libdrm is a little lax and does not allocate sufficient space for us to
safely use buffers on old gen. So compute the size we want for
ourselves.
Reported-by: Cem Aydin <cem.aydin@gmx.ch>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=80088
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Unlike GLXWindows, GLXPixmaps are rendered directly into, without a
staging copy. Therefore we must treat those carefully when exported and
clear our hints everytime control passes back to the Client.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=79999
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Panel Self-Refresh requires us to avoid frontbuffer rendering in order
to be power efficient. This is a job for TearFree!
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
The upload caches are special, along with having a bo->proxy, they also
claim to be completely damaged on both the GPU and CPU. Allow that to
pass through when discarding the proxy.
Reported-by: Nick Bowler <nbowler@draconx.ca>
Bugzilla; https://bugs.freedesktop.org/show_bug.cgi?id=79992
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
The region here has yet to be clipped, and so the only valid is-clipped
hint is from the flags computed from the PolyRect extents. Make sure we
use those when determining whether it is valid to discard damage.
Fixes regression from
commit ad03900688 [2.99.903]
Author: Chris Wilson <chris@chris-wilson.co.uk>
Date: Tue Sep 24 10:00:03 2013 +0100
sna: Separate out copy preferrence from operating in place decision
Reported-by: Nick Bowler <nbowler@draconx.ca>
Tested-by: Nick Bowler <nbowler@draconx.ca>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=79992
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
The call to GETPARAM requires either a rendernode or authorisation.
Therefore we can only assert that the fd is a valid i915 handle after
authorise() and not before.
Reported-by: Hans de Goede <hdegoede@redhat.com>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
As a last resort, continue to display the old cursor if we tried and
failed to create a new cursor.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Tricksy asserts making sure that the correct arguments are passed
around!
Reported-by: Arkadiusz Miskiewicz <arekm@maven.pl
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
It appears the kernel has a better idea when WT is supported on which
gen8 parts. I assumed it was always supported, but experience suggests
otherwise, so only use WT support when advertised.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=79967
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
As the BLT upload implies a sync (and a likely GPU boost), only do so if
we cannot do an immediate upload with the CPU, either directly to
backing memory or through the GTT.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Under a compositor, the current deltas may invert the sense of the copy
direction, causing scrolling corruption. Simplify handling those flags
by making them invariant for the function.
Reported-by: Bruno Prémont <bonbons@linux-vserver.org>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=79843
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
A few more places could use the common pixmap creator, just remember to
handle the SHMPixmap complication!
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
If the memory bw is constrained on the GPU avoid doing the 2-pass copy
for overlaps on the render ring, and do the single pass slower BLT copy
instead - as since it has to transfer less data it will be faster.
Reported-by: Ildar Nurislamov <absorbb@gmail.com>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=77436
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
sna_damage_contains_box() routine implicitly reduces the damage before
performing its check. This may alter and even destroy the damage entry,
so pass in the handle so that it can be updated correctly.
References: https://bugs.freedesktop.org/show_bug.cgi?id=77436
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
With an old kernel, we would fail to actually mark the display as part
of the scanout domain, but proceed to assign it a fb id. This caused our
asserts to report our bookkeeping error.
Reported-by: Pavel Ondračka <pavel.ondracka@email.cz>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=79909
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Given a kernel patch to fix coherency of physical objects, we can use a
more efficient method for updating cursors.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
As the kernel does not send out an explicit DISCOVER event, we have to
reinspect every connector after every hotplug event to detect topology
changes.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
In order to provide persistent names for MST devices, which may change
topology dynamically, the kernel exports a PATH blob. We can then use
this path to generate the output name, and so reuse existing XID when a
monitor is plugged back in.
Based on the patch for -modesetting by Dave Airlie.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Chris Wilson