous '&'.
Reviewed-by: Keith Packard <keithp@keithp.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
After we share the pixmap, the backing storage may have changed,
and we need to invalidate and buffers pointing at it.
This fixes GL compositors and prime windows lacking contents initially.
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Dave Airlie <airlied@redhat.com>
Otherwise we can't do fast user switch properly for multiple GPUs.
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
This respects the seat tag for hotplugged video devices at X start.
Reviewed-by: Keith Packard <keithp@keithp.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
this is a simple clean-up that is useful to stop further propogation
of this construct.
Reviewed-by: Keith Packard <keithp@keithp.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
If we don't free this here, it gets freed later in the resource
cleanups, however it then looks up up pmap->pScreen, which we
freed already in this function. So free the default colormap
when we should.
This fixes a bug after a couple of hotplug cycles when you try
to exit the X server and it crashes.
Reviewed-by: Keith Packard <keithp@keithp.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
This is set by pre_init not screen init, so if we free it here
and then recycle the server, we lose all the providers.
I think we need to wrap FreeScreen here to do this properly,
will investigate for 1.14 most likely, safer to just leak this
on server exit for now.
Reviewed-by: Keith Packard <keithp@keithp.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
Some compilers have difficulty with the previous implementation which
relies on undefined behavior according to the C standard. Using
offsetof() from <stddef.h> (which most likely just uses
__builtin_offsetof on modern compilers) allows us to accomplish this
without ambiguity.
This fix also requires support for typeof(). If your compiler does not
support typeof(), then the old implementation will be used. If you see
failures in test/list, please try a more modern compiler.
v2: Added fallback if typeof() is not present.
Signed-off-by: Jeremy Huddleston Sequoia <jeremyhu@apple.com>
Reviewed-by: Peter Hutterer <peter.hutterer@who-t.net>
This is a a gcc 4.6+ feature.
signal-logging.c:210: error: #pragma GCC diagnostic not allowed inside
functions
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Reviewed-by: Dave Airlie <airlied@redhat.com>
Signed-off-by: Keith Packard <keithp@keithp.com>
We never use child[2], so it's state is undefined.
This issue seems to have existed since the test was first
written: 92788e677b
Signed-off-by: Jeremy Huddleston Sequoia <jeremyhu@apple.com>
Reviewed-by: Peter Hutterer <peter.hutterer@who-t.net>
sighandler_t is not UNIX.
Regression from: 7f09126e06
Signed-off-by: Jeremy Huddleston Sequoia <jeremyhu@apple.com>
Reviewed-by: Peter Hutterer <peter.hutterer@who-t.net>
Introduced in 164b38c72f
Reported-by: Jon TURNEY <jon.turney@dronecode.org.uk>
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Reviewed-by: Keith Packard <keithp@keithp.com>
As of 81cfe44b1e, miPointerSetPosition now
returns the screen pointer of the device. This broke floating slave devices,
as soon as a motion event was submitted, miPointerSetPosition returned NULL,
crashing the server.
dev->coreEvents is only false if the device is a floating slave, in which
case it has a sprite.
X.Org Bug 53568 <http://bugs.freedesktop.org/show_bug.cgi?id=53568>
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Reviewed-by: Keith Packard <keithp@keithp.com>
In WarpPointer calls, we get input in screen coordinates. They must be
scaled to device coordinates, and then back to screen coordinates for screen
crossing and root coordinates in events.
The rounding errors introduced (and clipping in core/XI 1.x events) can lead
to the actual position being different to the requested input coordinates.
e.g. 200 scales to 199.9999, truncated to 199 in the event.
Avoid this by simply overwriting the scaled screen coordinates with the
input coordinates for the POINTER_SCREEN case.
X.Org Bug 53037 <http://bugs.freedesktop.org/show_bug.cgi?id=53037>
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Reviewed-by: Keith Packard <keithp@keithp.com>
Throw an error into the log file, but continue anyway. And after three
warnings, stop complaining. Not all input drivers will be fixed in time (or
ever) and our printf implementation is vastly inferior, so there is still a
use-case for non-sigsafe logging.
This also adds more linebreaks to the message.
CC: Chase Douglas <chase.douglas@canonical.com>
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Reviewed-by: Chase Douglas <chase.douglas@canonical.com>
The mouse driver uses %i in some debug messages
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Reviewed-by: Chase Douglas <chase.douglas@canonical.com>
Throw an assert when the conversion fails instead of just returning. Asserts
are more informative.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Reviewed-by: Chase Douglas <chase.douglas@canonical.com>
Commit 9d457f9c55 added an array of
DevPrivateSetRec structures in the middle of the ScreenRec, which throws off
extension modules trying to call things like pScreen->DestroyPixmap.
Signed-off-by: Aaron Plattner <aplattner@nvidia.com>
Signed-off-by: Keith Packard <keithp@keithp.com>
This change is #if'd out due to a bug in asl_log_descriptor, but
it is left here as reference to be enabled in the future.
Signed-off-by: Jeremy Huddleston Sequoia <jeremyhu@apple.com>
Prior to this change, it was possible that a large message would have some
of its data prepended to subsequent messages due to our not incorrectly
setting the location to write into the buffer.
Signed-off-by: Jeremy Huddleston Sequoia <jeremyhu@apple.com>
In __glXScreenInit() we generate the set of GLX visuals in two steps:
first we match each pre-existing X visual with a corresponding
FBConfig, then we generate a new X visual to correspond to all the
remaining FBConfigs.
The first step is used for the two default 24-bit visuals (true color
and direct color) and for the 32-bit visual. If windowsystem
multisampling is enabled in Mesa, we need to ensure that none of these
three visuals gets matched to a multisampled config.
Fixes a bug with windowsystem multisampling in gnome-shell. If the X
server happens to match up a multisampled FBConfig to the 32-bit
visual, gnome-shell will try to use it to read pixels from
alpha-blended windows (such as gnome-terminal), resulting in no window
appearing on screen.
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Chad Versace <chad.versace@linux.intel.com>
Signed-off-by: Keith Packard <keithp@keithp.com>
quartz.c:153:6: error: use of undeclared identifier 'GlxExtensionInit'; did you mean 'GEExtensionInit'? [Semantic Issue]
{GlxExtensionInit, "GLX", &noGlxExtension},
Regression-from: aad428b8e2
Signed-off-by: Jeremy Huddleston <jeremyhu@apple.com>
The core RandR screen cleanup now involves cleaning up any GPU screen
associations, and those call down into DDX to clean up the driver. If
the pointers from the xf86 structures back to the core randr
structures are set to NULL at that point, bad things happen.
This patch "knows" that the core RandR close screen is underneath the
xf86 randr close screen function, and so makes sure it gets called
first.
Signed-off-by: Keith Packard <keithp@keithp.com>
Reviewed-by: Dave Airlie <airlied@redhat.com>
This should make cleaning up the GPU screens easier as the core
screens they are associated with will still be around.
Signed-off-by: Keith Packard <keithp@keithp.com>
Reviewed-by: Dave Airlie <airlied@redhat.com>
Hotplugging screens causes the render filter names to get freed while
still in use; wait for the last core screen to be closed before
freeing them. That only happens at server reset, when we want them to
be freed.
Signed-off-by: Keith Packard <keithp@keithp.com>
Reviewed-by: Dave Airlie <airlied@redhat.com>
When fbBresSolid draws a line, it can happen that after the last
pixel, the Bresenham error term overflows, and fbBresSolid paints
another pixel before adjusting the error term.
However, if this happens on the last pixel (len=0), this extra pixel
might overshoot the boundary, and, in rare cases, lead to a segfault.
Fix this issue by adjusting for the Bresenham error term before
drawing the main pixel, not after.
Fixes: https://bugs.freedesktop.org/show_bug.cgi?id=24274
Signed-off-by: Simon Schubert <2@0x2c.or>
Tested-by: Mitch Davis <mjd+freedesktop.org@afork.com>
Signed-off-by: Matt Turner <mattst88@gmail.com>
Signed-off-by: Keith Packard <keithp@keithp.com>