During the present unflip path we blindly try to restore the original
mode after a flip failure. However, it confuses flipping zero CRTC with
a genuine failure. This has the result of undoing a DPMS change (e.g.
xset dpms force dpms) under a DRI3 compositor.
Reported-by: Jiri Slaby <jirislaby@gmail.com>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=81456
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
If the kernel reports that it is busy, it has not yet finished
processing a pending flip and we have multiple CRTC queued, just wait
for the kernel to clear its backlog before submitting the next flip. On
the other hand, if we can just overwrite the pending flip results.
However, the EBUSY may actually be a genuine report by the kernel of an
error, so check for an invalid CRTC first.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Since we only have the single callsite for do_page_flip, the wrapper is
not adding any meaningful information.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
We assert that damage is valid as we delete the Pixmap, and so we cannot
assert that the Pixmap is still valid itself.
Reported-by: Jiri Slaby <jirislaby@gmail.com>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
I overlooked the tiling-fill path passing down a temporary DrawableRec
into the fill_boxes callback - invalidating the assertion.
Fixes regression from
commit 43176b9bfa
Author: Chris Wilson <chris@chris-wilson.co.uk>
Date: Mon Jun 30 21:01:11 2014 +0100
sna/dri2: Pass around the correct DrawableRec for sampling from the foriegn bo
Reported-by: Jiri Slaby <jirislaby@gmail.com>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=70461#c62
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
In order to support a wider range of rotation/reflections, perform a
simple reduction of the requested rotation first.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
uClibc is one such library that doesn't implement getline()
Reported-by: Ben Widawsky <benjamin.widawsky@intel.com>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
The kernel interface has changed slightly since the early proposals. Now
the rotation property is only on the plane and so we have to lookup the
primary prime as well.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
If we already have a buffer that represents the data on the GPU, we can
simply use that when we need to promote the pixmap onto the GPU.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Note sure if this is strictly required -- but at the moment it fails for
1x1R solids, causing us to skip glyphs. The simulator doesn't complain,
so just skip the check for now.
Reported-by: Ben Widawsky <ben@bwidawsk.net>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
After relaxing some of the rules on when to discard the upload proxies,
we also need to relax some of the complementary asserts.
Reported-by: Zdenek Kabelac <zkabelac@redhat.com>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
We need to wrap xf86DPMSSet() so that we can reintialize our bookkeeping
and auxiliary planes after disabling/re-enabling CRTC during DPMS
operations.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
When we tile, we do so in order to fit an operation involving two
objects larger than the aperture. If we then choose an intermediate
tiling object that is larger than either of those two, the error will
persist and we will be forced to recuse.
In the worst case, this will provide an upper bound to the recursion.
Reported-by: Bruno Prémont <bonbons@linux-vserver.org>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Eeek, when not using LinearFramebuffer we still want to create the GPU
bo: s/,/;/
Reported-by: Ilia Mirkin <imirkin@alum.mit.edu>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
This option should only be used for compatibility. Previously this was
done at a high level, this changes it to enforce the tiling as we apply
the CRTC.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
We need to initialise both depth and bitsPerPixel on the drawable struct
we pass around as they are used for selecting for the format when
copying.
Reported-by: Vedran Rodic <vrodic@gmail.com>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
I assume the intention was to provide a different structure for each of
the gen 2 devices.
This doesn't change anything really.
Signed-off-by: Damien Lespiau <damien.lespiau@intel.com>
One day, we will move the width/height/bpp to the bo itself...
Reported-by: Ilia Mirkin <imirkin@alum.mit.edu>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
We need to only clip to the extents of the copy in the buffer space -
which implies that we need to translate the region into that space before
doing the clip.
Reported-by: Ilia Mirkin <imirkin@alum.mit.edu>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
If we are not actually accessing the memory through the pointer, we do
not care if it not currently coherent.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Since we now prefer CPU detiling, exactly when we want active/inactive
buffers is a little more complex - and we also need to take into account
when we want to use the CPU bo as a render target.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Fixes regression from
commit 961139f587 [2.99.912]
Author: Chris Wilson <chris@chris-wilson.co.uk>
Date: Fri May 30 09:45:15 2014 +0100
sna: Use manual detiling for downloads
Reported-by: Harald Judt <h.judt@gmx.at>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=80560
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
We don't really want to rendering into SHM pixmaps except for copying
back due to the strict serialisation requirements.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Chris Wilson
Damien Lespiau