sna: Be more flexible in converting random subsamples to 8x8 tiles

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>

sna: Mark up const boxes to keep the compiler happy

Pass around const to suppress a few compiler warnings.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>

sna: Feed more operations into the 8x8 BLT tiler

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>

sna: Color patterns for BLT are required to be aligned to 256 byte boundaries

This so far appears to be the most restrictive alignment required, so
simply increase the upload alignment to 256 bytes.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=80033
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>

test: Increase number of tiled sources

Significantly improve the stress impose upon the tiled BLT operations.
Also start dumping pngs of the failures.

References: https://bugs.freedesktop.org/show_bug.cgi?id=80033
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>

sna/dri2: Don't decouple front buffer tracking for invalid SwapBuffer calls

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>

sna: Mark upload from SHM segment as read-only

As this may be mapped by the Xserver using a read-only SHM segment, we
are forced to treat it always as read-only. And this being X, that it is
using a SHM segment is opaque to the driver. Fantastic middlelayer.

This was incorrectly removed in
commit e680e54eab6ffa72e5e1eb6cc0e3fe4b235b06a1
Author: Chris Wilson <chris@chris-wilson.co.uk>
Date:   Wed Jun 11 10:48:26 2014 +0100

    sna: Ignore setting read-only for temporary userptr maps

Also let's not forget the ShmPutImage -> CopyArea path.

Reported-by: Christoph Haag <haagch.christoph@googlemail.com>
Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=78411
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>

sna: Mark upload from SHM segment as read-only

As this may be mapped by the Xserver using a read-only SHM segment, we
are forced to treat it always as read-only. And this being X, that it is
using a SHM segment is opaque to the driver. Fantastic middlelayer.

This was incorrectly removed in
commit e680e54eab6ffa72e5e1eb6cc0e3fe4b235b06a1
Author: Chris Wilson <chris@chris-wilson.co.uk>
Date:   Wed Jun 11 10:48:26 2014 +0100

    sna: Ignore setting read-only for temporary userptr maps

Reported-by: Christoph Haag <haagch.christoph@googlemail.com>
Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=78411
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>

sna: Missed fixing y tile offset in last commit

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>

sna: Use the right pattern origin for tiled 8x8 extraction

References: https://bugs.freedesktop.org/show_bug.cgi?id=80033
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>

sna: Fix up small extents for 8x8 tile construction

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>

sna: Wrap pointer access for 8x8 tile construction

The kernel can quite happily send a SIGBUS whenever we read from an bo
(due to oom), so catch it and handle it.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>

sna: Always readback shadow damage after replacing the buffer

Instead track what we explicitly want to discard for the next operation,
so that we don't have to copy back the whole buffer if we have to
replace the shadow and intend to overwrite it all.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>

sna/dri2: Replace assertion with code to skip updating the back buffer

After a window resize or redirection, the DRI2BufferPtr become stale but
it may take a while for the Client to notice and reconfigure. Just
skip touching stale backbuffers and let the Client catch up.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>

sna: Fix modulus for partial conversion of tiles to 8x8 patterns

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>

sna: Delete the silly typo

Not checking DAMAGE_IS_ALL() correctly leads to an assertion failure.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>

sna: Do not treat both src/dst having no clones as being the same

Fixes regression from
commit 62aaf2ff4f8597067cf387865707baa00ed9a123
Author: Chris Wilson <chris@chris-wilson.co.uk>
Date:   Wed Jun 18 18:41:15 2014 +0100

    sna: Skip redundant copies when already cloned

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>

sna: Add log breadcrumbs with valgrind

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>

sna: Fix DBG compilation

Missed updaing the DBG message in

commit 83c0f034454ef0f474126a3398e5e790ac5ef842
Author: Chris Wilson <chris@chris-wilson.co.uk>
Date:   Wed Jun 18 16:51:46 2014 +0100

    sna: Pass desired CRTC viewport for completing single CRTC flips

References: https://bugs.freedesktop.org/show_bug.cgi?id=80157
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>

sna: Skip redundant copies when already cloned

References: https://bugs.freedesktop.org/show_bug.cgi?id=77436
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>

sna: Pass desired CRTC viewport for completing single CRTC flips

We cannot simply compare against the crtc->shadow_bo for all our needs,
so pass in exactly the setup we want and apply that. In particular this
is required when flipping away from the single CRTC setup on secondary
pipes.

Fixes TearFree multi-monitor regression from
commit 3932e97057fca16615adaefbc1eb25a0d51a1d8b [2.99.912]
Author: Chris Wilson <chris@chris-wilson.co.uk>
Date:   Mon Jun 9 08:58:15 2014 +0100

    sna/dri2: Allow TearFree flipping to individual CRTC

Reported-by: Leo Wolf <jclw@ymail.com>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=80191

uxa: Mark the output as off before making pipe changes

This is a partial "backport" of commit c6cd10f536, which makes the same
change for sna, to avoid users still using uxa ending up with a blackscreen
after plugging in an external monitor.

Bugzilla: https://bugzilla.redhat.com/show_bug.cgi?id=1032978
Bugzilla: https://bugzilla.redhat.com/show_bug.cgi?id=1103806
Signed-off-by: Hans de Goede <hdegoede@redhat.com>

uxa: When the output is off, report the cached backlight value

This is a "backport" of commit b545e10c50cb to uxa, so that users who are
still using uxa, don't end up with a black screen after suspend / resume.

Bugzilla: https://bugzilla.redhat.com/show_bug.cgi?id=1032978
Bugzilla: https://bugzilla.redhat.com/show_bug.cgi?id=1103806
Signed-off-by: Hans de Goede <hdegoede@redhat.com>

sna/gen8: Fix cleanup after failing to convert channel

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>

sna/gen8: Fix unaligned() to return true when not aligned

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>

sna: Assert that we do not try to mmap a proxy

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>

backlight: Fix compiler error

/usr/include/xorg/os.h around line 579 reads:

extern _X_EXPORT char *
strndup(const char *str, size_t n);

However strndup is already defined by glibc, and this redefine causes a
compile error.

This gets triggered because backlight.c does:

Without first doing:

Causing HAVE_STRNDUP to not be defined.

Signed-off-by: Hans de Goede <hdegoede@redhat.com>

sna: Use nxm tiled blits for small regions of large tiles

If the region we are blitting is either narrow or short, we may be able
to construct a tiling pattern out of a large pixmap.

In the process, spot a regression due to

commit 542aeca6e67fc64f9133ca3e27ac8eca28af6d25
Author: Chris Wilson <chris@chris-wilson.co.uk>
Date:   Tue Jun 17 12:26:30 2014 +0100

    sna: Tweak creation 8x8 tiled patterns

which copied too much from the source line.

References: https://bugs.freedesktop.org/show_bug.cgi?id=79888
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>

sna: Simplify sna_pixmap_get_source

kgem_bo_upload_image was open-coded, so replace it with a call to the
common function

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>

sna: Add DBG option to disable snoop bo cacheing

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>

sna/dri2: fix invalid DBG string

References: https://bugs.freedesktop.org/show_bug.cgi?id=80157
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>

sna: Set the correct mapping type after calling mmap__async

References: https://bugs.freedesktop.org/show_bug.cgi?id=80001
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>

sna: Fix computing source bo for BLT operations

If we migrate the pixmap to the GPU, use the GPU bo. This may fix an issue
where we might end up using the CPU bo in a rare circumstance.

References: https://bugs.freedesktop.org/show_bug.cgi?id=80033
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>

sna: Align some function parameters

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>

sna: Remove invalid assert

Present and flip fallback may queue self-flips, so the assertion that
the crtc->flip_bo != crtc->bo is false.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>

sna: Cache small pixmaps rather than forcing a large fallback for BLT operations

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>

sna: Relax PREFER_GPU so that we don't needless create small GPU bo

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>

sna: Tweak creation 8x8 tiled patterns

Avoid reading back from a GTT mmapping.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>

sna/dri2: Fixup failure to do a render copy for DRI2CopyRegion

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>

sna: Add a few missing checks for all-damaged

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>

sna: Discard GPU damage for the magic upload proxy

The magic upload proxy is initially considered damaged in both the GPU
and CPU.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>

sna/glyphs: DBG compile fix

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>

sna: Regularly check the cache level on bo

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>

sna/dri2: Allow single crtc exchanges to be queued up

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>

uxa: Allocate frontbuffer to meet old fence constaints

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>

sna/glyphs: Add DBG option to force use of the glyph image mask

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>

uxa: Add some explanation to why bo were rejected

References: https://bugs.freedesktop.org/show_bug.cgi?id=80088
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>

sna: Fix logic inversion for final placement in is_gpu_dst()

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>

sna: Promote better active buffer reuse

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>

sna/glyphs: Show number of glyphs in DBG

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>

sna: Fix some batch DBG

Helps to check for a successful read when we want to show the batch
after execution.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>

sna: Clear our private hints about front rendering exported bo

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>

sna: Add more DBG to track transitions between CRTC and its shadow or flip bo

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>

test: Add a simple SHM test skeleton

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>

sna: Enable TearFree by default for systems with PSR

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>

sna: Fix assertions for discarding upload caches

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>

sna: Use the right is-clipped hint

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 ad0390068832ad4727371902fe41a85a53de1894 [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>

sna: Only discard the CPU bo if we have a GPU bo to transfer the damage to

References: https://bugs.freedesktop.org/show_bug.cgi?id=79992
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>

intel-virtual-output: Compile fixes

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>

intel-virtual-output: Disable outputs on remote displays upon shutdown

References: https://bugs.freedesktop.org/show_bug.cgi?id=79994
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>

intel-virtual-output: Initialise image after clone

As the image inherits its width/height from the clone, we need to set
those first on the clone.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=79994
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>

intel-virtual-output: Only call XShmDetach on a valid SHM segment

Saves a couple of extraneous XErrors.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>

intel-virtual-output: Check for an error creating the DRI3 fd

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=79994
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>

sna: Refine decision whether to use the GPU or render inplace

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>

sna: Track the pageflip bo's busyness

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>

intel: Check that the fd points to i915 after authorising

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>

sna/gen6+: Prefer the BLT for large overlapping copies

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>

sna: Try to reuse the current cursor if we fail to update

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>

xvmc: Refactor mutex locking

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>

xvmc: Sanitize compile and library flags

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>

configure: Fix DRI2 build

One s/DRI2/have_dri2/ too far.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>

sna: Reorder ignore_cpu_damage() to avoid asserts

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>

sna/gen8: Allow reads from scanout to be cached

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>

sna/gen8: Check for WT support before attempting to use the WT mocs

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>

configure: Allow manual disabling of each DRI level

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>

sna: Try to use CPU uploads before BLT uploads

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>

sna: Refactor USE_INPLACE debug macro

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>

sna/gen8: Clear a couple more push constants

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>

sna: Remove bogus assertion

The sna_pixmap is cleared afterwards, it does not need to be clear on
free.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>

sna: Fix direction flags for memmove

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>

sna: Use the pixmap header creation routine

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>

sna/gen6+: On small GT avoid the intermediate copy for overlaps

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/gen6+: Overlapping bo by definition have compatible depths

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>

sna: First copy to overlap temporary must be using GXcopy

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>

sna: Remove debugging hunk

Stray hunk in self-copy boxes.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>

sna: Update damage pointer after the implicit reduction

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>

sna: Fix scanout creation routine for old kernels <= 3.11

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>

sna: Add some DBG to the memmove path

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>

sna: Tweak self-copy boxes to hit the GPU more often

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>

sna: Recheck source bo after migrating dst (in case of src == dst)

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>

sna: Skip redundant clears

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>

sna/glyphs: Always print some DBG when performing fallback

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>

sna: Add some asserts to track redundant damage operations

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>

sna/gen8: Check offset alignment before using a userptr for RENDER

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=79053
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>

sna: Support efficient cursor updates for old machines

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>

sna: MST topologies may change on any hotplug event

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>

sna: Create a stable output name based on MST topology

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>

sna: Ignore setting read-only for temporary userptr maps

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>

sna: Prefer to operate inplace on the GPU rather than stall on the CPU

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>

sna: Discard TearFree damage before checking for an overwriting upload

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>