Adam Jackson notes that what appeared to be my paranoid ramblings in SNA
actually served a purpose - it prevents a server crash following
server regen if an indirect client happened to be running at the time
(e.g. LIBGL_INDIRECT_ALWAYS=1 glxgears).
Reported-by: Adam Jackson <ajax@redhat.com>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Since RandR itself is disabled if Xinerama is enabled, for example with
ZaphodHeads, calling RRGetInfo() upon a hotplug event generates an
assertion.
Reported-by: Stephen Liang <inteldriver@angrywalls.com>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=55260
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Damage is processed in two phases, with the actual Damage being appended
before the operation is performed so that a copy can be made before
modification (e.g. software cursors).
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
On gen6+, bo are expected to be LLC by default. However, as using the bo
for the scanout causes it to be moved into the uncached domain, this
assumption is then false and we should release the bo back to the system
rather than spread the uncached buffers around. The most common
allocator of scanouts is for pageflipping which are already non-reusable
due to the DRI2 export, so there should actually be little impact.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
When we return NULL from sna_drawable_use_bo(), the expectation is that
the damage pointer is also NULL. However, one SHM path leaked.
References: https://bugs.freedesktop.org/show_bug.cgi?id=56180
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Fixes regression from commit c51aaa731e
Author: Chris Wilson <chris@chris-wilson.co.uk>
Date: Thu Oct 11 11:36:00 2012 +0100
sna/gen7: Replace bogus state tracking assertion
The assumption being that we only used the encoded flags for determining
the composite state is false for the magic CA pass.
Reported-by: Oleksij Rempel <bug-track@fisher-privat.net>
Reported-by: Eyal Lotem <eyal.lotem@gmail.com>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=56037
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
The only path where this is correct already handles it as the special
case that it is, everywhere else it just nonsense.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
As the maximum reported image sizes are for the source image, we should
be careful not to recommend the application use an output Window larger
than can be handled by the overlay hardware. So shrink it to fit, whilst
preserving the aspect ratio.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Libdrm's possible_clones is a mask of encoders. Xorg's possible_clones
is a mask of outputs, so we just can't do the following:
output->possible_clones = kencoder->possible_clones;
This is a problem on Haswell because, at least with the current
patches floating on the mailing list, there is more than one connector
per encoder.
This patch writes the code to properly translate libdrm's encoder mask
into Xorg's output mask.
Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
If we are forced to use the GPU bo as the target because the CPU bo is
busy, the priv->cpu flag is likely to remain set, so we need to clear it
rather than fail the assertion that is false.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Since pre-g33 chipsets impose massive alignment restrictions on objects
within the aperture we need to further restrict the amount of available
space to be sure we have sufficient room to accommodate the alignment.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
This helps with fitting larger operations into the small apertures of
gen2, which due to the lax accounting could trigger ENOSPC.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
While playing with MPX and sw cursor I noticed page flips won't
end up misrendering some bits, so the sw cursor was replacing the
bits on the wrong pixmap.
Fix the damage handling to be correct and append damage before swapping
the pointers and process damage after.
This fixes misrendering with MPX cursors under a fullscreen compositor,
that pageflips.
Signed-off-by: Dave Airlie <airlied@redhat.com>
[ickle: The secret is that damage is sometimes reported before the
rendering in DamageRegionAppend, and sometimes afterwards in
DamageRegionProcessPending. For instance, the software cursor operates
prior to being rendered over.]
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
In case we need to redirect the rendering for a large render target, we
need to initialize the damage pointer.
Reported-by: Clemens Eisserer <linuxhippy@gmail.com>
Bugzilla: ttps://bugs.freedesktop.org/show_bug.cgi?id=55812
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reported-by: Clemens Eisserer <linuxhippy@gmail.com>
Bugzilla: ttps://bugs.freedesktop.org/show_bug.cgi?id=55812
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
When is a pipelined operation, not pipelined? That is the mystery posed
by our hardware!
Reported-by: Clemens Eisserer <linuxhippy@gmail.com>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=51422
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
As we use pkg-config to determine whether to use intel-gen4asm, we
should also use it to locate the right version of intel-gen4asm to use.
This allows the user to install the assembler in a non-standard path for
cross-builds and similar.
Reported-by: Josh Tripplet <josh@freedesktop.org>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=55646
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Chris Wilson
Paulo Zanoni
Dave Airlie