This is an important optimization for avoiding read-after-write (RAW)
stalls in the HiZ buffer.  Certain workloads would run very slowly with
HiZ enabled, but run much faster with the "hiz=false" driconf option.
With this patch, they run at full speed even with HiZ.

Improves performance in OglVSInstancing by 3.2x on Broadwell GT3e
(Iris Pro 6200).

Thanks to Jesse Barnes and Ben Widawsky for their help in tracking this
down.  Thanks to Chris Wilson for showing me the new workarounds system.
Signed-off-by: NKenneth Graunke <kenneth@whitecape.org>
Cc: Jesse Barnes <jbarnes@virtuousgeek.org>
Reviewed-by: NVille Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

It is useful to know at debug time if we are keeping main link on.
Signed-off-by: NRodrigo Vivi <rodrigo.vivi@intel.com>
Reviewed-by: NDurgadoss R <durgadoss.r@intel.com>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

No functional changes on this patch. Just grouping the link_standy decision
to avoid miss any change. Also making this info available everywhere
which will help to decide when to use vbt's tp time on following patch.
Signed-off-by: NRodrigo Vivi <rodrigo.vivi@intel.com>
Reviewed-by: NDurgadoss R <durgadoss.r@intel.com>
[danvet: Slight editing of the commit message which was one huge
run-on sentence.]
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

Let's respect vbt full_link (link_standby) on source side as well.
Signed-off-by: NRodrigo Vivi <rodrigo.vivi@intel.com>
Reviewed-by: NDurgadoss R <durgadoss.r@intel.com>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

We have only two possible states with so many names and combinations that
might be confusing.

1 - Main link active / enabled / stand by / on
2 - Main link disabled / off / full off

Let's start organizing it by fixing a inverted logic when setting the sink bit.
Signed-off-by: NRodrigo Vivi <rodrigo.vivi@intel.com>
Reviewed-by: NDurgadoss R <durgadoss.r@intel.com>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

These conditions applies only to Haswell and we were also checking for them
on Valleyview/Cherryview.
Signed-off-by: NRodrigo Vivi <rodrigo.vivi@intel.com>
Reviewed-by: NDurgadoss R <durgadoss.r@intel.com>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

ON these platforms we don't have hardware tracking working for any case.
So we need to fake this on software by forcing psr to exit on every
flush.

Manual tests indicated this was needed.
Signed-off-by: NRodrigo Vivi <rodrigo.vivi@intel.com>
Reviewed-by: NDurgadoss R <durgadoss.r@intel.com>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

Description of the 'state' parameter for intel_plane_destroy_state() was
missing and the intel_atomic_plane.c file section heading did not match
drm.tmpl.
Signed-off-by: NMatt Roper <matthew.d.roper@intel.com>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

915 doens't support hotplug at all, so we shouldn't try to pretend
otherwise in the SDVO code.

Note: i915 does have hotplug support in hw, we simply never enabled it
in i915.ko: There's only one hpd bit for all outputs, so not worth the
bother to add this special case for this rather old platform.
Signed-off-by: NVille Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: NJani Nikula <jani.nikula@intel.com>
[danvet: Clarify that only i915.ko doesn't support hpd on i915g.]
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

If we determine that a specific port is eDP, don't register the HDMI
connector/encoder for it. The reason being that we want to disable
HPD interrupts for eDP ports when the display is off, but the presence
of the extra HDMI connector would demand the HPD interrupt to remain
enabled all the time.
Signed-off-by: NVille Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: NJani Nikula <jani.nikula@intel.com>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

The dev_priv->display.hpd_irq_setup hook is optional, so we can move the
I915_HAS_HOTPLUG() check out of i915_hpd_irq_setup() and only set up the
hook when hotplug support is present.
Signed-off-by: NVille Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: NJani Nikula <jani.nikula@intel.com>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

intel_hpd_irq_handler() walks the passed in hpd[] array assuming it
contains HPD_NUM_PINS elements. Currently that's not true as we don't
specify an explicit size for the arrays when initializing them. Avoid
the out of bounds accesses by specifying the size for the arrays.
Signed-off-by: NVille Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: NJani Nikula <jani.nikula@intel.com>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

This reverts commit 5a0afd4b.

Although timeout mode allows higher residency it impact badly on performance.
I believe while we don't have a way to balance between performance and
power savings at runtime I believe we have to revert and prioritize
performance that was impacted a lot.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=88103
Cc: Deepak S <deepak.s@linux.intel.com>
Cc: Wendy Wang <wendy.wang@intel.com>
Cc: Jani Nikula <jani.nikula@intel.com>
Signed-off-by: NRodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

Found by reading the HIZ_CHICKEN documentation.

Improves performance in a HiZ microbenchmark by around 50%.
Improves performance in OglZBuffer by around 18%.

Thanks to Chris Wilson for helping me figure out where to put this.
Signed-off-by: NKenneth Graunke <kenneth@whitecape.org>
Reviewed-by: NVille Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

A previous commit enabled execlists by default:

       commit 27401d12 ("drm/i915/bdw: Enable execlists by default where supported")

This allowed routine testing of execlists which exposed a regression when
resuming from suspend.  The cause was tracked down the to recent changes to the
ring init sequence:

       commit 35a57ffb ("drm/i915: Only init engines once")

During a suspend/resume cycle the hardware Context Status Buffer write pointer
is reset.  However since the recent changes to the init sequence the software CSB
read pointer is no longer reset.  This means that context status events are not
handled correctly and new contexts are not written to the ELSP, resulting in an
apparent GPU hang.

Pending further changes to the ring init code, just move the
ring->next_context_status_buffer initialization into gen8_init_common_ring to
fix this regression.

v2: Moved init into gen8_init_common_ring rather than context_enable after
feedback from Daniel Vetter.  Updated commit msg to reflect this and also cite
commits related to the regression.  Fixed bz link to correct bug.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=88096
Cc: Paulo Zanoni <paulo.r.zanoni@intel.com>
Cc: Daniel Vetter <daniel.vetter@ffwll.ch>
Cc: Dave Gordon <david.s.gordon@intel.com>
Signed-off-by: NThomas Daniel <thomas.daniel@intel.com>
Reviewed-by: NDave Gordon <david.s.gordon@intel.com>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

The userspace-requested plane coordinates are now always available via
plane->state.base (and the i915-adjusted values are stored in
plane->state), so we no longer use the coordinate fields in intel_plane
and can drop them.

Also, note that the error case for pageflip calls update_plane() to
program the values from plane->state; it's simpler to just call
intel_plane_restore() which does the same thing.

v2: Replace manual update_plane() with intel_plane_restore() in pageflip
    error handler.

Reviewed-by(v1): Bob Paauwe <bob.j.paauwe@intel.com>
Signed-off-by: NMatt Roper <matthew.d.roper@intel.com>
Reviewed-by: NAnder Conselvan de Oliveira <ander.conselvan.de.oliveira@intel.com>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

Switch plane handling to use the atomic plane helpers.  This means that
rather than provide our own implementations of .update_plane() and
.disable_plane(), we expose the lower-level check/prepare/commit/cleanup
entrypoints and let the DRM core implement update/disable for us using
those entrypoints.

The other main change that falls out of this patch is that our
drm_plane's will now always have a valid plane->state that contains the
relevant plane state (initial state is allocated at plane creation).
The base drm_plane_state pointed to holds the requested source/dest
coordinates, and the subclassed intel_plane_state holds the adjusted
values that our driver actually uses.

v2:
 - Renamed file from intel_atomic.c to intel_atomic_plane.c (Daniel)
 - Fix a copy/paste comment mistake (Bob)

v3:
 - Use prepare/cleanup functions that we've already factored out
 - Use newly refactored pre_commit/commit/post_commit to avoid sleeping
   during vblank evasion

v4:
 - Rebase to latest di-nightly requires adding an 'old_state' parameter
   to atomic_update;

v5:
 - Must have botched a rebase somewhere and lost some work.  Restore
   state 'dirty' flag to let begin/end code know which planes to
   run the pre_commit/post_commit hooks for.  This would have actually
   shown up as broken in the next commit rather than this one.

v6:
 - Squash kerneldoc patch into this one.
 - Previous patches have now already taken care of most of the
   infrastructure that used to be in this patch.  All we're adding here
   now is some thin wrappers.

v7:
 - Check return of intel_plane_duplicate_state() for allocation
   failures.

v8:
 - Drop unused drm_plane_state -> intel_plane_state cast.  (Ander)
 - Squash in actual transition to plane helpers.  Significant
   refactoring earlier in the patchset has made the combined
   prep+transition much easier to swallow than it was in earlier
   iterations. (Ander)

v9:
 - s/track_fbs/disabled_planes/ in the atomic crtc flags.  The only fb's
   we need to update frontbuffer tracking for are those on a plane about
   to be disabled (since the atomic helpers never call prepare_fb() when
   disabling a plane), so the new name more accurately describes what
   we're actually tracking.

Testcase: igt/kms_plane
Testcase: igt/kms_universal_plane
Testcase: igt/kms_cursor_crc
Signed-off-by: NMatt Roper <matthew.d.roper@intel.com>
Reviewed-by: NAnder Conselvan de Oliveira <conselvan2@gmail.com>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

A few of the sprite-related function names in i915 are very similar
(e.g., intel_enable_planes() vs intel_crtc_enable_planes()) and don't
make it clear whether they only operate on sprite planes, or whether
they also apply to all universal plane types.  Rename a few functions to
be more consistent with our function naming for primary/cursor planes or
to clarify that they apply specifically to sprite planes:

 - s/intel_disable_planes/intel_disable_sprite_planes/
 - s/intel_enable_planes/intel_enable_sprite_planes/

Also, drop the sprite-specific intel_destroy_plane() and just use
the type-agnostic intel_plane_destroy() function.  The extra 'disable'
call that intel_destroy_plane() did is unnecessary since the plane will
already be disabled due to framebuffer destruction by the point it gets
called.

v2: Earlier consolidation patches have reduced the number of functions
    we need to rename here.

v3: Also rename intel_plane_funcs vtable to intel_sprite_plane_funcs
    for consistency with primary/cursor.  (Ander)

v4: Convert comment for intel_plane_destroy() to kerneldoc now that it
    is no longer a static function.  (Ander)

Reviewed-by(v1): Bob Paauwe <bob.j.paauwe@intel.com>
Signed-off-by: NMatt Roper <matthew.d.roper@intel.com>
Reviewed-by: NAnder Conselvan de Oliveira <conselvan2@gmail.com>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

Move the vblank evasion up from the low-level, hw-specific
update_plane() handlers to the general plane commit operation.
Everything inside commit should now be non-sleeping, so this brings us
closer to how vblank evasion will behave once we move over to atomic.

v2:
 - Restore lost intel_crtc->active check on vblank evasion

v3:
 - Replace assert_pipe_enabled() in intel_disable_primary_hw_plane()
   with an intel_crtc->active test; it turns out assert_pipe_enabled()
   grabs some mutexes and can sleep, which we can't do with interrupts
   disabled.

v4:
 - Equivalent to v2; v3 change is now squashed into an earlier patch
   of the series.  (Ander).
Signed-off-by: NMatt Roper <matthew.d.roper@intel.com>
Reviewed-by: NAnder Conselvan de Oliveira <conselvan2@gmail.com>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

Once we integrate our work into the atomic pipeline, plane commit
operations will need to happen with interrupts disabled, due to vblank
evasion.  Our commit functions today include sleepable work, so those
operations need to be split out and run either before or after the
atomic register programming.

The solution here calculates which of those operations will need to be
performed during the 'check' phase and sets flags in an intel_crtc
sub-struct.  New intel_begin_crtc_commit() and
intel_finish_crtc_commit() functions are added before and after the
actual register programming; these will eventually be called from the
atomic plane helper's .atomic_begin() and .atomic_end() entrypoints.

v2: Fix broken sprite code split

v3: Make the pre/post commit work crtc-based to match how we eventually
    want this to be called from the atomic plane helpers.

v4: Some platforms that haven't had their watermark code reworked were
    waiting for vblank, then calling update_sprite_watermarks in their
    platform-specific disable code.  These also need to be flagged out
    of the critical section.

v5: Sprite plane test for primary show/hide should just set the flag to
    wait for pending flips, not actually perform the wait.  (Ander)

v6:
 - Rebase onto latest di-nightly; picks up an important runtime PM fix.
 - Handle 'wait_for_flips' flag in intel_begin_crtc_commit(). (Ander)
 - Use wait_for_flips flag for primary plane update rather than
   performing the wait in the check routine.
 - Added kerneldoc to pre_disable/post_enable functions that are no
   longer static.  (Ander)
 - Replace assert_pipe_enabled() in intel_disable_primary_hw_plane()
   with an intel_crtc->active test; it turns out assert_pipe_enabled()
   grabs some mutexes and can sleep, which we can't do with interrupts
   disabled.

v7:
 - Check for fb != NULL when deciding whether the sprite plane hides the
   primary plane during a sprite update.  (PRTS)
Signed-off-by: NMatt Roper <matthew.d.roper@intel.com>
Reviewed-by: NAnder Conselvan de Oliveira <conselvan2@gmail.com>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

If CONFIG_BUG=n __WARN_printf won't be defined leading to the below
build failure. The double underscores should have told us to steer clear
of it anyway.

drivers/gpu/drm/i915/intel_display.c: In function ‘assert_pll’:
drivers/gpu/drm/i915/intel_display.c:1027:2: error: implicit declaration
of function ‘__WARN_printf’ [-Werror=implicit-function-declaration]
  I915_STATE_WARN(cur_state != state,

Use WARN(1, ...) instead. It handles CONFIG_BUG=n gracefully and, with
the constant condition, a sane compiler should reduce it to
__WARN_printf.

This is a regression introduced by

commit e2c719b7
Author: Rob Clark <robdclark@gmail.com>
Date:   Mon Dec 15 13:56:32 2014 -0500

    drm/i915: tame the chattermouth (v2)
Reported-by: NJim Davis <jim.epost@gmail.com>
Reference: http://mid.gmane.org/CA+r1ZhgHTi7bS2irhtuSUs9aO=Br1dumN8=oAOeaMJDZ_ZhwBw@mail.gmail.com
Cc: Rob Clark <robdclark@gmail.com>
Signed-off-by: NJani Nikula <jani.nikula@intel.com>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

After switching to using the component interface this API isn't needed
any more.

v2-3: unchanged
v4:
- move the removal of i915_powerwell.h to this patch (Takashi)
Signed-off-by: NImre Deak <imre.deak@intel.com>
Reviewed-by: NTakashi Iwai <tiwai@suse.de>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

Register a component to be used to interface with the snd_hda_intel
driver. This is meant to replace the same interface that is currently
based on module symbol lookup.

v2:
- change roles between the hda and i915 components (Daniel)
- add the implementation to a new file (Jani)
- use better namespacing (Jani)
v3:
- move the implementation to intel_audio.c (Daniel)
- rename display_component to audio_component (Daniel)
- add kerneldoc (Daniel)
v4:
- run forgotten git rm i915_component.c (Jani)
Signed-off-by: NImre Deak <imre.deak@intel.com>
Reviewed-by: NTakashi Iwai <tiwai@suse.de>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

This will be needed by later patches, so factor it out.

No functional change.

v2:
- s/dev_to_i915_priv/dev_to_i915/ (Jani)
- don't use the helper in i915_pm_suspend (Chris)
- simplify the helper (Chris)
v3:
- remove redundant upcasting in the helper (Daniel)
Signed-off-by: NImre Deak <imre.deak@intel.com>
Reviewed-by: NTakashi Iwai <tiwai@suse.de>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

If not pinned VMA can become an eviction target just before it needs to be
executed which breaks the internal object lifetime rules.
Signed-off-by: NTvrtko Ursulin <tvrtko.ursulin@intel.com>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=87399Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

Sometimes we wish to tweak how an individual context behaves. Since we
always create a context for every filp, this means that individual
processes can fine tune their behaviour even if they do not explicitly
create a context.

The first example parameter here is to enable multi-process GPU testing,
but the interface should be able to cope with passing arbitrarily complex
parameters.
Signed-off-by: NChris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: NRodrigo Vivi <rodrigo.vivi@intel.com>
Reviewed-by: NMika Kuoppala <mika.kuoppala@intel.com>
Testcase: igt/gem_reset_stats/ban-period-*
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

It is platform/output depenedent when exactly the pipe will start
running. Sometimes we just need the (cpu) pipe enabled, in other cases
the pch transcoder is enough and in yet other cases the (DP) port is
sending the frame start signal.

In a perfect world we'd put the drm_crtc_vblank_on call exactly where
the pipe starts running, but due to cloning and similar things this
will get messy. And the current approach of picking the most
conservative place for all combinations also doesn't work since that
results in legit vblank waits (in encoder->enable hooks, e.g. the 2
vblank waits for sdvo) failing.

Completely going back to the old world before

commit 51e31d49
Author: Daniel Vetter <daniel.vetter@ffwll.ch>
Date:   Mon Sep 15 12:36:02 2014 +0200

    drm/i915: Use generic vblank wait

isn't great either since screaming when the vblank wait work because
the pipe is off is kinda nice.

Pick a compromise and move the drm_crtc_vblank_on right before the
encoder->enable call. This is a lie on some outputs/platforms, but
after the ->enable callback the pipe is guaranteed to run everywhere.
So not that bad really. Suggested by Ville.

v2: Same treatment for drm_crtc_vblank_off and encoder->disable: I've
missed the ibx pipe B select w/a, which also has a vblank wait in the
disable function (while the pipe is obviously still running).

Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Acked-by: NVille Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: NDaniel Vetter <daniel.vetter@intel.com>

This will allow us to set per-file, or even per-context, periods in the
future.
Signed-off-by: NChris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: NRodrigo Vivi <rodrigo.vivi@intel.com>
Reviewed-by: NMika Kuoppala <mika.kuoppala@intel.com>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

This reverts commit 371abae8.

This data seems unreliable and causing many issues and blocking other
teams and feature implementation. Safest way is to revert that for now.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=88081
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=88039
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=87671
Cc: Vandana Kannan <vandana.kannan@intel.com>
Cc: Deepak M <m.deepak@intel.com>
Cc: Jani Nikula <jani.nikula@intel.com>
Cc: Ben Widawsky <ben@bwidawsk.net>
Cc: Kristian Høgsberg <hoegsberg@gmail.com>
Signed-off-by: NRodrigo Vivi <rodrigo.vivi@intel.com>
Reviewed-by: NJani Nikula <jani.nikula@intel.com>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

This patch changes kfd_ioctl() to be very similar to drm_ioctl().

The patch defines an array of amdkfd_ioctls, which maps IOCTL definition to the
ioctl function.

The kfd_ioctl() uses that mapping to call the appropriate ioctl function,
through a function pointer.

This patch also declares a new typedef for the ioctl function pointer.

v2: Renamed KFD_COMMAND_(START|END) to AMDKFD_...
Signed-off-by: NOded Gabbay <oded.gabbay@amd.com>
Acked-by: NChristian König <christian.koenig@amd.com>

This patch reformats the ioctl definitions in kfd_ioctl.h to be similar to the
drm ioctls definition style.

v2: Renamed KFD_COMMAND_(START|END) to AMDKFD_...
Signed-off-by: NOded Gabbay <oded.gabbay@amd.com>
Acked-by: NChristian König <christian.koenig@amd.com>

This patch moves the copy_to_user() and copy_from_user() calls from the
different ioctl functions in amdkfd to the general kfd_ioctl() function, as
this is a common code for all ioctls.

This was done according to example taken from drm_ioctl.c
Signed-off-by: NOded Gabbay <oded.gabbay@amd.com>
Reviewed-by: NChristian König <christian.koenig@amd.com>

This patch provides support to create write-combining virtual mappings of
GEM object. It intends to provide the same funtionality of 'mmap_gtt'
interface without the constraints and contention of a limited aperture
space, but requires clients handles the linear to tile conversion on their
own. This is for improving the CPU write operation performance, as with such
mapping, writes and reads are almost 50% faster than with mmap_gtt. Similar
to the GTT mmapping, unlike the regular CPU mmapping, it avoids the cache
flush after update from CPU side, when object is passed onto GPU.  This
type of mapping is specially useful in case of sub-region update,
i.e. when only a portion of the object is to be updated. Using a CPU mmap
in such cases would normally incur a clflush of the whole object, and
using a GTT mmapping would likely require eviction of an active object or
fence and thus stall. The write-combining CPU mmap avoids both.

To ensure the cache coherency, before using this mapping, the GTT domain
has been reused here. This provides the required cache flush if the object
is in CPU domain or synchronization against the concurrent rendering.
Although the access through an uncached mmap should automatically
invalidate the cache lines, this may not be true for non-temporal write
instructions and also not all pages of the object may be updated at any
given point of time through this mapping.  Having a call to get_pages in
set_to_gtt_domain function, as added in the earlier patch 'drm/i915:
Broaden application of set-domain(GTT)', would guarantee the clflush and
so there will be no cachelines holding the data for the object before it
is accessed through this map.

The drm_i915_gem_mmap structure (for the DRM_I915_GEM_MMAP_IOCTL) has been
extended with a new flags field (defaulting to 0 for existent users). In
order for userspace to detect the extended ioctl, a new parameter
I915_PARAM_MMAP_VERSION has been added for versioning the ioctl interface.

v2: Fix error handling, invalid flag detection, renaming (ickle)

v3: Rebase to latest drm-intel-nightly codebase

The new mmapping is exercised by igt/gem_mmap_wc,
igt/gem_concurrent_blit and igt/gem_gtt_speed.

Change-Id: Ie883942f9e689525f72fe9a8d3780c3a9faa769a
Signed-off-by: NAkash Goel <akash.goel@intel.com>
Signed-off-by: NChris Wilson <chris@chris-wilson.co.uk>
Cc: Daniel Vetter <daniel.vetter@ffwll.ch>
Reviewed-by: NTvrtko Ursulin <tvrtko.ursulin@linux.intel.com>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

Previously, this was restricted to only operate on bound objects - to
make pointer access through the GTT to the object coherent with writes
to and from the GPU. A second usecase is drm_intel_bo_wait_rendering()
which at present does not function unless the object also happens to
be bound into the GGTT (on current systems that is becoming increasingly
rare, especially for the typical requests from mesa). A third usecase is
a future patch wishing to extend the coverage of the GTT domain to
include objects not bound into the GGTT but still in its coherent cache
domain. For the latter pair of requests, we need to operate on the
object regardless of its bind state.

v2: After discussion with Akash, we came to the conclusion that the
get-pages was required in order for accurate domain tracking in the
corner cases (like the shrinker) and also useful for ensuring memory
coherency with earlier cached CPU mmaps in case userspace uses exotic
cache bypass (non-temporal) instructions.

v3: Fix the inactive object check.

v4: Rebase to latest drm-intel-nightly codebase
Signed-off-by: NChris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: NAkash Goel <akash.goel@intel.com>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

v2: Use WARN_ONs (Daniel)

Cc: Daniel Vetter <daniel@ffwll.ch>
Signed-off-by: NBen Widawsky <ben@bwidawsk.net>
Signed-off-by: NMichel Thierry <michel.thierry@intel.com>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

I've written these long before we've had a reasonable docbook
structure, and naturally they've gone stale. Fix this up asap.
Signed-off-by: NDaniel Vetter <daniel.vetter@intel.com>

Haswell significantly improved the performance of sampler_c messages,
but the optimization appears to be off by default.  Later platforms
remove this bit, and apparently always enable the optimization.

Improves performance in "Counter Strike: Global Offensive" by 18%
at default settings on Iris Pro.

This may break sampling of paletted formats (P8/A8P8/P8A8).  It's
unclear whether it affects sampling of paletted formats in general,
or just the sample_c message (which is never used).

While libva does have support for using paletted formats (primarily
for OSDs), that support appears to have been broken for at least a
year, so I couldn't observe a regression from this:

I tried to get libva-intel to use paletted formats, and observe a
regression...but the only thing I found that used it was mplayer's OSD
(on screen display).  Even without my patch, the colors were totally
wrong with that, and it's according to a few distro wikis, that's been
the case for over a year.

If libva's code for paletted formats /is/ broken, they could always
add code to disable this bit using the command validator when fixing
it.

Further investigation from Haihao shows that libva mplayer OSD seems
to work at least on his setup (still unclear what's wron with Ken's),
and that it's not affected by this patch. Quoting the discussion
between Haihao and Ken:

> > > If you use "-vo gl" or "-vo xv", the OSD is solid white text with a black
> > > border around it.  I presume that it's supposed to be white with vaapi as
> > > well, but I guess I'm not entirely sure.
> > >
> > > It's possible that the optimization doesn't affect the palette as long as
> > > you never use sample_c with the paletted textures.
> >
> > I verified the palette takes effect in the following way:
> >
> > 1. Only support P8A8 format in the driver
> >
> > 2. ran the above command and I saw white OSD text
> >
> > 3. Only support P4A4 format in the driver and don't use
> > 3DSTATE_SAMPLER_PALETTE_LOAD0 to load the value to the texture palette,
> > so the palette keeps unchanged.
> >
> > 4. ran the above command and I saw black OSD text.
> >
> > 5. Load the right value to the texture palette and ran the above command
> > again, I saw white OSD text.
> >
> > Hence I think sample_c with the paletted textures is used in the driver.
>
> That sounds like the palette is actually working, then.  Great :)
>
> I doubt that libva would use sample_c - sampling with a shadow comparison?
> It looks like it just uses sample and sample+killpix.

You are right, libva driver doesn't use sample_c message.

> I'm pretty sure the sample_c optimization just uses the palette memory as
> storage for some stuff, so it's quite possible it just works if you're
> only using sample and sample+killpix.

Thanks for the explanation, it makes sense to me.
Signed-off-by: NKenneth Graunke <kenneth@whitecape.org>
[danvet: Add wa name from Ville's review to the comment and copypaste
the explanation why we don't care about libva (already broken) from
Ken. Also add conclusion from libva devs that&why this is all fine.]
Reviewed-by: NVille Syrjälä <ville.syrjala@linux.intel.com>
Cc: "Xiang, Haihao" <haihao.xiang@intel.com>
Cc: libva@lists.freedesktop.org
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

The test:

	if (size > RADEON_MAX_TEXTURE_SIZE) {

"size" is an integer and it's controled by the user so it can be
negative and the test can underflow.  Later we use "size" in:

	dwords = size / 4;
	...
	RADEON_COPY_MT(buffer, data, (int)(dwords * sizeof(u32)));

It causes memory corruption to copy a negative size buffer.
Signed-off-by: NDan Carpenter <dan.carpenter@oracle.com>
Reviewed-by: NChristian König <christian.koenig@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

Enabling bapm seems to cause clocking problems on some
KV configurations.  Disable it by default for now.
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>
Cc: stable@vger.kernel.org

The check was already in place in the dp mode_valid check, but
radeon_dp_get_dp_link_clock() never returned the high clock
mode_valid was checking for because that function clipped the
clock based on the hw capabilities.  Add an explicit check
in the mode_valid function.

bug:
https://bugs.freedesktop.org/show_bug.cgi?id=87172Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>
Cc:stable@vge.kernel.org