We no longer unconditionally clear the CPU hint when moving to the GPU
for readonly updates, so we can no longer assert that the priv->cpu ==
false afterwards.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
We need to also take account of upload buffers which include an offset
to the base map. To do so, we also need to stop conflating the cpu hint
with the current mapping flag.
References: https://bugs.freedesktop.org/show_bug.cgi?id=70461
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
It didn't consider the height of the glyph above the baseline, i.e. it
was fundamentally flawed. The only thing to do is to make sure that
glyph_extents() is sufficiently fast never to show up in profiles.
(Until then QA should spot the ~10% regression.) An alternative would be
to feed the drawable clip extents to the render engine and avoid manual
clipping if the clip region covers the whole drawable.
Reported-by: Clemens Eisserer <linuxhippy@gmail.com>
Reported-by: Jiri Slaby <jirislaby@gmail.com>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=70552
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
And make sure we consider such overflowing strings for correct clipping
against Windows.
To offset the cost of doing a full extents check (~10% on aa10text), we
introduce an approximate extents query (~1% on aa10text). The disparity
should be rare, and should be an overestimate to force redundant
clipping.
Reported-by: Clemens Eisserer <linuxhippy@gmail.com>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=70541
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Pavel Roskin found that with a 32-bit build of the DDX with a 64-bit
kernel that the call to GETCONNECTOR was overwriting the 4 bytes beyond
the end of the drm_mode_get_connector structure. This would appear to be
due to the surreptious padding inserted by the compiler so that the
structure would be naturally aligned on a 64-bit system. To compenstate
we need to insert padding between the adjacent 32-bit structures on the
stack.
As usual, be paranoid and make sure that all the adjacent KMS structs we
use are padded out to an 64-bit boundary.
Reported-by: Pavel Roskin <proski@gnu.org>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
This reverts commit 41b6b792d8 in favour
of the better fix to not ask RRChangeOutputProperty to reset the known
hardware values.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
When we assign the hardware values to the output properties, we do not
need to process the set_property callback to write those values back to
hardware. This callback is triggered by the pending update flag passed
to RRChangeOutputProperty.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=70406
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
When querying the current backlight value, be sure not to overwrite the
last user set value by the call to RRChangeOutputProperty.
RRChangeOutputProperty calls into set_backlight_property, tricking us
into believing that the user has set a new backlight value. The result
is that we end up believing that the user chooses a 0 backlight if she
should happen to query the property whilst the output is disabled.
Reported-by: reztho@archlinux.us
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=70406
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Remember to set it initially to false so that we randomly do not start
using fallbacks.
Fixes regression from
commit 0cd2c43fa8
Author: Chris Wilson <chris@chris-wilson.co.uk>
Date: Sun Oct 13 14:46:45 2013 +0100
sna/trapezoids: Use the aligned fast path for fallbacks
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
As we use the current ring to encode upon the bo relocs, we need that
ring to be properly setup prior to performing relocations.
References: https://bugs.freedesktop.org/show_bug.cgi?id=70461
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Pass along the flags for read/write direction so that we know whether or
not we expect the gpu bo to be in the CPU write domain.
Reported-by: Pavel Ondračka <pavel.ondracka@email.cz>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=70436
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
commit 951f969fa6
Author: Chris Wilson <chris@chris-wilson.co.uk>
Date: Sun Oct 13 12:47:20 2013 +0100
sna: Update DPMS on attached outputs before disabling the CRT
left behind a couple of variables now deceased.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
We may want to take action such as preserving the current user value of
the backlight before disabling it whilst forcing a CRTC off. This
requires us to record that value first.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
If the second syscall fails (presumably as a deferred allocation failure
check), cleanup the allocations made so far before reporting the
failure.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
For a large screen, we have to create a temporary surface for rendering
the textured video. If this pixmap creation fails we may be left with a
system memory only pixmap leading to a segfault.
Reported-by: Bas Wijnen <wijnen@debian.org>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
One of UXA's invarients is that the ScreenPixmap is complete (i.e. has
an intel_pixmap private with a bo attached). If we fail to create that
private during CreateScreenResources we will die very soon afterwards,
so just report the failure and shutdown gracefully.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Recent kernels gained the ability to report the actual size of the
dma-buf through an lseek. We can use this to set the correct size of the
bo when available, overriding the guess provided by the caller.
Suggested-by: Kristian Høgsberg <krh@bitplanet.net>
Suggested-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
During initialisation, we stash the currently attached CRTC id in
output->crtc. This is fine as ordinarily we would not dereference
output->crtc until after it had been assigned a real CRTC. However,
commit 6fda305e2f
Author: Chris Wilson <chris@chris-wilson.co.uk>
Date: Wed Oct 9 15:59:42 2013 +0100
sna: Append the current mode to the output list if not found
introduces such an early dereference and causes a crash if we fail to
probe the KMS configuration (usually due to a user override).
Reported-by: Łukasz Maśko <ed@yen.ipipan.waw.pl>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Add xfixes to the list of PKG_CHECK_MODULES for X11. '-lXfixes' was
hardcoded in test/Makefile.am before. This could lead to a broken build
in very rare cases where the build environment has all specified X
libraries but Xfixes.
Signed-off-by: Daniel Martin <consume.noise@gmail.com>
An extra caveat to these generations for
commit 97d809c26b
Author: Chris Wilson <chris@chris-wilson.co.uk>
Date: Thu Oct 10 00:15:55 2013 +0100
sna: Pass usage hint down to render fill routines
is that we don't want to incur ring switch overheads that may overwhelm
any advantages from using the BLT.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
For the scanlines emitted for rendering Core drawing primitives, it is
preferable to use the BLT engine, so pass those hints down.
Reported-by: Jiri Slaby <jirislaby@gmail.com>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
If we cancel an operation after partially committing it, we may leave
the batch bookkeeping in an inconsistent state with an exec object with
a zero-length batch. Ordinarily, this would not be an issue as we could
pass the extra object to the next batch. However, if we switch rings, we
need to clear the extra objects as they are currently flagged as being
on the wrong ring, leading to hilarity.
Reported-by: Jiri Slaby <jirislaby@gmail.com>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Rather than duplicating a string, we can simply transfer ownership from
the temporary mode to the mode list.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
If for some reason the current mode on the CRTC (inherited most likely
from fastboot) doesn't match any of the modes reported by the output, we
end up with a stray mode that the client cannot control.
Reported-by: Jiri Slaby <jirislaby@gmail.com>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=70132
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Upon aligning the buffer, we may enlarge the vbo to accomodate the
vertex alignment and push the current index past the end of the buffer.
Move the space check from before the alignment computation to
afterwards.
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>
We need to be careful not to execute threads past the end of the alloted
buffer by making sure the clip extents correctly align.
Reported-by: Joseph Yasi <joe.yasi@gmail.com>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=70204
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Instead trying to allocate 4100 bytes, fix the logic to only require a
maximum of 4096 bytes in the cache buffer.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
If the trapezoids are rectilinear, they should hit a fast path through
the span compositors and so threading them seems pointless. Expect
possibily for inplace pixman operations.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Chris Wilson
Daniel Martin