firefly-linux-kernel-4.4.55.git
10 years agodrm/i915: unify remaining register save/restore code a bit
Jani Nikula [Wed, 12 Nov 2014 15:01:10 +0000 (17:01 +0200)]
drm/i915: unify remaining register save/restore code a bit

Use the same conditions, group by features, add comments.

Signed-off-by: Jani Nikula <jani.nikula@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: don't save/restore backlight hist ctl registers
Jani Nikula [Wed, 12 Nov 2014 14:25:43 +0000 (16:25 +0200)]
drm/i915: don't save/restore backlight hist ctl registers

This is not used within the driver, and merely saving/restoring these
registers isn't going to do any good anyway. In fact, it's possible it's
actively harmful. Any code enabling the feature should handle this
completely in the regular platform specific enable/disable backlight
functions.

Signed-off-by: Jani Nikula <jani.nikula@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: don't save/restore panel fitter registers
Jani Nikula [Wed, 12 Nov 2014 14:25:42 +0000 (16:25 +0200)]
drm/i915: don't save/restore panel fitter registers

AFAICT i9xx_pfit_disable() on the GMCH display crtc disable path in
i9xx_crtc_disable() will always disable the panel fitter by writing 0 to
PFIT_CONTROL. The register save will always save/restore 0. Also we
completely recompue both in intel_gmch_panel_fitting so there's no way
we depend upon leftover bits.

Move the PFIT_CONTROL and PFIT_PGM_RATIOS save/restore to UMS
code. While at it, save/restore them both under the same conditions.

Signed-off-by: Jani Nikula <jani.nikula@intel.com>
[danvet: Make it a bit clearer that we nowhere depend upon these
bits.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: remove the unnecessary block around display.hpd_irq_setup
Jani Nikula [Wed, 12 Nov 2014 12:48:52 +0000 (14:48 +0200)]
drm/i915: remove the unnecessary block around display.hpd_irq_setup

The block was added for spin_lock_irqsave flags, but since the locking
was converted to spin_lock_irq variant, the block is no longer needed.

Signed-off-by: Jani Nikula <jani.nikula@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: restore RSTDBYCTL only on non-KMS paths
Jani Nikula [Tue, 11 Nov 2014 14:48:04 +0000 (16:48 +0200)]
drm/i915: restore RSTDBYCTL only on non-KMS paths

Since RSTDBYCTL is only saved on non-KMS path in within i915_save_state,
move the restore in i915_restore_state for symmetry.

Signed-off-by: Jani Nikula <jani.nikula@intel.com>
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915/vlv: don't save panel power sequencer registers on suspend
Jani Nikula [Tue, 11 Nov 2014 14:48:03 +0000 (16:48 +0200)]
drm/i915/vlv: don't save panel power sequencer registers on suspend

Don't save the panel power sequencer register on vlv/chv for two simple
reasons. First, these are the wrong registers to save to begin
with. Second, they are not restored anyway.

Signed-off-by: Jani Nikula <jani.nikula@intel.com>
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Wait thread status on gen8+ fw sequence
Mika Kuoppala [Mon, 10 Nov 2014 12:52:50 +0000 (04:52 -0800)]
drm/i915: Wait thread status on gen8+ fw sequence

As per latest pm guide, we need to do this also on
past hsw.

Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Mika Kuoppala <mika.kuoppala@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Initialize workarounds in logical ring mode too
Michel Thierry [Tue, 11 Nov 2014 16:47:33 +0000 (16:47 +0000)]
drm/i915: Initialize workarounds in logical ring mode too

Following the legacy ring submission example, update the
ring->init_context() hook to support the execlist submission mode.

v2: update to use the new workaround macros and cleanup unused code.
This takes care of both bdw and chv workarounds.

v2.1: Add missing call to init_context() during deferred context creation.

v3: Split init_context (emit) in legacy/lrc modes. For lrc, get the ringbuf
from the context (Mika/Daniel).

v4: Merge init_context interfaces back, the legacy mode only needs the ring,
but the lrc mode needs the ring and context (Mika).

Issue: VIZ-4092
Issue: GMIN-3475
Change-Id: Ie3d093b2542ab0e2a44b90460533e2f979788d6c
Cc: Deepak S <deepak.s@intel.com>
Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Cc: Daniel Vetter <daniel@ffwll.ch>
Signed-off-by: Michel Thierry <michel.thierry@intel.com>
Signed-off-by: Arun Siluvery <arun.siluvery@linux.intel.com>
Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com>
[danvet: Align function paramater lists properly.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Read the CCK fuse register from CCK
Ville Syrjälä [Fri, 7 Nov 2014 19:33:43 +0000 (21:33 +0200)]
drm/i915: Read the CCK fuse register from CCK

When reading a CCK register we should obviously read it from CCK not
Punit. This problem has been present ever since this of code was
introduced in

 commit 67c3bf6f55a97a0915a0f9ea07278a3073cc9601
 Author: Deepak S <deepak.s@linux.intel.com>
 Date:   Thu Jul 10 13:16:24 2014 +0530

    drm/i915: populate mem_freq/cz_clock for chv

The problem was raised during review by Mika [1] but somehow slipped
through the cracks, and the patch got applied with the problem unfixed.

[1] http://lists.freedesktop.org/archives/intel-gfx/2014-July/048937.html

Cc: Deepak S <deepak.s@linux.intel.com>
Cc: Mika Kuoppala <mika.kuoppala@intel.com>
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Deepak S <deepak.s@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Add the predicate source registers to the register whitelist
Neil Roberts [Fri, 7 Nov 2014 19:00:26 +0000 (19:00 +0000)]
drm/i915: Add the predicate source registers to the register whitelist

The predicate source registers are needed to implement conditional
rendering without stalling. The two source registers are used to load
the previous values of the PS_DEPTH_COUNT register saved from
PIPE_CONTROL commands. These can then be compared and used to set the
predicate enable bit via the MI_PREDICATE command.

The command parser version number is increased to 2 to make it easier
to detect the new functionality in user space.

Signed-off-by: Neil Roberts <neil@linux.intel.com>
Reviewed-by: Brad Volkin <bradley.d.volkin@intel.com> (v1)
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org> (v1)
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: update pipe size at set_config time
Jesse Barnes [Wed, 5 Nov 2014 22:26:10 +0000 (14:26 -0800)]
drm/i915: update pipe size at set_config time

This only affects the fastboot path as-is.  In that case, we simply need
to make sure that we update the pipe size at the first mode set.  Rather
than putting it off until we decide to flip (if indeed we do end up
flipping), update the pipe size as appropriate a bit earlier in the
set_config call.

This sets us up for better pipe tracking in later patches.

Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Reviewed-by: Ander Conselvan de Oliveira <ander.conselvan.de.oliveira@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: check for audio and infoframe changes across mode sets v2
Jesse Barnes [Wed, 5 Nov 2014 22:26:09 +0000 (14:26 -0800)]
drm/i915: check for audio and infoframe changes across mode sets v2

If these change (e.g. after a modeset following a fastboot), we need to
do a full mode set.

v2:
  - put under pipe_config check so we don't deref a null state (Jesse)

Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Reviewed-by: Ander Conselvan de Oliveira <ander.conselvan.de.oliveira@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915/hdmi: fetch infoframe status in get_config v2
Jesse Barnes [Wed, 5 Nov 2014 22:26:08 +0000 (14:26 -0800)]
drm/i915/hdmi: fetch infoframe status in get_config v2

This is useful for checking things later.

v2:
  - fix hsw infoframe enabled check (Ander)

Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Reviewed-by: Ander Conselvan de Oliveira <ander.conselvan.de.oliveira@intel.com>
[danvet: Add the missing PIPE_CONF_CHECK_I(has_infoframe); line to the
hw state cross-checker.]
[danet: Squash in fixup from Jesse to correctly compute has_infoframe
in the hdmi compute_config function.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: use compute_config in set_config v4
Jesse Barnes [Fri, 7 Nov 2014 21:11:00 +0000 (13:11 -0800)]
drm/i915: use compute_config in set_config v4

This will allow us to consult more info before deciding whether to flip
or do a full mode set.

v2:
  - don't use uninitialized or incorrect pipe masks in set_config
    failure path (Ander)
v3:
  - fixup for pipe_config changes in compute_config (Jesse)
v4:
  - drop spurious hunk in force restore path (Ander)

Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Reviewed-by: Ander Conselvan de Oliveira <ander.conselvan.de.oliveira@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: factor out compute_config from __intel_set_mode v3
Jesse Barnes [Wed, 5 Nov 2014 22:26:06 +0000 (14:26 -0800)]
drm/i915: factor out compute_config from __intel_set_mode v3

This allows us to calculate the full pipe config before we do any mode
setting work.

v2:
  - clarify comments about global vs. per-crtc mode set (Ander)
  - clean up unnecessary pipe_config = NULL setting (Ander)
v3:
  - fix pipe_config handling (alloc in compute_config, free in set_mode) (Jesse)
  - fix arg order in set_mode (Jesse)
  - fix failure path of set_config (Ander)

Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Reviewed-by: Ander Conselvan de Oliveira <ander.conselvan.de.oliveira@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Remove most INVALID_PIPE checks from the backlight code
Ville Syrjälä [Fri, 7 Nov 2014 13:20:23 +0000 (15:20 +0200)]
drm/i915: Remove most INVALID_PIPE checks from the backlight code

Now that the backlight device no longer gets registered too early we
should be able to drop most of the INVALID_PIPE checks from the backlight
code.

The only exceptio is the opregion stuff where we may (in theory at
least) get a request from the BIOS already during driver init as soon as
the backlight setup has been done. In which case we can still get the
INVALID_PIPE from intel_get_pipe_from_connector(). So leave that check
in place, and add a comment explaining why.

For the rest, if we still manage to get here with INVALID_PIPE on
VLV/CHV we will now get a WARN from the lower level functions and
can then actually investigate further.

v2: Leave the check in the BIOS related code (Jani)

Cc: Jani Nikula <jani.nikula@intel.com>
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Jani Nikula <jani.nikula@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Register the backlight device after the modeset init
Ville Syrjälä [Fri, 7 Nov 2014 13:19:46 +0000 (15:19 +0200)]
drm/i915: Register the backlight device after the modeset init

Currently we register the backlight device as soon as we register the
connector. That means we can get backlight requests from userspace
already before reading out the current modeset hardware state.

That means we don't yet know the current crtc->encoder->connector mapping,
which causes problems for VLV/CHV which need to know the current pipe in
order to figure out which BLC registers to poke. Currently we just
ignore such requests fairly deep in the backlight code which means the
backlight device brightness property will get out of sync with our
backlight.level and the actual hardware state.

Fix the problem by delaying the backlight device registration until the
entire modeset init has been performed. And we also move the
backlight unregisteration to happen as the first thing during the
modeset cleanup so that we also won't be bothered with userspace
backlight requested during teardown.

This is a real world problem on machines using systemd, because systemd,
for some reason, wants to restore the backlight to the level it used last
time. And that happens as soon as it sees the backlight device appearing
in the system. Sometimes the userspace access makes it through before
the modeset init, sometimes not.

v2: Do not lie to the user in the debug prints (Jani)
    Include connector name in the prints (Jani)
    Fix a typo in the commit message (Jani)

Reviewed-by: Jani Nikula <jani.nikula@intel.com>
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Pass the current pipe from eDP init to backlight setup
Ville Syrjälä [Fri, 7 Nov 2014 09:16:02 +0000 (11:16 +0200)]
drm/i915: Pass the current pipe from eDP init to backlight setup

On VLV/CHV both pipes A and B have their own backlight control
registers. In order to correctly read out the current hardware state at
init we need to know which pipe is driving the eDP port. Pass that
information down from the eDP init code into the backlight code.

To determine the correct pipe we first look at which pipe is currently
configured in the port control register, if that look invalid we look
at which pipe's PPS is currently controlling the port, and if that
too looks invalid we just assume pipe A.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Jani Nikula <jani.nikula@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Don't deref NULL crtc in intel_get_pipe_from_connector()
Ville Syrjälä [Fri, 7 Nov 2014 09:16:01 +0000 (11:16 +0200)]
drm/i915: Don't deref NULL crtc in intel_get_pipe_from_connector()

If the connector would have an encoder but the encoder didn't have a
crtc we might dereference a NULL crtc here. I suppose that should never
happen due to intel_sanitize_encoder(), but let's be a bit paranoid
print a warning if we ever hit this and return INVALID_PIPE to the
caller.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Jani Nikula <jani.nikula@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Skip .get_backlight() when backlight isn't enabled
Ville Syrjälä [Fri, 7 Nov 2014 13:18:45 +0000 (15:18 +0200)]
drm/i915: Skip .get_backlight() when backlight isn't enabled

On VLV/CHV when the display is off, we can't read out the current
backlight level from the hardware since we have no pipe to do so.
Currently we end up reading a bigus register due to passing
INVALID_PIPE to VLV_BLC_PWM_CTL().

Skip the entire .get_backlight() call if the backlight isn't enabled
according to backlight.enabled.

This problem can be reproduced simply by reading the backlight device
actual_brightness file while the display is off.

Cc: Jani Nikula <jani.nikula@intel.com>
Suggested-by: Jani Nikula <jani.nikula@intel.com>
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Jani Nikula <jani.nikula@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Warn if trying to poke a VLV backlight on invalid pipe
Ville Syrjälä [Fri, 7 Nov 2014 09:15:59 +0000 (11:15 +0200)]
drm/i915: Warn if trying to poke a VLV backlight on invalid pipe

VLV/CHV have backlight controls only on pipes A and B. Bail out
without touching registers that don't exist, and print a warning.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Jani Nikula <jani.nikula@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Make the physical object coherent with GTT
Chris Wilson [Tue, 4 Nov 2014 12:51:40 +0000 (04:51 -0800)]
drm/i915: Make the physical object coherent with GTT

Currently objects for which the hardware needs a contiguous physical
address are allocated a shadow backing storage to satisfy the contraint.
This shadow buffer is not wired into the normal obj->pages and so the
physical object is incoherent with accesses via the GPU, GTT and CPU. By
setting up the appropriate scatter-gather table, we can allow userspace
to access the physical object via either a GTT mmaping of or by rendering
into the GEM bo. However, keeping the CPU mmap of the shmemfs backing
storage coherent with the contiguous shadow is not yet possible.
Fortuituously, CPU mmaps of objects requiring physical addresses are not
expected to be coherent anyway.

This allows the physical constraint of the GEM object to be transparent
to userspace and allow it to efficiently render into or update them via
the GTT and GPU.

v2: Fix leak of pci handle spotted by Ville
v3: Remove the now duplicate call to detach_phys_object during free.
v4: Wait for rendering before pwrite. As this patch makes it possible to
render into the phys object, we should make it correct as well!

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: WARN if we receive any gen9 rps interrupts
Imre Deak [Mon, 10 Nov 2014 13:34:33 +0000 (15:34 +0200)]
drm/i915: WARN if we receive any gen9 rps interrupts

Paulo noticed that we don't support RPS on GEN9 yet, so WARN for and
ignore any RPS interrupts on that platform.

Signed-off-by: Imre Deak <imre.deak@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: move rps irq enable/disable to i915_irq.c
Imre Deak [Wed, 5 Nov 2014 18:48:48 +0000 (20:48 +0200)]
drm/i915: move rps irq enable/disable to i915_irq.c

The logical place for these functions is in i915_irq.c next to the rest of
PM interrupt handling functions.

No functional change.

Signed-off-by: Imre Deak <imre.deak@intel.com>
Reviewed-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: unify gen6/gen8 rps irq enable/disable
Imre Deak [Wed, 5 Nov 2014 18:48:42 +0000 (20:48 +0200)]
drm/i915: unify gen6/gen8 rps irq enable/disable

The GEN6 and GEN8 versions differ only in the PM IIR and IER register
addresses and that on GEN8 we need to keep the
GEN8_PMINTR_REDIRECT_TO_NON_DISP PM interrupt unmasked. Abstract away
these 3 things in the GEN6 versions of the helpers and use them
everywhere.

No functional change.

Signed-off-by: Imre Deak <imre.deak@intel.com>
Reviewed-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: unify gen6/gen8 rps irq handler
Imre Deak [Wed, 5 Nov 2014 18:48:37 +0000 (20:48 +0200)]
drm/i915: unify gen6/gen8 rps irq handler

After the previous patch the GEN8 RPS handler became very similar to the
GEN6 version, so unify the two functions.

No functional change.

Signed-off-by: Imre Deak <imre.deak@intel.com>
Reviewed-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
[danvet: Move one misplaced hunk from a later patch to fix a bisect
issue as reported by Wu Fengguang's 0-day builder and fix suggested by
Imre.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: unify gen6/gen8 pm irq helpers
Imre Deak [Wed, 5 Nov 2014 18:48:31 +0000 (20:48 +0200)]
drm/i915: unify gen6/gen8 pm irq helpers

The helpers to enable/disable PM IRQs for GEN6 and GEN8 are the same
except for the PM interrupt mask register, so abstract away this
register in the GEN6 versions and use these everywhere.

No functional change.

Signed-off-by: Imre Deak <imre.deak@intel.com>
Reviewed-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915/chv: Add new workarounds for chv
Arun Siluvery [Tue, 28 Oct 2014 18:33:14 +0000 (18:33 +0000)]
drm/i915/chv: Add new workarounds for chv

+WaForceEnableNonCoherent:chv
+WaHdcDisableFetchWhenMasked:chv

For: VIZ-4090
Signed-off-by: Arun Siluvery <arun.siluvery@linux.intel.com>
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915/chv: Combine GEN8_ROW_CHICKEN w/a
Arun Siluvery [Tue, 28 Oct 2014 18:33:13 +0000 (18:33 +0000)]
drm/i915/chv: Combine GEN8_ROW_CHICKEN w/a

WaDisablePartialInstShootdown:chv and
WaDisableThreadStallDopClockGating:chv are related to the same
register so combine them.

Signed-off-by: Arun Siluvery <arun.siluvery@linux.intel.com>
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915/chv: Remove pre-production workarounds
Arun Siluvery [Tue, 28 Oct 2014 18:33:12 +0000 (18:33 +0000)]
drm/i915/chv: Remove pre-production workarounds

-WaDisableDopClockGating:chv
-WaDisableSamplerPowerBypass:chv
-WaDisableGunitClockGating:chv
-WaDisableFfDopClockGating:chv
-WaDisableDopClockGating:chv

v2: Remove pre-production WA instead of restricting them
based on revision id (Ville)

For: VIZ-4090
Signed-off-by: Arun Siluvery <arun.siluvery@linux.intel.com>
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: use the correct obj when preparing the sprite plane
Paulo Zanoni [Mon, 10 Nov 2014 16:47:30 +0000 (14:47 -0200)]
drm/i915: use the correct obj when preparing the sprite plane

Commit "drm/i915: create a prepare phase for sprite plane updates"
changed the old_obj pointer we use when committing sprite planes,
which caused a WARN() and a BUG() to be triggered. Later, commit
"drm/i915: use intel_fb_obj() macros to assign gem objects" introduced
the same problem to function intel_commit_sprite_plane().

Regression introduced by:
    commit ec82cb793c9224e0692eed904f43490cf70e8258
    Author: Gustavo Padovan <gustavo.padovan@collabora.co.uk>
    Date:   Fri Oct 24 14:51:32 2014 +0100
        drm/i915: create a prepare phase for sprite plane updates
and:
    commit 77cde95217484e845743818691df026cec2534f4
    Author: Gustavo Padovan <gustavo.padovan@collabora.co.uk>
    Date:   Fri Oct 24 14:51:33 2014 +0100
        drm/i915: use intel_fb_obj() macros to assign gem objects

Credits to Imre Deak for pointing out the exact lines that were wrong.

v2: Also fix intel_commit_sprite_plane() (Ville)

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=85634
Testcase: igt/pm_rpm/legacy-planes
Testcase: igt/pm_rpm/legacy-planes-dpms
Testcase: igt/pm_rpm/universal-planes
Testcase: igt/pm_rpm/universal-planes-dpms
Credits-to: Imre Deak <imre.deak@intel.com>
Cc: Gustavo Padovan <gustavo.padovan@collabora.co.uk>
Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Add tracepoints to track a vm during its lifetime
Daniele Ceraolo Spurio [Mon, 10 Nov 2014 13:44:31 +0000 (13:44 +0000)]
drm/i915: Add tracepoints to track a vm during its lifetime

- ppgtt init/release: these tracepoints are useful for observing the
  creation and destruction of Full PPGTTs.

- ctx create/free: we can use the ctx_free trace in combination with the
  ppgtt_release one to be sure that the ppgtt doesn't stay alive for too
  long after the ctx is destroyed. ctx_create is there for simmetry

- switch_mm: important point in the lifetime of the vm

v4: add DOC information
v5: pull the DOC in drm.tmpl
v6: clean ppgtt init/release traces + add ctx create/free and switch_mm
    tracepoints (Chris)
v7: drop execlist_submit_context tracepoint

Cc: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/edid: fix Baseline_ELD_Len field in drm_edid_to_eld()
Jani Nikula [Tue, 28 Oct 2014 14:20:48 +0000 (16:20 +0200)]
drm/edid: fix Baseline_ELD_Len field in drm_edid_to_eld()

The Baseline_ELD_Len field does not include ELD Header Block size.

From High Definition Audio Specification, Revision 1.0a:

The header block is a fixed size of 4 bytes. The baseline block
is variable size in multiple of 4 bytes, and its size is defined
in the header block Baseline_ELD_Len field (in number of
DWords).

Do not include the header size in Baseline_ELD_Len field. Fix all known
users of eld[2].

While at it, switch to DIV_ROUND_UP instead of open coding it.

Signed-off-by: Jani Nikula <jani.nikula@intel.com>
Acked-by: Ben Skeggs <bskeggs@redhat.com>
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Acked-by: Dave Airlie <airlied@linux.ie>
[danvet: Fix compile fail in nouveau.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: avoid deadlock on failure paths in __intel_framebuffer_create()
Alexey Khoroshilov [Fri, 7 Nov 2014 22:41:23 +0000 (01:41 +0300)]
drm/i915: avoid deadlock on failure paths in __intel_framebuffer_create()

Since a8bb6818270c __intel_framebuffer_create() is called
with struct_mutex held, so it should use drm_gem_object_unreference()
instead of drm_gem_object_unreference_unlocked().

Found by Linux Driver Verification project (linuxtesting.org).

This regression has been introduced in

commit a8bb6818270c32126dba0fd2ddb139d885c5687d
Author: Daniel Vetter <daniel.vetter@ffwll.ch>
Date:   Mon Feb 10 18:00:39 2014 +0100

    drm/i915: Fix error path leak in fbdev fb allocation

Signed-off-by: Alexey Khoroshilov <khoroshilov@ispras.ru>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Use correct pipe config to update pll dividers. V2
Bob Paauwe [Tue, 11 Nov 2014 17:29:18 +0000 (09:29 -0800)]
drm/i915: Use correct pipe config to update pll dividers. V2

Use the new pipe config values to calculate the updated pll dividers.

This regression was introduced in

commit 0dbdf89f27b17ae1eceed6782c2917f74cbb5d59
Author: Ander Conselvan de Oliveira <ander.conselvan.de.oliveira@intel.com>
Date:   Wed Oct 29 11:32:33 2014 +0200

    drm/i915: Add infrastructure for choosing DPLLs before disabling crtcs

and

commit 00d958817dd3daaa452c221387ddaf23d1e4c06f
Author: Ander Conselvan de Oliveira <ander.conselvan.de.oliveira@intel.com>
Date:   Wed Oct 29 11:32:36 2014 +0200

    drm/i915: Covert remaining platforms to choose DPLLS before disabling CRTCs

v2: Use intel_pipe_will_have_type() to look at new configuration - Ander

Signed-off-by: Bob Paauwe <bob.j.paauwe@intel.com>
CC: Ander Conselvan de Oliveira <ander.conselvan.de.oliveira@intel.com>
Tested-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Reviewed-by: Ander Conselvan de Oliveira <ander.conselvan.de.oliveira@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Plug memory leak in intel_shared_dpll_start_config()
Ander Conselvan de Oliveira [Fri, 7 Nov 2014 12:07:41 +0000 (14:07 +0200)]
drm/i915: Plug memory leak in intel_shared_dpll_start_config()

The cleanup path would reset pll->new_config to NULL but wouldn't free
the allocated memory.

Signed-off-by: Ander Conselvan de Oliveira <ander.conselvan.de.oliveira@intel.com>
Reviewed-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agoMerge remote-tracking branch 'airlied/drm-next' into HEAD
Daniel Vetter [Mon, 10 Nov 2014 09:55:35 +0000 (10:55 +0100)]
Merge remote-tracking branch 'airlied/drm-next' into HEAD

Backmerge drm-next so that I can keep merging patches. Specifically I
want:
- atomic stuff, yay!
- eld parsing patch from Jani.

Signed-off-by: Daniel Vetter <daniel.vetter@intel.com>
10 years agodrm/mode: document path property and function to set it. (v1.1)
Dave Airlie [Wed, 22 Oct 2014 02:03:04 +0000 (12:03 +1000)]
drm/mode: document path property and function to set it. (v1.1)

These two didn't get documented properly, do so.

Pointed out by Daniel.

v1.1: add missing boilerplate (Daniel)

Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Dave Airlie <airlied@redhat.com>
10 years agoMerge tag 'topic/atomic-helpers-2014-11-09' of git://anongit.freedesktop.org/drm...
Dave Airlie [Sun, 9 Nov 2014 23:59:16 +0000 (09:59 +1000)]
Merge tag 'topic/atomic-helpers-2014-11-09' of git://anongit.freedesktop.org/drm-intel into drm-next

So here's my atomic series, finally all debugged&reviewed. Sean Paul has
done a full detailed pass over it all, and a lot of other people have
commented and provided feedback on some parts. Rob Clark also converted
msm over the w/e and seems happy. The only small thing is that Rob wants
to export the wait_for_vblank, which imo makes sense. Since there's other
stuff still to do I think we should apply Rob's patch (once it has grown
appropriate kerneldoc) later on top of this.

This is just the core<->driver interface plus a big pile of helpers. Short
recap of the main ideas:

- There are essentially three helper libraries in this patch set:

  * Transitional helpers to use the new plane callbacks for legacy plane
    updates and in the crtc helper's ->mode_set callback. These helpers are
    only temporarily used to convert drivers to atomic, but they allow a
    nice separation between changing the driver backend and switching to
    the atomic commit logic.

  * Legacy helpers to implement all the legacy driver entry points
    (page_flip, set_config, plane vfuncs) on top of the new atomic driver
    interface. These are completely driver agnostic. The reason for having
    the legacy support as helpers is that drivers can switch step-by-step.
    And they could e.g. even keep the legacy page_flip code around for some
    old platforms where converting to full-blown atomic isn't worth it.

  * Atomic helpers which implement the various new ->atomic_* driver
    interfaces in terms of the revised crtc helper and new plane helper
    hooks.

- The revised crtc helper implemenation essentially implements all the
  lessons learned in the i915 modeset rework (when using the atomic helpers
  only):

  * Enable/disable sequence for a given config are always the same and
    callbacks are always called in the same order. This contrast starkly
    with the crtc helpers, where the sequence of operations is heavily
    dependent on the previous config.

    One corollary of this is that if the configuration of a crtc only
    partially changes (e.g. a connector moves in a cloned config) the
    helper code will still disable/enable the full display pipeline. This
    is the only way to ensure that the enable/disable sequence is always
    the same.

  * It won't call disable or enable hooks more than once any more because
    it lost track of state, thanks to the atomic state tracking. And if
    drivers implement the ->reset hook properly (by either resetting the hw
    or reading out the hw state into the atomic structures) this even
    extends to the hardware state. So no more disable-me-harder kind of
    nonsense.

  * The only thing missing is the hw state readout/cross-check support, but
    if drivers have hw state readout support in their ->reset handlers it's
    simple to extend that to cross-check the hw state.

  * The crtc->mode_set callback is gone and its replacement only sets crtc
    timings and no longer updates the primary plane state. This way we can
    finally implement primary planes properly.

- The new plane helpers should be suitable enough for pretty much
  everything, and a perfect fit for hardware with GO bits. Even if they
  don't fit the atomic helper library is rather flexible and exports all
  the functions for the individual steps to drivers. So drivers can pick
  what matches and implement their own magic for everything else.

- A big difference compared to all previous atomic series is that this one
  doesn't implement async commit in a generic way. Imo driver requirements
  for that are too diverse to create anything reasonable sane which would
  actually work on a reasonable amount of different drivers. Also, we've
  never had a helper library for page_flips even, so it's really hard to
  know what might work and what's stupid without a bit of experience in the form
  of a few driver implementations.

  I think with the current flexibility for drivers to pick individual
  stages and existing helpers like drm_flip_queue it's rather easy though
  to implement proper async commit.

- There's a few other differences of minor importance to earlier atomic
  series:

  * Common/generic properties are parsed in the callers/core and not in
    drivers, and passed to drivers by directly setting the right members in
    atomic state structures. That greatly simplifies all the transitional
    and legacy helpers an removes a lot of boilerplate code.

  * There's no crazy trylock mode used for the async commit since these
    helpers don't do async commit. A simple ordered flip queue of atomic
    state updates should be sufficient for preventing concurrent hw access
    anyway, as long as synchronous updates stall correctly with e.g.
    flush_work_queue or similar function. Abusing locks to enforce ordering
    isn't a good idea imo anyway.

  * These helpers reuse the existing ->mode_fixup hooks in the atomic_check
    callback. Which means that drivers need to adapat and move a lot less code
    into their atomic_check callbacks.

Now this isn't everything needed in the drm core and helpers for full
atomic support. But it's enough to start with converting drivers, and
except for actually testing multiplane and multicrtc updates also enough to
implement full atomic updates. Still missing are:

- Per-plane locking. Since these helpers here encapsulate the locking
  completely this should be fairly easy to implement.

- fbdev support for atomic_check/commit, so that multi-pipe finally works
  sanely in fbcon.

- Adding and decoding shared/core properties. That just needs to be rebased
  from Rob's latest patch series, with minor adjustments so that the
  decoding happens in the core instead of in drivers.

- Actually adding the atomic ioctl. Again just rebasing Rob's latest patch
  should be all that's needed.

- Resolving how to deal with DPMS in atomic. Atomic is a good excuse to fix up
  the crazy semantics dpms currently has. I'm floating an RFC about this topic
  already.

- Finally I couldn't test connector/encoder stealing properly since my test
  vehicle here doesn't allow a connector on different crtcs. So drivers
  which support this might see some surprises in that area. There is no semantic
  change though in how encoder stealing and assignment works (or at least no
  intended one), so I think the risk is minimal.

As just mentioned I've done a fake conversion of an existing driver using
crtc helpers to debug the helper code and validate the smooth transition
approach. And that smooth transition was the really big motivation for
this. It seems to actually work and consists of 3 phases:

Phase 1: Rework driver backend for crtc/plane helpers

The requirement here is that universal plane support is already implement. If
universal plane support isn't implement yet it might be better though to just do
it as part of this phase, directly using the new plane helpers. There are two
big things to do:

- Split up the existing ->update/disable_plane hooks into check/commit
  hooks and extract the crtc-wide prep/flush parts (like setting/clearing
  GO bits).

- The other big change is to split the crtc->mode_set hook into the plane
  update (done using the plane helpers) and the crtc setup in a new
  ->mode_set_nofb hook.

When phase 1 is complete the driver implements all the new callbacks which
push the software state into hardware, but still using all the legacy entry
points and crtc helpers. The transitional helpers serve as impendance
mismatch here.

Phase 2: Rework state handling

This consists of rolling out the state handling helpers for planes, crtcs
and connectors and reviewing all ->mode_fixup and similar hooks to make
sure they don't depend upon implicit global state which might change in the
atomic world. Any such code must be moved into ->atomic_check functions which
just rely on the free-standing atomic state update structures.

This phase also adds a few small pieces of fixup code to make sure the
atomic state doesn't get out of sync in the legacy driver callbacks.

Phase 3: Roll out atomic support

Now it's just about replacing vfuncs with the ones provided by the helper
and filling out the small missing pieces (like atomic_check logic or async
commit support needed for page_flips). Due to the prep work in phase 1 no
changes to the driver backend functions should be required, and because of
the prep work in phase 2 atomic implementations can be rolled out
step-by-step. So if async commit ins't implemented yet page_flip can be
implemented with the legacy functions without wreaking havoc in the other
operations.

* tag 'topic/atomic-helpers-2014-11-09' of git://anongit.freedesktop.org/drm-intel:
  drm/atomic: Refcounting for plane_state->fb
  drm: Docbook integration and over sections for all the new helpers
  drm/atomic-helpers: functions for state duplicate/destroy/reset
  drm/atomic-helper: implement ->page_flip
  drm/atomic-helpers: document how to implement async commit
  drm/atomic: Integrate fence support
  drm/atomic-helper: implementatations for legacy interfaces
  drm: Atomic crtc/connector updates using crtc/plane helper interfaces
  drm/crtc-helper: Transitional functions using atomic plane helpers
  drm/plane-helper: transitional atomic plane helpers
  drm: Add atomic/plane helpers
  drm: Global atomic state handling
  drm: Add atomic driver interface definitions for objects
  drm/modeset_lock: document trylock_only in kerneldoc
  drm: fixup kerneldoc in drm_crtc.h
  drm: Pull drm_crtc.h into the kerneldoc template
  drm: Move drm_crtc_init from drm_crtc.h to drm_plane_helper.h

10 years agodrm/i915: Update DRIVER_DATE to 20141107
Daniel Vetter [Fri, 7 Nov 2014 18:03:19 +0000 (19:03 +0100)]
drm/i915: Update DRIVER_DATE to 20141107

Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Add gen to the gpu hang ecode
Mika Kuoppala [Thu, 6 Nov 2014 11:03:46 +0000 (13:03 +0200)]
drm/i915: Add gen to the gpu hang ecode

for the Brothers in Triage

Signed-off-by: Mika Kuoppala <mika.kuoppala@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Cache HPLL frequency on VLV/CHV
Ville Syrjälä [Tue, 7 Oct 2014 14:41:22 +0000 (17:41 +0300)]
drm/i915: Cache HPLL frequency on VLV/CHV

We need the HPLL frequency when calculating cdclk. Currently we read
that out from the hardware every single time, which isn't going to fly
very well if the device is runtime suspended. So cache the HPLL
frequency in dev_priv and use the cached value.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reference: https://bugs.freedesktop.org/show_bug.cgi?id=82939
Reviewed-by: Imre Deak <imre.deak@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agoRevert "drm/i915/vlv: Remove check for Old Ack during forcewake"
Mika Kuoppala [Wed, 5 Nov 2014 15:30:52 +0000 (17:30 +0200)]
Revert "drm/i915/vlv: Remove check for Old Ack during forcewake"

This reverts commit 5cb13c07dae73380d8b3ddc792740487b8742938.

While the relevance for WaRsDontPollForAckOnClearingFWBits is under
investigation, revert this as regression.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=85684
Tested-by: Tested-by: lu hua <huax.lu@intel.com>
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
Cc: S, Deepak <deepak.s@intel.com>
Signed-off-by: Mika Kuoppala <mika.kuoppala@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Make mmio flip wait for seqno in the work function
Ander Conselvan de Oliveira [Thu, 6 Nov 2014 09:03:40 +0000 (11:03 +0200)]
drm/i915: Make mmio flip wait for seqno in the work function

This simplifies the code quite a bit compared to iterating over all
rings during the ring interrupt.

Also, it allows us to drop the mmio_flip spinlock, since the mmio_flip
struct is only accessed in two places. The first is when the flip is
queued and the other when the mmio writes are done. Since a flip cannot
be queued while there is a pending flip, the two paths shouldn't ever
run in parallel. We might need to revisit that if support for replacing
flips is implemented though.

v2: Don't hold dev->struct_mutext while waiting (Chris)

v3: Make the wait uninterruptable (Chris)

Signed-off-by: Ander Conselvan de Oliveira <ander.conselvan.de.oliveira@intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Make __wait_seqno non-static and rename to __i915_wait_seqno
Ander Conselvan de Oliveira [Thu, 6 Nov 2014 07:26:38 +0000 (09:26 +0200)]
drm/i915: Make __wait_seqno non-static and rename to __i915_wait_seqno

So that it can be used by the flip code to wait for rendering without
holding any locks.

Signed-off-by: Ander Conselvan de Oliveira <ander.conselvan.de.oliveira@intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Move the .global_resources() hook call into modeset_update_crtc_power_domains()
Ville Syrjälä [Thu, 6 Nov 2014 12:49:12 +0000 (14:49 +0200)]
drm/i915: Move the .global_resources() hook call into modeset_update_crtc_power_domains()

We may need to access various hardware bits in the .global_resources()
hook, so move the call to occur after enabling all the newly required
power wells, but before disabling all the now unneeded wells. This
should guarantee that we have all the sufficient hardware resources
available during the .global_resources() call. And if not, any additional
resources must be explicitly acquired by the .global_resorces() hook.

For instance on VLV/CHV we need to access the gunit mailbox so that we
can talk to punit/cck over sideband. In addition some PFI credit
reprogramming may need to be addes as well, which may require the disp2d
well.

This should also make the power domain refcounts consistent on platforms
which don't have a .global_resource() hook since now they too will
call modeset_update_crtc_power_domains() which will drop the init power.
Previously init power was just left enabled for such platforms.

Cc: Imre Deak <imre.deak@intel.com>
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915/audio: add DOC comment describing HDA over HDMI/DP
Jani Nikula [Mon, 27 Oct 2014 14:27:00 +0000 (16:27 +0200)]
drm/i915/audio: add DOC comment describing HDA over HDMI/DP

v2: include the section in the drm docbook.

Signed-off-by: Jani Nikula <jani.nikula@intel.com>
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: make pipe/port based audio valid accessors easier to use
Jani Nikula [Mon, 27 Oct 2014 14:26:59 +0000 (16:26 +0200)]
drm/i915: make pipe/port based audio valid accessors easier to use

Signed-off-by: Jani Nikula <jani.nikula@intel.com>
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915/audio: add audio codec enable debug log for g4x
Jani Nikula [Mon, 27 Oct 2014 14:26:58 +0000 (16:26 +0200)]
drm/i915/audio: add audio codec enable debug log for g4x

Signed-off-by: Jani Nikula <jani.nikula@intel.com>
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915/audio: add audio codec disable on g4x
Jani Nikula [Mon, 27 Oct 2014 14:26:57 +0000 (16:26 +0200)]
drm/i915/audio: add audio codec disable on g4x

This not based on any documentation...

Signed-off-by: Jani Nikula <jani.nikula@intel.com>
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: enable audio codec after port
Jani Nikula [Mon, 27 Oct 2014 14:26:56 +0000 (16:26 +0200)]
drm/i915: enable audio codec after port

As per spec, and similar to DDI.

Signed-off-by: Jani Nikula <jani.nikula@intel.com>
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915/audio: add vlv/chv/gen5-7 audio codec disable sequence
Jani Nikula [Mon, 27 Oct 2014 14:26:55 +0000 (16:26 +0200)]
drm/i915/audio: add vlv/chv/gen5-7 audio codec disable sequence

Add support for disabling the audio codec on vlv/chv/gen5-7, similar to
hsw/bdw.

Signed-off-by: Jani Nikula <jani.nikula@intel.com>
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915/audio: rewrite vlv/chv and gen 5-7 audio codec enable sequence
Jani Nikula [Tue, 4 Nov 2014 08:31:28 +0000 (10:31 +0200)]
drm/i915/audio: rewrite vlv/chv and gen 5-7 audio codec enable sequence

Similar to the hsw/bdw enable sequence rewrite.

v3: replace vblank wait with a comment

v4: expand the comment on what should be done with the vblank wait

Signed-off-by: Jani Nikula <jani.nikula@intel.com>
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915/skl: Enable Gen9 RC6
Zhe Wang [Tue, 4 Nov 2014 17:07:05 +0000 (17:07 +0000)]
drm/i915/skl: Enable Gen9 RC6

Configure and enable RC6 for Gen9.

v2: Rebase on top of BDW rc6 support (Damien)

Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com>
Signed-off-by: Zhe Wang <zhe1.wang@intel.com>
Signed-off-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915/skl: Gen9 Forcewake
Zhe Wang [Tue, 4 Nov 2014 17:07:04 +0000 (17:07 +0000)]
drm/i915/skl: Gen9 Forcewake

Implement common forcewake functions shared by Gen9 features.

v2: Make the focewake_{get,put} functions static (Mika)
    Small coding style fix in the function definition (Damien)

Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com>
Signed-off-by: Zhe Wang <zhe1.wang@intel.com> (v1)
Signed-off-by: Damien Lespiau <damien.lespiau@intel.com> (v2)
Signed-off-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915/skl: Log the order in which we flush the pipes in the WM code
Damien Lespiau [Tue, 4 Nov 2014 17:07:03 +0000 (17:07 +0000)]
drm/i915/skl: Log the order in which we flush the pipes in the WM code

Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915/skl: Flush the WM configuration
Damien Lespiau [Tue, 4 Nov 2014 17:07:02 +0000 (17:07 +0000)]
drm/i915/skl: Flush the WM configuration

When we write new values for the DDB allocation and WM parameters, we now
need to trigger the double buffer update for the pipe to take the new
configuration into account.

As the DDB is a global resource shared between planes, enabling or
disabling one plane will result in changes for all planes that are
currently in use, thus the need write PLANE_SURF/CUR_BASE for more than
the plane we're touching.

v2: Don't wait for pipes that are off

v3: Split the staging results structure to not exceed the 1Kb stack
    allocation in skl_update_wm()

v4: Rework and document the algorithm after Ville found that it was all
    wrong.

Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915/skl: Stage the pipe DDB allocation
Damien Lespiau [Tue, 4 Nov 2014 17:07:01 +0000 (17:07 +0000)]
drm/i915/skl: Stage the pipe DDB allocation

To correctly flush the new DDB allocation we need to know about the pipe
allocation layout inside the DDB in order to sequence the re-allocation
to not cause a newly allocated pipe to fetch from a space that was
previously allocated to another pipe.

This patch preserves the per-pipe (start,end) allocation to be used in
the flush.

Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915/skl: Reduce the indentation level in skl_write_wm_values()
Damien Lespiau [Tue, 4 Nov 2014 17:07:00 +0000 (17:07 +0000)]
drm/i915/skl: Reduce the indentation level in skl_write_wm_values()

We can reduce the indentation level by continuing early.

Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915/skl: Correctly align skl_compute_plane_wm() arguments
Damien Lespiau [Tue, 4 Nov 2014 17:06:59 +0000 (17:06 +0000)]
drm/i915/skl: Correctly align skl_compute_plane_wm() arguments

Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915/skl: Rework when the transition WMs are computed
Damien Lespiau [Tue, 4 Nov 2014 17:06:58 +0000 (17:06 +0000)]
drm/i915/skl: Rework when the transition WMs are computed

The transition WMs code was doing a shortcut and the values were copied
from the WM0 ones at compute_wm_results() time. Going forward, we want
to compute them like the other WMs and resolve their final register
values in the same way as well.

This patch does just that and isolate the transtion WM compute code in
skl_compute_transition_wm() while skl_compute_wm_results() takes care of
the register values.

We also take the opportunity to disable the transition WMs for now.
We've noticed underruns and they seem to be the culprit.

Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915/skl: Move all the WM compute functions in one place
Damien Lespiau [Tue, 4 Nov 2014 17:06:57 +0000 (17:06 +0000)]
drm/i915/skl: Move all the WM compute functions in one place

The DDB allocation code managed to split in two the compute functions.

Bring back skl_compute_transition_wm() and skl_compute_linetime_wm()
with their little friends.

Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915/skl: Reduce the number of holes in struct skl_wm_level
Damien Lespiau [Tue, 4 Nov 2014 17:06:56 +0000 (17:06 +0000)]
drm/i915/skl: Reduce the number of holes in struct skl_wm_level

Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915/skl: Make res_blocks/lines intermediate values 32 bits
Damien Lespiau [Tue, 4 Nov 2014 17:06:55 +0000 (17:06 +0000)]
drm/i915/skl: Make res_blocks/lines intermediate values 32 bits

To align with the ilk WM code and because it makes sense to test against
the upper bounds as soon as possible on variables that are bigger than
the number of bits in the register, let's move the maximum checks from
skl_compute_wm_results() to skl_compute_plane_wm().

v2: Leave the result values to 0 when overflowing the limits (Ville)
    Use 32 bits intermediate variables (Damien)

Instead of using the 16 and 8 bits space we have in the result
structure, use 32 bits local variables until we're sure they fit into
the constraints.

Suggested-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915/skl: Use a more descriptive parameter name in skl_compute_plane_wm()
Damien Lespiau [Tue, 4 Nov 2014 17:06:54 +0000 (17:06 +0000)]
drm/i915/skl: Use a more descriptive parameter name in skl_compute_plane_wm()

What we're talking about here is the DDB allocation (in blocks). That's
more descriptive than 'max_page_buff_alloc'.

Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915/skl: Make 'end' of the DDB allocation entry exclusive
Damien Lespiau [Tue, 4 Nov 2014 17:06:53 +0000 (17:06 +0000)]
drm/i915/skl: Make 'end' of the DDB allocation entry exclusive

Ville suggested that we should use the same semantics as C arrays to
reduce the number of those pesky +1/-1 in the allocation code.

This patch leaves the debugfs file as is, showing the internal DDB
allocation structure, not the values written in the registers.

v2: Remove the test on ->end in skl_ddb_entry_size() (Ville)

Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Suggested-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915/skl: Check the DDB state at modeset
Damien Lespiau [Tue, 4 Nov 2014 17:06:52 +0000 (17:06 +0000)]
drm/i915/skl: Check the DDB state at modeset

v2: Don't check DDB on pre-SKL platforms
    Don't check DDB state on disabled pipes

v3: Squash "Expose skl_ddb_get_hw_state()"

Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915/skl: Add a debugfs file to dump the DDB allocation
Damien Lespiau [Tue, 4 Nov 2014 17:06:51 +0000 (17:06 +0000)]
drm/i915/skl: Add a debugfs file to dump the DDB allocation

v2: minor conflict in i915_debugfs.c
v3: Rebase on top of the for_each_pipe() change adding dev_priv as first
    argument.
v4: minor conflict in the i915_debugfs_files array
v5: minor conflict in the i915_debugfs_files array

Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915/skl: Augment the latency debugfs files for SKL
Damien Lespiau [Tue, 4 Nov 2014 17:06:50 +0000 (17:06 +0000)]
drm/i915/skl: Augment the latency debugfs files for SKL

v2: Use the gen >= 9 in the debugfs file condition (Ville)

Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915/skl: Read back the DDB allocation hw state
Damien Lespiau [Tue, 4 Nov 2014 17:06:49 +0000 (17:06 +0000)]
drm/i915/skl: Read back the DDB allocation hw state

This logically belongs to the WM state, so do it there.

Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915/skl: Store the new WM state at the very end of the update
Damien Lespiau [Tue, 4 Nov 2014 17:06:48 +0000 (17:06 +0000)]
drm/i915/skl: Store the new WM state at the very end of the update

We're going to add a new step, let's not hide the copy of the new WM
state inside one inner function, but as a 1st level operation in the WM
update.

v2: Split the staging results structure to not exceed the 1Kb stack
    allocation in skl_update_wm()

Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915/gen9: Disable WM if corresponding latency is 0
Vandana Kannan [Tue, 4 Nov 2014 17:06:47 +0000 (17:06 +0000)]
drm/i915/gen9: Disable WM if corresponding latency is 0

According to updated BSpec, If level 1 or any higher level has a value of 0x00,
that level and any higher levels are unused and the associated watermark
registers must not be enabled.

This patch checks for latency 0 for level >=1 and does not enable WM
corresponding to level m | m>=n, if level n (n != 0) has a 0us latency.

v2: Satheesh's review comments
- zero-out latency values (for all higher levels if latency of given
level is zero ) in read_wm_latency() function itself

v3: removed redundant check as per Satheesh's observation.
v4: rebase on top before merging (Damien)
v5: Rebase on top of the default value removal (Ville)

Reviewed-by: Satheeshakrishna M <satheeshakrishna.m@intel.com> (v3)
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Vandana Kannan <vandana.kannan@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915/gen9: Add 2us read latency to WM level
Vandana Kannan [Tue, 4 Nov 2014 17:06:46 +0000 (17:06 +0000)]
drm/i915/gen9: Add 2us read latency to WM level

According to the updated Bspec, The mailbox response data is not currently
accounting for memory read latency. Add 2 microseconds to the result for
each level.
This patch adds 2us to latency of level 0 for all cases and
for all other levels (1-7) only if latency[level] > 0.

v2: Slightly rework the patch and add a big comment (Damien)
v3: Rebase on top of the renames of the memory latency defines

Signed-off-by: Vandana Kannan <vandana.kannan@intel.com> (v1)
Signed-off-by: Damien Lespiau <damien.lespiau@intel.com>
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com> (v2)
Reviewed-by: M, Satheeshakrishna <satheeshakrishna.m@intel.com> (v1)
Cc: Lespiau, Damien <damien.lespiau@intel.com>
Cc: M, Satheeshakrishna <satheeshakrishna.m@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915/skl: Read the pipe WM HW state
Pradeep Bhat [Tue, 4 Nov 2014 17:06:45 +0000 (17:06 +0000)]
drm/i915/skl: Read the pipe WM HW state

This patch provides the implementation for reading the pipe wm HW
state.

v2: Incorporated Damien's review comments and also made modifications
    to incorporate the plane/cursor split.

v3: No need to ident a line that was fitting 80 chars
    Return early instead of indenting the remaining of a function
    (Damien)

v4: Rebase on top of nightly (minor conflict in intel_drv.h)
v5: Rebase on top of nightly (minor conflict in intel_drv.h)
v6: Rebase on top of nightly (minor conflict in intel_drv.h)

Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Pradeep Bhat <pradeep.bhat@intel.com>
Signed-off-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915/skl: Program the DDB allocation
Damien Lespiau [Tue, 4 Nov 2014 17:06:44 +0000 (17:06 +0000)]
drm/i915/skl: Program the DDB allocation

v2: Adapt to the planes/cursor split

Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915/skl: Allocate DDB portions for display planes
Damien Lespiau [Tue, 4 Nov 2014 17:06:43 +0000 (17:06 +0000)]
drm/i915/skl: Allocate DDB portions for display planes

v2: Fix the 3rd plane/cursor logic (Pradeep Bhat)
v3: Fix one-by-one error in the DDB allocation code
v4: Rebase on top of the skl_pipe_pixel_rate() argument change
v5: Replace the available/start/end output parameters of
    skl_ddb_get_pipe_allocation_limits() by a single ddb entry constify
    a few arguments
    Make nth_active_pipe 0 indexed
    Use sizeof(variable) instead of sizeof(type)
    (Ville)
v6: Use the for_each_crtc() macro instead of list_for_each_entry()

Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915/skl: SKL Watermark Computation
Pradeep Bhat [Tue, 4 Nov 2014 17:06:42 +0000 (17:06 +0000)]
drm/i915/skl: SKL Watermark Computation

This patch implements the watermark algorithm and its necessary
functions. Two function pointers skl_update_wm and
skl_update_sprite_wm are provided. The skl_update_wm will update
the watermarks for the crtc provided as an argument and then
checks for change in DDB allocation for other active pipes and
recomputes the watermarks for those Pipes and planes as well.
Finally it does the register programming for all dirty pipes.
The trigger of the Watermark double buffer registers will have
to be once the plane configurations are done by the caller.

v2: fixed the divide-by-0 error in the results computation func.
    Also reworked the PLANE_WM register values computation func to
    make it more compact. Incorporated all other review comments
    from Damien.

v3: Changed the skl_compute_plane_wm function to now return success
    or failure. Also the result blocks and lines are computed here
    instead of in skl_compute_wm_results function.

v4: Adjust skl_ddb_alloc_changed() to the new planes/cursor split
    (Damien)

v5: Reworked the affected functions to implement new plane/cursor
    split.

v6: Rework the logic that triggers the DDB allocation and WM computation
    of skl_update_other_pipe_wm() to not depend on non-computed DDB
    values.
    Always give a valid cursor_width (at boot it's 0) to keep the
    invariant that we consider the cursor plane always enabled.
    Otherwise we end up dividing by 0 in skl_compute_plane_wm()
    (Damien Lespiau)

v7: Spell out allocation
    skl_ddb_ functions should have the ddb as first argument
    Make the skl_ddb_alloc_changed() parameters const
    (Damien)

v8: Rebase on top of the crtc->primary changes

v9: Split the staging results structure to not exceed the 1Kb stack
    allocation in skl_update_wm()

v10: Make skl_pipe_pixel_rate() take a pointer to the pipe config
     Add a comment about overflow considerations for skl_wm_method1()
     Various additions of const
     Various use of sizeof(variable) instead of sizeof(type)
     Various move of variable definitons to a narrower scope
     Zero initialize some stack allocated structures to make sure we
     don't have garbage in case we don't write all the values
     (Ville)

v11: Remove non-necessary default number of blocks/lines when the plane
     is disabled (Ville)

Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Pradeep Bhat <pradeep.bhat@intel.com>
Signed-off-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915/skl: Add DDB allocation management structures
Damien Lespiau [Tue, 4 Nov 2014 17:06:41 +0000 (17:06 +0000)]
drm/i915/skl: Add DDB allocation management structures

We now need to allocate space in the DDB for planes being scanned out
ourselves. The data structure to represent an allocation mirrors what
we'll need to write in the registers later on: (start, end).

We add that allocation datat to the skl_wm_values structure as part of
the values to program the hardware with.

v2: Split planes and cursor for consistency.

v3: Make the skl_ddb_entry_size() parameter const

Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915/skl: Definition of SKL WM param structs for pipe/plane
Pradeep Bhat [Tue, 4 Nov 2014 17:06:40 +0000 (17:06 +0000)]
drm/i915/skl: Definition of SKL WM param structs for pipe/plane

This patch defines the structures needed for computation of
watermarks of pipes and planes for SKL.

v2: Incorporated Damien's review comments and removed unused fields
    in structs for future features like rotation, drrs and scaling.
    The skl_wm_values struct is now made more generic across planes
    and cursor planes for all pipes.

v3: implemented the plane/cursor split.

v4: Change the wm union back to a structure (Ville, Daniel)

Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Pradeep Bhat <pradeep.bhat@intel.com>
Signed-off-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915/skl: Register definitions and macros for SKL Watermark regs
Pradeep Bhat [Tue, 4 Nov 2014 17:06:39 +0000 (17:06 +0000)]
drm/i915/skl: Register definitions and macros for SKL Watermark regs

This patch defines SKL specific PLANE_WM Watermark registers. It also
defines macros to get the addresses of different LP levels within a pipe.

v2: Reworked the register definitions and associated macros to make it more
    generic and be able to use for_each_pipe in values computation.
    Incorporated Damien's review comments and indentation.

v3: Added default values for lines and blocks. Provided mask for blocks.

v4: Prefix intermedidate (internal-only) macros with _ (Ville)

v5: Remove the lines and block defaults value (Ville)

Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com> (v4)
Signed-off-by: Pradeep Bhat <pradeep.bhat@intel.com>
Signed-off-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915/skl: Read the Memory Latency Values for WM computation
Pradeep Bhat [Tue, 4 Nov 2014 17:06:38 +0000 (17:06 +0000)]
drm/i915/skl: Read the Memory Latency Values for WM computation

This patch reads the memory latency values for all the 8 levels for
SKL. These values are needed for the Watermark computation.

v2: Incorporated the review comments from Damien on register
    indentation.

v3: Updated the code to use the sandybridge_pcode_read for reading
    memory latencies for GEN9.

v4: Don't put gen 9 in the middle of an ordered list of ifs
    (Damien)

v5: take the rps.hw_lock around sandybridge_pcode_read() (Damien)

v6: Use gen >= 9 in the pcode_read() function for data1.
    Move the defines near the gen6 ones and prefix them with PCODE.
    Remove unused timeout define (the pcode_read() code has a larger
    timeout already).

Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Pradeep Bhat <pradeep.bhat@intel.com>
Signed-off-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: rewrite hsw/bdw audio codec enable/disable sequences
Jani Nikula [Tue, 4 Nov 2014 08:30:23 +0000 (10:30 +0200)]
drm/i915: rewrite hsw/bdw audio codec enable/disable sequences

There's some serious confusion regarding ELD valid bit that gets set and
cleared back and forth etc. Rewrite it all based on the documented audio
codec enable/disable sequences.

v3: replace vblank wait with a comment

v4: expand the comment on what should be done with the vblank wait

Signed-off-by: Jani Nikula <jani.nikula@intel.com>
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: clean up and clarify audio related register defines
Jani Nikula [Mon, 27 Oct 2014 14:26:52 +0000 (16:26 +0200)]
drm/i915: clean up and clarify audio related register defines

Make audio related register defines conform to existing style: Add _MASK
where relevant, indent the defines for register contents, don't indent
the defines for register addresses, prefix pipe specific register
address defines with underscores, drop self explanatory comments.

No functional changes.

Signed-off-by: Jani Nikula <jani.nikula@intel.com>
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Report the actual swizzling back to userspace
Chris Wilson [Fri, 24 Oct 2014 11:11:11 +0000 (12:11 +0100)]
drm/i915: Report the actual swizzling back to userspace

Userspace cares about whether or not swizzling depends on the page
address for its direct access into bound objects. Extend the get_tiling
ioctl to report the physical swizzling value in addition to the logical
swizzling value so that userspace can accurately determine when it is
possible for manual detiling.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Akash Goel <akash.goel@intel.com>
Cc: Daniel Vetter <daniel.vetter@ffwll.ch>
Testcase: igt/gem_tiled_wc
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Request PIN_GLOBAL when pinning a vma for GTT relocations
Chris Wilson [Fri, 31 Oct 2014 13:53:53 +0000 (13:53 +0000)]
drm/i915: Request PIN_GLOBAL when pinning a vma for GTT relocations

Always require PIN_GLOBAL when we want a mappable offset (PIN_MAPPABLE).
This causes the pin to fixup the global binding in cases were the vma
was already bound (and due to the proceeding bug, we considered it to be
already mappable).

References: https://bugs.freedesktop.org/show_bug.cgi?id=85671
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
[danvet: Add WARN_ON to check that PIN_MAP implies PIN_GLOBAL as
discussed on irc.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Only mark as map-and-fenceable when bound into the GGTT
Chris Wilson [Fri, 31 Oct 2014 13:53:52 +0000 (13:53 +0000)]
drm/i915: Only mark as map-and-fenceable when bound into the GGTT

We use the obj->map_and_fenceable hint for when we already have a
valid mapping of this object in the aperture. This hint can only apply
to the GGTT and not to the aliasing-ppGTT. One user of the hint is
execbuffer relocation, which began to fail when it tried to follow the
hint and perform the relocate through the non-existent GGTT mapping.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=85671
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Use vblank evade mechanism in mmio_flip
Ander Conselvan de Oliveira [Tue, 28 Oct 2014 13:10:14 +0000 (15:10 +0200)]
drm/i915: Use vblank evade mechanism in mmio_flip

Currently we program just DPSCNTR and DSPSTRIDE directly from the ring
interrupt handler, which is fine since the hardware guarantees that
those are update atomically. When we have atomic page flips we'll want
to be able to update also the offset registers, and then we need to use
the vblank evade mechanism to guarantee atomicity. Since that mechanism
introduces a wait, we need to do the actual register write from a work
when it is triggered by the ring interrupt.

v2: Explain the need for mmio_flip.work in the commit message (Paulo)
    Initialize the mmio_flip work in intel_crtc_init() (Paulo)
    Prevent new flips the previous flip work finishes (Paulo)
    Don't acquire modeset locks for mmio flip work

Note: Paulo had reservations about the work item leaking over a plane
disable. But insofar as we do lack these checks that issue is already
present with the existing code.

Signed-off-by: Ander Conselvan de Oliveira <ander.conselvan.de.oliveira@intel.com>
Reviewed-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Remove modeset lock check from intel_pipe_update_start()
Ander Conselvan de Oliveira [Tue, 28 Oct 2014 13:10:13 +0000 (15:10 +0200)]
drm/i915: Remove modeset lock check from intel_pipe_update_start()

A follow up patch will call this funcion from a work context for the
mmio flip, in which case we cannot acquire the modeset locks. That's
not a problem though, since the check is there to protect vblank and
the mode, but the code that changes that waits for pending flips
first.

Signed-off-by: Ander Conselvan de Oliveira <ander.conselvan.de.oliveira@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Add kerneldoc for intel_pipe_update_{start, end}
Ander Conselvan de Oliveira [Tue, 28 Oct 2014 13:10:12 +0000 (15:10 +0200)]
drm/i915: Add kerneldoc for intel_pipe_update_{start, end}

Note that a later patch will use these functions in some other file
and drop the static. Hence the kerneldoc looks appropriate.

Signed-off-by: Ander Conselvan de Oliveira <ander.conselvan.de.oliveira@intel.com>
Reviewed-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
[danvet: Add comment that the functions will become non-static
shortly.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Remove redundant parameter to i915_gem_object_wait_rendering__tail()
John Harrison [Thu, 30 Oct 2014 18:40:53 +0000 (18:40 +0000)]
drm/i915: Remove redundant parameter to i915_gem_object_wait_rendering__tail()

An earlier commit (c8725f3dc0911d4354315a65150aecd8b7d0d74a: Do not call
retire_requests from wait_for_rendering) removed the use of the ring parameter
within wait_rendering__tail() but did not remove the parameter itself. As the
plan is to remove obj->ring which is where this parameter comes from, it is
simpler to just remove the parameter completely than to update it with a new
source.

For: VIZ-4377
Signed-off-by: John Harrison <John.C.Harrison@Intel.com>
CC: Chris Wilson <chris@chris-wilson.co.uk>
CC: Brad Volkin <bradley.d.volkin@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: fix RPS on runtime suspend
Paulo Zanoni [Thu, 30 Oct 2014 17:59:31 +0000 (15:59 -0200)]
drm/i915: fix RPS on runtime suspend

With this patch, the RPS sequence for runtime suspend/resume is
exactly like the sequence for S3 suspend/resume:
 - flush_delayed_work(&dev_priv->rps.delayed_resume_work)
 - intel_runtime_pm_disable_interrupts()
 - intel_suspend_gt_powersave()
   (suspended)
 - intel_runtime_pm_enable_interrupts()
 - intel_enable_gt_powersave()

With this, we get rid of WARNs that are currently intermittently
triggered by the system-suspend-execbuf subtest of runtime PM. Notice
that these WARNs could also be triggered in other ways that involved
doing lots of RPM suspend/resume cycles just after a system S3 resume.

Testcase: igt/pm_rpm/system-suspend-execbuf
Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Reference: https://bugs.freedesktop.org/show_bug.cgi?id=82939
Reviewed-by: Imre Deak <imre.deak@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: fix "Unexpected fault" error message line break
Paulo Zanoni [Thu, 30 Oct 2014 17:52:45 +0000 (15:52 -0200)]
drm/i915: fix "Unexpected fault" error message line break

Fix the message, not the fault :)

This is what I see:
[  282.108597] [drm:i915_check_and_clear_faults] Unexpected fault
[  282.108597]  Addr: 0x00000000\n Address space: PPGTT
[  282.108597]  Source ID: 24
[  282.108597]  Type: 0

Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Kill leftover GTIIR writes from valleyview_irq_preinstall()
Ville Syrjälä [Fri, 31 Oct 2014 09:53:06 +0000 (11:53 +0200)]
drm/i915: Kill leftover GTIIR writes from valleyview_irq_preinstall()

There are two leftover GTIIR writes in valleyview_irq_preinstall().
Looks like the were originally left behind by:

 commit d18ea1b58a5003eb6fca03aff03c4c01321e6cb1
 Author: Daniel Vetter <daniel.vetter@ffwll.ch>
 Date:   Fri Jul 12 22:43:25 2013 +0200

    drm/i915: unify PM interrupt preinstall sequence

and then the GTIIR reset was added back here:

 commit f86f3fb005d0c907285fa8685badcb24ec31ee59
 Author: Paulo Zanoni <paulo.r.zanoni@intel.com>
 Date:   Tue Apr 1 15:37:14 2014 -0300

    drm/i915: properly clear IIR at irq_uninstall on Gen5+

so we can kill the leftovers from the vlv code.

Cc: Paulo Zanoni <przanoni@gmail.com>
Suggested-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Drop useless VLV_IIR writes from vlv_display_irq_postinstall()
Ville Syrjälä [Thu, 30 Oct 2014 17:43:01 +0000 (19:43 +0200)]
drm/i915: Drop useless VLV_IIR writes from vlv_display_irq_postinstall()

The extra VLV_IIR writes at the end of vlv_display_irq_postinstall()
serve no purpose. Remove them.

The VLV_IMR/IER/IIR setup at the start of the function also seems a bit
pointless since it doesn't unmask/enable anything. But leave it be for
now.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i914: Refactor vlv_display_irq_postinstall()
Ville Syrjälä [Thu, 30 Oct 2014 17:43:00 +0000 (19:43 +0200)]
drm/i914: Refactor vlv_display_irq_postinstall()

Split the vlv display irq postinstall code to a separate function so
that we can share it with chv.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Refactor vlv_display_irq_reset()
Ville Syrjälä [Thu, 30 Oct 2014 17:42:58 +0000 (19:42 +0200)]
drm/i915: Refactor vlv_display_irq_reset()

Pull the vlv display irq reset code to a new functions. The aim is to
share the code with chv.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Make valleyview_display_irqs_(un)install() work for chv
Ville Syrjälä [Thu, 30 Oct 2014 17:42:57 +0000 (19:42 +0200)]
drm/i915: Make valleyview_display_irqs_(un)install() work for chv

Genralize valleyview_display_irqs_install() and
valleyview_display_irqs_uninstall() enough so that they work on chv.
The only difference to vlv here being the third pipe that chv brings.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Call gen5_gt_irq_reset() from valleyview_irq_uninstall()
Ville Syrjälä [Thu, 30 Oct 2014 17:42:56 +0000 (19:42 +0200)]
drm/i915: Call gen5_gt_irq_reset() from valleyview_irq_uninstall()

Looks like we forgot to call gen5_gt_irq_reset() for vlv in the
uninstall phase. Do so.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Use GEN5_IRQ_RESET() on vlv/chv
Ville Syrjälä [Thu, 30 Oct 2014 17:42:55 +0000 (19:42 +0200)]
drm/i915: Use GEN5_IRQ_RESET() on vlv/chv

Replace the hand rolled IIR,IER,IMR disable sequences with
GEN5_IRQ_RESET().

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Use a consistent order between IIR, IER, IMR writes on vlv/chv
Ville Syrjälä [Thu, 30 Oct 2014 17:42:54 +0000 (19:42 +0200)]
drm/i915: Use a consistent order between IIR, IER, IMR writes on vlv/chv

Follow the same ordering rules for the IIR,IER,IMR writes on vlv/chv
that we do on other gen5+ platforms.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>