As we may hold a cached reference to an upload buffer whilst it is
in-flight, the assertion that there are no such references to a buffer
being reused is no longer true. Those cached references will be released
as soon as we retire the buffer during the readback and so we are free
to reuse such an upload buffer for immediate readback.
Reported-by: Jiri Slaby <jirislaby@gmail.com>
References: https://bugs.freedesktop.org/show_bug.cgi?id=47597
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Same change for intel_display.c and sna_display.c.
Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
If we lack the ability to use a shader to compute the gradients
per-pixel, we need to use pixman to render a fallback texture. We can
reduce the size of this texture and upsample to reduce the cost with
hopefully imperceptible loss of quality.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
It was broken and not flushing damage correctly. With the
improvements made to the kernel, it is no longer a significant advantage
per se and not worth its additional complexity.
Reported-by: Tilman Sauerbeck <tilman@code-monkey.de>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=32547
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Another case where I passed an empty string believing that would be
sufficient to replace the error path...
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
With xterm, it is quite common for it to redraw itself by using lots of
spaces and so it is efficient for us if we ellide those clear glyphs and
only draw the backing boxes. However, we were only checking the first 8
pixels in each line because of a missing pointer increment.
Fixes absent '=' characters when using a compositor and ImageText.
Reported-by: Jiri Slaby <jirislaby@gmail.com
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=47735
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Since the removal of the ability to create a backing pixmap after the
creation of its parent, it no longer becomes practical to attempt
rendering with the GPU to unattached pixmaps. So having made the
decision never to render to that pixmap, perform the test explicitly
along the render paths.
This fixes a segmentation fault introduced in 8a303f195 (sna: Remove
existing damage before overwriting with a composite op) which assumed
the existence of a backing pixmap along a render path.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=47700
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Double check that the maximum access size computed from the bo
parameters is within the allocated size for the bo.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
The danger now is that we may have either discarded the shadow pixels or
have replaced them with a GTT mapping, either way undesirable and so we
should reconstruct the shadow mapping.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
As the font is kept around and reused after UnrealizeFont, we need to
nullify the pointer to our private data in order to prevent the later
use-after-free.
Reported-by: Peter Jordan
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
As we only enter that path for singular unbounded boxes, we are
guaranteed to fill the entire trapezoid extents and so do not need the
unbounded fixup the assertion was fretting about.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
We need to do this apparently or else we never perform the VT switch.
However, we can not do it too early, especially not before we have
finished intialising the outputs.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=47395
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
In some cases off-screen is stil visible, for example under a rotation.
As such xrandr -o left; glxgears -fullscreen was broken.
Reported-by: Phillip Haddad <phillip.haddad@gmail.com>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Mark vertical edges with dy==0 to reduce structure size and reduce
memory load during edge walking.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
As the extraction routine is now smarter and can construction
subsurfaces without copying we do not need to force tiling.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
The compiler presumes that the uint64_t write is naturally aligned and
so may emit code that crashes with an unaligned moved. To workaround
this, make sure the write is so aligned.
References: https://bugs.freedesktop.org/show_bug.cgi?id=47418
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Confusion reigns between using the backing pixmap for the drawable for
the front buffer, and a fake pixmap for the auxiliary buffers.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
So that we do not keep a stale references of the last cached pixmap
across server regeneration (or shutdown).
Reported-by: Thierry Reding <thierry.reding@avionic-design.de>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=47357
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
If we choose not to use the CPU bo for the upload (because we fear the
subsequent synchronisation cost), we need to fixup the shadow pointer
before dereferencing it.
On the move-to-cpu side, the fixup is already performed as we will need
to access the shadow pixels for the subsequent drawing operation.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Chris Wilson
Paulo Zanoni