GitHub/moto-9609/android_kernel_motorola_exynos9610.git
10 years agodrm/i915: Power gating display wells during i915_pm_suspend
Borun Fu [Sat, 12 Jul 2014 04:32:27 +0000 (10:02 +0530)]
drm/i915: Power gating display wells during i915_pm_suspend

On VLV, after i915_pm_suspend display power wells are staying
power ungated. So, after initiating mem sleep "echo mem > /sys/power/state"
Display is staing D0 State. There might be better way/place to power gate
these wells. Also, we need to make sure that if wells are power gated due to
DPMS OFF sequence, they need not be turned off by i915_pm_suspend again.

v2: Extracted helper for intel_crtc_disable and power gating CRTC power wells.
[Daniel]

Cc: Imre Deak <imre.deak@intel.com>
Cc: Paulo Zanoni <paulo.r.zanoni@intel.com>
Cc: Daniel Vetter <daniel.vetter@ffwll.ch>
Cc: Jani Nikula <jani.nikula@linux.intel.com>
Change-Id: I34c80da66aa24c423a5576c68aa1f3a8d0f43848
Signed-off-by: Borun Fu <borun.fu@intel.com>
Signed-off-by: Sagar Kamble <sagar.a.kamble@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: byt_gpu_freq() can be static
Fengguang Wu [Sat, 12 Jul 2014 09:21:39 +0000 (11:21 +0200)]
drm/i915: byt_gpu_freq() can be static

CC: Deepak S <deepak.s@linux.intel.com>
CC: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Fengguang Wu <fengguang.wu@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915/chv: calculate rc6 residency correctly
Mika Kuoppala [Wed, 9 Jul 2014 11:55:56 +0000 (14:55 +0300)]
drm/i915/chv: calculate rc6 residency correctly

The register to read cz count is different from vlv. Also
the counts returned from CCK_CTL1 for BSW are (ticks in 30ns - 1).
czcount_30ns of value 1 is a special case for 320Mhz.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=80703
Suggested-by: Deepak S <deepak.s@linux.intel.com>
Cc: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Mika Kuoppala <mika.kuoppala@intel.com>
Tested-by: Guo Jinxian <jinxianx.guo@intel.com>
Reviewed-by: Deepak S <deepak.s@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agoDocumentation: drm: describing rotation property
Sagar Kamble [Tue, 8 Jul 2014 05:02:03 +0000 (10:32 +0530)]
Documentation: drm: describing rotation property

Cc: damien.lespiau@intel.com
Cc: daniel.vetter@ffwll.ch
Cc: ville.syrjala@linux.intel.com
Signed-off-by: Sagar Kamble <sagar.a.kamble@intel.com>
Reviewed-by: Damien Lespiau <damien.lespiau@intel.com>
Acked-by: Dave Airlie <airlied@linux.ie>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agoDocumentation: drm: Removing placeholders for generic drm properties description
Sagar Kamble [Tue, 8 Jul 2014 05:02:02 +0000 (10:32 +0530)]
Documentation: drm: Removing placeholders for generic drm properties description

These property descriptions were kept as placeholder. Removing them for simplicity.

Cc: damien.lespiau@intel.com
Cc: daniel.vetter@ffwll.ch
Cc: ville.syrjala@linux.intel.com
Signed-off-by: Sagar Kamble <sagar.a.kamble@intel.com>
Reviewed-by: Damien Lespiau <damien.lespiau@intel.com>
Acked-by: Dave Airlie <airlied@linux.ie>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm: Add drm_rotation_simplify()
Ville Syrjälä [Tue, 8 Jul 2014 05:01:56 +0000 (10:31 +0530)]
drm: Add drm_rotation_simplify()

drm_rotation_simplify() can be used to eliminate unsupported rotation
flags. It will check if any unsupported flags are present, and if so
it will modify the rotation to an alternate form by adding 180 degrees
to rotation angle, and flipping the reflect x and y bits. The hope is
that this identity transform will eliminate the unsupported flags.

Of course that might not result in any more supported rotation, so
the caller is still responsible for checking the result afterwards.

Cc: dri-devel@lists.freedesktop.org
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Imre Deak <imre.deak@intel.com>
Acked-by: Dave Airlie <airlied@linux.ie>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm: Add drm_rect rotation functions
Ville Syrjälä [Tue, 8 Jul 2014 05:01:55 +0000 (10:31 +0530)]
drm: Add drm_rect rotation functions

Add some helper functions to move drm_rects between different rotated
coordinate spaces. One function does the forward transform and
another does the inverse.

Cc: dri-devel@lists.freedesktop.org
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Imre Deak <imre.deak@intel.com>
Acked-by: Dave Airlie <airlied@linux.ie>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/omap: Switch omapdrm over to drm_mode_create_rotation_property()
Ville Syrjälä [Tue, 8 Jul 2014 05:01:54 +0000 (10:31 +0530)]
drm/omap: Switch omapdrm over to drm_mode_create_rotation_property()

Use the new drm_mode_create_rotation_property() in omapdrm.

Cc: dri-devel@lists.freedesktop.org
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Rob Clark <robdclark@gmail.com>
Reviewed-by: Imre Deak <imre.deak@intel.com>
Reviewed-by: Sagar Kamble <sagar.a.kamble@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm: Add drm_mode_create_rotation_property()
Ville Syrjälä [Tue, 8 Jul 2014 05:01:53 +0000 (10:31 +0530)]
drm: Add drm_mode_create_rotation_property()

Add a function to create a standards compliant rotation property.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Imre Deak <imre.deak@intel.com>
Acked-by: Dave Airlie <airlied@linux.ie>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm: Add support_bits parameter to drm_property_create_bitmask()
Ville Syrjälä [Tue, 8 Jul 2014 05:01:52 +0000 (10:31 +0530)]
drm: Add support_bits parameter to drm_property_create_bitmask()

Make drm_property_create_bitmask() a bit more generic by allowing the
caller to specify which bits are in fact supported. This allows multiple
callers to use the same enum list, but still create different versions
of the same property with different list of supported bits.

v2: Populate values[] array as non-sparse
    Make supported_bits 64bit
    Fix up omapdrm call site (Rob)

Cc: dri-devel@lists.freedesktop.org
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Imre Deak <imre.deak@intel.com>
Reviewed-by: Sagar Kamble <sagar.a.kamble@intel.com>
Acked-by: Dave Airlie <airlied@linux.ie>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm: Move DRM_ROTATE bits out of omapdrm into drm_crtc.h
Ville Syrjälä [Tue, 8 Jul 2014 05:01:51 +0000 (10:31 +0530)]
drm: Move DRM_ROTATE bits out of omapdrm into drm_crtc.h

The rotation property stuff should be standardized among all drivers.
Move the bits to drm_crtc.h from omap_drv.h.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Imre Deak <imre.deak@intel.com>
Acked-by: Dave Airlie <airlied@linux.ie>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Add RP1 render P state thresholds in CHV
Deepak S [Sat, 12 Jul 2014 13:16:14 +0000 (18:46 +0530)]
drm/i915: Add RP1 render P state thresholds in CHV

This is useful for userspace utilities to verify and micromanaging
the increase/decrease frequncy.

v2: Use vlv_gpu_freq to get freq (Deepak)

Signed-off-by: Deepak S <deepak.s@linux.intel.com>
Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915/chv: Add basic PM interrupt support for CHV
Deepak S [Thu, 10 Jul 2014 07:46:26 +0000 (13:16 +0530)]
drm/i915/chv: Add basic PM interrupt support for CHV

Enabled PM interrupt programming for CHV. Re-using gen8 code and extending same for CHV.

Signed-off-by: Deepak S <deepak.s@linux.intel.com>
Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: CHV GPU frequency to opcode functions
Deepak S [Sat, 12 Jul 2014 09:24:33 +0000 (14:54 +0530)]
drm/i915: CHV GPU frequency to opcode functions

Adding chv specific fre/encode conversion.

v2: Remove generic function and platform check (Daniel)

Signed-off-by: Deepak S <deepak.s@linux.intel.com>
Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: populate mem_freq/cz_clock for chv
Deepak S [Thu, 10 Jul 2014 07:46:24 +0000 (13:16 +0530)]
drm/i915: populate mem_freq/cz_clock for chv

We need mem_freq or cz clock for freq/opcode conversion

Signed-off-by: Deepak S <deepak.s@linux.intel.com>
Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Add RP0/RP1/RPn render P state thresholds in VLV sysfs
Deepak S [Thu, 10 Jul 2014 07:46:22 +0000 (13:16 +0530)]
drm/i915: Add RP0/RP1/RPn render P state thresholds in VLV sysfs

This is useful for userspace utilities to verify and micromanaging the
increase/decrease frequncy.

Signed-off-by: Deepak S <deepak.s@linux.intel.com>
Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Read guaranteed freq for valleyview
Deepak S [Thu, 10 Jul 2014 07:46:21 +0000 (13:16 +0530)]
drm/i915: Read guaranteed freq for valleyview

Reading RP1 for valleyview to help us enable "pm_rps" i-g-t testcase
execution.

Signed-off-by: Deepak S <deepak.s@linux.intel.com>
Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com>
[danvet: Add missing static.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Call encoder->post_disable() in intel_sanitize_encoder()
Ville Syrjälä [Fri, 27 Jun 2014 23:04:01 +0000 (02:04 +0300)]
drm/i915: Call encoder->post_disable() in intel_sanitize_encoder()

VLV and CHV disable the DP port only in the .post_disable() hook, so we
need to make intel_sanitize_encoder() call that when it's trying to
disable encoders without an active pipes.

My bsw actaully hits this when an external display is connected. The
BIOS still likes to turn on the eDP port, but leaves the pipe disabled.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Rafael Barbalho <rafael.barbalho@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: ddi: enable runtime pm during dpms
Daniel Vetter [Wed, 25 Jun 2014 19:02:04 +0000 (22:02 +0300)]
drm/i915: ddi: enable runtime pm during dpms

Signed-off-by: Imre Deak <imre.deak@intel.com>
Reviewed-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
[danvet: Remove now bogus comment.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Use the cached min/min/rpe values in the vlv debugfs code
Ville Syrjälä [Fri, 27 Jun 2014 23:03:53 +0000 (02:03 +0300)]
drm/i915: Use the cached min/min/rpe values in the vlv debugfs code

No need to re-read the hardware rps fuses when we already have all the
values tucked away in dev_priv->rps.

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: Don't disable PPGTT for CHV based in PCI rev
Ville Syrjälä [Fri, 27 Jun 2014 23:03:56 +0000 (02:03 +0300)]
drm/i915: Don't disable PPGTT for CHV based in PCI rev

In
 commit 62942ed7279d3e06dc15ae3d47665eff3b373327
 Author: Jesse Barnes <jbarnes@virtuousgeek.org>
 Date:   Fri Jun 13 09:28:33 2014 -0700

    drm/i915/vlv: disable PPGTT on early revs v3

we forgot about CHV. IS_VALLEYVIEW() is true for CHV, so we need to
explicitly avoid disabling PPGTT on CHV.

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: Make the RPS interrupt generation mask handle the vlv wa
Chris Wilson [Thu, 10 Jul 2014 19:31:19 +0000 (20:31 +0100)]
drm/i915: Make the RPS interrupt generation mask handle the vlv wa

We can eliminate a lot of special case code by making the computation of
the interrupt mask be correct for all callers.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Move RPS evaluation interval counters to i915->rps
Chris Wilson [Thu, 10 Jul 2014 19:31:18 +0000 (20:31 +0100)]
drm/i915: Move RPS evaluation interval counters to i915->rps

Place the RPS counters inside the RPS struct.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Don't cast a pointer to void* unnecessarily
Damien Lespiau [Thu, 10 Jul 2014 13:52:43 +0000 (14:52 +0100)]
drm/i915: Don't cast a pointer to void* unnecessarily

C is super happy to asign anything pointer to void *. Don't pretend
otherwise.

Signed-off-by: Damien Lespiau <damien.lespiau@intel.com>
Reviewed-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: don't read LVDS regs at compute_config time
Paulo Zanoni [Fri, 4 Jul 2014 16:38:36 +0000 (13:38 -0300)]
drm/i915: don't read LVDS regs at compute_config time

We may reach this point while the machine is still runtime suspended,
so we'll hit a WARN. The other encoders also don't touch registers at
this point, so instead of waking the machine up, write some code to
keep the register always at the same state, including after we runtime
suspend/resume.

Testcase: igt/pm_rpm
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=80463
Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Reviewed-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: check the power domains in intel_lvds_get_hw_state()
Paulo Zanoni [Fri, 4 Jul 2014 16:38:35 +0000 (13:38 -0300)]
drm/i915: check the power domains in intel_lvds_get_hw_state()

Just like we do for the other encoders. This should fix some WARNs
when running pm_rpm on SNB.

Testcase: igt/pm_rpm
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=80463
Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Reviewed-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: check the power domains in ironlake_get_pipe_config()
Paulo Zanoni [Fri, 4 Jul 2014 16:38:34 +0000 (13:38 -0300)]
drm/i915: check the power domains in ironlake_get_pipe_config()

Just like we already do in haswell_get_pipe_config(). This should
prevent some WARNs when we run pm_rpm on SNB.

Testcase: igt/pm_rpm
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=80463
Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Reviewed-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: don't skip shared DPLL assertion on LPT
Paulo Zanoni [Fri, 4 Jul 2014 14:30:28 +0000 (11:30 -0300)]
drm/i915: don't skip shared DPLL assertion on LPT

Since we now have support for shared DPLLS.

Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Reviewed-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Only touch WRPLL hw state in enable/disable hooks
Daniel Vetter [Wed, 25 Jun 2014 19:02:03 +0000 (22:02 +0300)]
drm/i915: Only touch WRPLL hw state in enable/disable hooks

To be able to do this we need to separately keep track of how many
crtcs need a given WRPLL and how many actually actively use it. The
common shared dpll framework already has all this, including massive
state readout and cross checking. Which allows us to do this switch in
a fairly small patch.

Reviewed-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Switch to common shared dpll framework for WRPLLs
Daniel Vetter [Wed, 25 Jun 2014 19:02:02 +0000 (22:02 +0300)]
drm/i915: Switch to common shared dpll framework for WRPLLs

Mostly this patch is one big excersize in deleting code and asserts
which are no longer needed. Note that we still abuse the shared dpll
framework a bit since we call the enable/disable functions from the
crtc mode_set and off hooks. But changing the actual hardware sequence
will be done in the next step.

Note that besides the massive amount of changes in this patch the
places and order in which the low-level WRPLL code is called is
absolutely unchanged.

Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Reviewed-by: Damien Lespiau <damien.lespiau@intel.com>
[imre: rebased on patchset version w/o pch/crt/fdi refactoring]
Signed-off-by: Imre Deak <imre.deak@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: ->enable hook for WRPLLs
Daniel Vetter [Wed, 25 Jun 2014 19:02:01 +0000 (22:02 +0300)]
drm/i915: ->enable hook for WRPLLs

This time around another cute hack to pre-fill the pll->hw_state with
the right values. And also remove a bunch of checks which will be
replaced by lots more checks in the common framework.

Reviewed-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: ->disable hook for WRPLLs
Daniel Vetter [Wed, 25 Jun 2014 19:02:00 +0000 (22:02 +0300)]
drm/i915: ->disable hook for WRPLLs

Currently still with a redudant WARN_ON in there, the common shared
dpll code will take care of this in the future.

Also we need to flip the switch for the transitional hack now to make
sure that we disable the right pll.

Reviewed-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: State readout support for WRPLLs
Daniel Vetter [Fri, 4 Jul 2014 14:27:39 +0000 (11:27 -0300)]
drm/i915: State readout support for WRPLLs

Still tacked onto the side, but slowly getting there.

v2: Don't forget the debugfs file.

v3 (from Paulo): Don't forget to check the power domains.

Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Reviewed-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: add POWER_DOMAIN_PLLS
Paulo Zanoni [Fri, 4 Jul 2014 14:27:38 +0000 (11:27 -0300)]
drm/i915: add POWER_DOMAIN_PLLS

And get/put it when needed. The special thing about this commit is
that it will now return false in ibx_pch_dpll_get_hw_state() in case
the power domain is not enabled. This will fix some WARNs we have when
we run pm_rpm on SNB.

Testcase: igt/pm_rpm
Bugzilla:https://bugs.freedesktop.org/show_bug.cgi?id=80463
Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Reviewed-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Document that the pll->mode_set hook is optional
Daniel Vetter [Wed, 25 Jun 2014 19:01:58 +0000 (22:01 +0300)]
drm/i915: Document that the pll->mode_set hook is optional

The WRPLLs won't use them.

Reviewed-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Basic shared dpll support for WRPLLs
Daniel Vetter [Wed, 25 Jun 2014 19:01:57 +0000 (22:01 +0300)]
drm/i915: Basic shared dpll support for WRPLLs

Just filing in names and ids, but not yet officially registering them
so that the hw state cross checker doesn't completely freak out about
them. Still since we do already read out and cross check
config->shared_dpll the basics are now there to flesh out the wrpll
shared dpll implementation.

The idea is now to roll out all the callbacks step-by-step and then at
the end switch to the shared dpll framework. This way hw and sw
changes are clearly separated.

Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
[imre: added const to hsw_ddi_pll_names (Damien)]
Reviewed-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Precompute static ddi_pll_sel values in encoders
Daniel Vetter [Fri, 4 Jul 2014 14:26:04 +0000 (11:26 -0300)]
drm/i915: Precompute static ddi_pll_sel values in encoders

This way only the dynamic WRPLL selection for hdmi ddi mode is
done in intel_ddi_pll_select.

v2: Don't clobber the precomputed values when selecting clocks fro
hdmi encoders.
v3 (from Paulo): Rebase on top of the s/IS_HASWELL/HAS_DDI/ patch.

Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Paulo Zanoni <przanoni@gmail.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: BDW also has special-purpose DP DDI clocks
Paulo Zanoni [Fri, 4 Jul 2014 14:26:03 +0000 (11:26 -0300)]
drm/i915: BDW also has special-purpose DP DDI clocks

Don't let it fall in the HAS_PCH_SPLIT() case.

Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Reviewed-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: State readout and cross-checking for ddi_pll_sel
Daniel Vetter [Wed, 25 Jun 2014 19:01:55 +0000 (22:01 +0300)]
drm/i915: State readout and cross-checking for ddi_pll_sel

To make things a bit more manageable extract a new function for
reading out common ddi port state. This means a bit of duplication
between encoders and the core since both look at the same registers,
but doesn't seem worth to make a fuzz about.

We can also remove the state readout code in intel_ddi_setup_hw_pll_state.
That code is only called from the hardware take over and not the cross
check code, and only after the crtc state is reconstructed. So we can
rely on an accurate value of crtc->config.ddi_pll_sel already.

Compared to the old code also trust the hw state more and don't
special-case port A - we want to cross-check the actual-state, not
bake in our own assumptions about how this is supposed to all be
linked up.

v2: Make use of the read-out ddi_pll_sel in intel_ddi_clock_get.

Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Reviewed-by: Damien Lespiau <damien.lespiau@intel.com>
[imre: rebased on patchset version w/o pch/crt/fdi refactoring]
Signed-off-by: Imre Deak <imre.deak@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Move ddi_pll_sel into the pipe config
Daniel Vetter [Wed, 25 Jun 2014 19:01:54 +0000 (22:01 +0300)]
drm/i915: Move ddi_pll_sel into the pipe config

Just boring sed job for preparation.

Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Reviewed-by: Damien Lespiau <damien.lespiau@intel.com>
[imre: rebased on patchset version w/o pch/crt/fdi refactoring]
Signed-off-by: Imre Deak <imre.deak@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Add a debugfs file for the shared dpll state
Daniel Vetter [Wed, 25 Jun 2014 19:01:53 +0000 (22:01 +0300)]
drm/i915: Add a debugfs file for the shared dpll state

Reviewed-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Move SPLL disabling into hsw_crt_post_disable
Daniel Vetter [Wed, 25 Jun 2014 19:01:52 +0000 (22:01 +0300)]
drm/i915: Move SPLL disabling into hsw_crt_post_disable

Similar to how the ->crtc_mode_set hook should touch the hardware to
enable anything the ->crtc_off hook should disable anything in the
hardware. Otherwise runtime pm for dpms will not work.

Currently the only things left int the haswell_crtc_off hook is
disabling the ddi plls. We can't move the WRPLL enabling out yet
because the current ddi pll sharing code used by the haswell code
doesn't separately track active users and overall users. This must be
fixed by porting it to the generic shared display pll framework, which
is powerful enough.

But the SPLL source is only used by the crt encoder and so can be
moved already. We only need to make sure that the ddi port E is
already off, which hsw_fdi_disable does by calling
intel_ddi_post_disable.

With this the code reorg to shuffle hsw fdi/lpt specific code into a
new hsw-specific crt encoder type is now finally complete.

Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
[imre: rebased on patchset version w/o pch/crt/fdi refactoring]
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: Move the SPLL enabling into hsw_crt_pre_enable
Daniel Vetter [Wed, 25 Jun 2014 19:01:51 +0000 (22:01 +0300)]
drm/i915: Move the SPLL enabling into hsw_crt_pre_enable

The call to intel_ddi_pll_enable in haswell_crtc_mode_set is the only
function that still touches the hardware state from the crtc mode_set
callback on hsw. Since the SPLL isn't ever shared we can easily take
it out into the hsw crt encoder functions.

Temporarily we'll loose a bit of WARN_ON coverage with this, but once
the WRPLLs are switched over that will be restored. For the SPLL
selection add a WARN in the hsw fdi link training code.

Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
[imre: rebased on patchset version w/o pch/crt/fdi refactoring]
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: ddi: move pch cleanup before encoder->post_disable
Imre Deak [Wed, 25 Jun 2014 19:01:50 +0000 (22:01 +0300)]
drm/i915: ddi: move pch cleanup before encoder->post_disable

This is needed by an upcoming patch that moves the PCH/CRT PLL disabling
into the post_disable hook, after which we want to keep the modeset
sequence at its current state. At this point this won't have an effect
since the PCH/CRT post_disable hook is atm a NOP.

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: ddi: move pch setup after encoder->pre_enable
Imre Deak [Wed, 25 Jun 2014 19:01:49 +0000 (22:01 +0300)]
drm/i915: ddi: move pch setup after encoder->pre_enable

This is needed by an upcoming patch that moves the PCH/CRT PLL enabling
into the pre_enable hook, after which we want to keep the modeset
sequence at its current state. At this point this won't have an effect
since the PCH/CRT pre_enable hook is atm a NOP.

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: Clean up WRPLL/SPLL #defines
Daniel Vetter [Wed, 25 Jun 2014 19:01:48 +0000 (22:01 +0300)]
drm/i915: Clean up WRPLL/SPLL #defines

Luckily the bit definitions match, but it's still confusing
to use one when handling the other. So sprinkle some OCD over
the #defines to make them match and use the right version in
each place.

Maybe we should unify these definitions completely, but that
can always be done sometime in the future.

Reviewed-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Remove spll_refcount for hsw
Daniel Vetter [Wed, 25 Jun 2014 19:01:47 +0000 (22:01 +0300)]
drm/i915: Remove spll_refcount for hsw

SPLL would be a reference clock we could potentially share,
especially if we want to use the SSC mode. But currently we
don't, so let's rip out this complexity for a simpler conversion
to the new display pll framework.

Reviewed-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Check hw state in assert_can_disable_lcpll
Daniel Vetter [Wed, 25 Jun 2014 19:01:46 +0000 (22:01 +0300)]
drm/i915: Check hw state in assert_can_disable_lcpll

All the other checks also check hw state, so checking our software
refcounts for the plls looks a bit odd. Also this will simplify the
conversion over to the shared dpll framework, which itself has massive
amounts of checks to make sure that we never leave a display pll
enabled when we shouldn't.

So after that conversion we should stil have a good enough coverage of
asserts for entering pc8/runtime pm on hsw/bdw.

Reviewed-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Add a couple WARN()'s to catch missing locks
Matt Roper [Wed, 9 Jul 2014 23:22:11 +0000 (16:22 -0700)]
drm/i915: Add a couple WARN()'s to catch missing locks

Add !mutex_is_locked() checks to intel_pin_and_fence_fb_obj() and
intel_unpin_fb_obj() to help catch failures to grab struct_mutex when
operating on fb objects.

Signed-off-by: Matt Roper <matthew.d.roper@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Add missing locking to primary plane handlers
Matt Roper [Wed, 9 Jul 2014 23:22:10 +0000 (16:22 -0700)]
drm/i915: Add missing locking to primary plane handlers

intel_primary_plane_{setplane,disable} were lacking struct_mutex locking
around their GEM operations.

Signed-off-by: Matt Roper <matthew.d.roper@intel.com>
Reported-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: get/put runtime PM at i915_semaphore_status
Paulo Zanoni [Wed, 9 Jul 2014 17:31:57 +0000 (14:31 -0300)]
drm/i915: get/put runtime PM at i915_semaphore_status

Otherwise we will print some WARNs when we read registers and the
machine is suspended.

Testcase: igt/pm_rpm/debugfs-read
Cc: Ben Widawsky <ben@bwidawsk.net>
Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Reviewed-by: Ben Widawsky <ben@bwidawsk.net>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: fix D_COMP usage on BDW
Paulo Zanoni [Fri, 4 Jul 2014 14:59:58 +0000 (11:59 -0300)]
drm/i915: fix D_COMP usage on BDW

On HSW, the D_COMP register can be accessed through the mailbox (read
and write) or through MMIO on a MCHBAR offset (read only). On BDW, the
access should be done through MMIO on another address. So to account
for all these cases, create hsw_read_dcomp() with the correct
implementation for reading, and also fix hsw_write_dcomp() to do the
correct thing on BDW.

With this patch, we can now get back from the PC8+ state on BDW. We
were previously getting a black screen and lots of dmesg errors.
Please notice that the bug only happens when you actually reach the
PC8+ states, not when you only allow it.

Testcase: igt/pm_rpm/rte
Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Reviewed-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: fix hsw_write_dcomp() error message
Paulo Zanoni [Fri, 4 Jul 2014 14:59:57 +0000 (11:59 -0300)]
drm/i915: fix hsw_write_dcomp() error message

That function can be used to write anything on D_COMP, not just
disable it, so print a more appropriate message.

Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Reviewed-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: remove i915_rstdby_delays debugfs entry
Mika Kuoppala [Wed, 9 Jul 2014 12:10:46 +0000 (15:10 +0300)]
drm/i915: remove i915_rstdby_delays debugfs entry

CHV hard hangs on reading on 0x11100

References: https://bugs.freedesktop.org/show_bug.cgi?id=80893
Signed-off-by: Mika Kuoppala <mika.kuoppala@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: remove i915_gfxec debugfs entry
Mika Kuoppala [Wed, 9 Jul 2014 12:10:45 +0000 (15:10 +0300)]
drm/i915: remove i915_gfxec debugfs entry

CHV hard hangs on reading on 0x112f4.

References: https://bugs.freedesktop.org/show_bug.cgi?id=80893
Suggested-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Mika Kuoppala <mika.kuoppala@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: remove i915_inttoext_table debugfs entry
Mika Kuoppala [Wed, 9 Jul 2014 12:10:44 +0000 (15:10 +0300)]
drm/i915: remove i915_inttoext_table debugfs entry

CHV hard hangs on reading these registers. As these have not
been used since cantiga & ilk, remove the debugfs entry.

References: https://bugs.freedesktop.org/show_bug.cgi?id=80893
Suggested-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Mika Kuoppala <mika.kuoppala@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: remove i915_delayedfreq_table debugfs entry
Mika Kuoppala [Wed, 9 Jul 2014 12:10:43 +0000 (15:10 +0300)]
drm/i915: remove i915_delayedfreq_table debugfs entry

CHV hard hangs on reading these registers. As these have not
been used since cantiga & ilk, remove the debugfs entry.

References: https://bugs.freedesktop.org/show_bug.cgi?id=80893
Suggested-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Mika Kuoppala <mika.kuoppala@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Make use of intel_fb_obj() (v2)
Matt Roper [Tue, 8 Jul 2014 14:50:07 +0000 (07:50 -0700)]
drm/i915: Make use of intel_fb_obj() (v2)

This should hopefully simplify the display code slightly and also
solves at least one mistake in intel_pipe_set_base() where
to_intel_framebuffer(fb)->obj is referenced during local variable
initialization, before 'if (!fb)' gets checked.

Potential uses of this macro were identified via the following
Coccinelle patch:

        @@
        expression E;
        @@
        * to_intel_framebuffer(E)->obj

        @@
        expression E;
        identifier I;
        @@
          I = to_intel_framebuffer(E);
          ...
        * I->obj

v2: Rewrite some NULL tests in terms of the obj rather than the fb.
    Also add a WARN() if trying to pageflip with a disabled primary
    plane.  [Suggested by Chris Wilson]

Signed-off-by: Matt Roper <matthew.d.roper@intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Introduce intel_fb_obj() macro
Matt Roper [Tue, 8 Jul 2014 01:21:47 +0000 (18:21 -0700)]
drm/i915: Introduce intel_fb_obj() macro

Add an intel_fb_obj() macro that returns the GEM object associated with
a DRM framebuffer.  This macro is safe to call on NULL framebuffers (a
NULL object pointer will be returned in this case).

Signed-off-by: Matt Roper <matthew.d.roper@intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: HWS must be in the mappable region for g33
Chris Wilson [Thu, 3 Jul 2014 21:33:03 +0000 (17:33 -0400)]
drm/i915: HWS must be in the mappable region for g33

On g33, the documentation states

"HWS_PGA:
 Format = Bits 28:12 of graphics memory address (bits 31:29 MBZ)."

which translates to that the address of the HWS must be below 256MiB,
which is conveniently the mappable aperture.

This also appears to be true (but not documented as so) for gen4 and
gen5. To generalise we force it into the low mappable region for all
non-LLC platforms. If we locate the HWS at the top of the GTT the
machine will hard hang during boot (fails on pnv, gm45, ilk and byt,
but works on snb, ivb, hsw).

v2: Add comments to explain why use PIN_MAPPABLE even though we have
    no intention of mapping the object. (Ville)

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: 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/vlv: WA for Turbo and RC6 to work together.
Deepak S [Thu, 3 Jul 2014 21:33:01 +0000 (17:33 -0400)]
drm/i915/vlv: WA for Turbo and RC6 to work together.

With RC6 enabled, BYT has an HW issue in determining the right
Gfx busyness.
WA for Turbo + RC6: Use SW based Gfx busy-ness detection to decide
on increasing/decreasing the freq. This logic will monitor C0
counters of render/media power-wells over EI period and takes
necessary action based on these values

v2: Refactor duplicate code. (Ville)

v3: Reformat the comments. (Ville)

v4: Enable required counters and remove unwanted code (Ville)

v5: Added frequency change acceleration support and remove kernel-doc
style comments. (Ville)

v6: Updated comment section and Fix w/a comment. (Ville)

Signed-off-by: Deepak S <deepak.s@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: Use mmio flips to change tiling mode on Baytrail
Chris Wilson [Tue, 8 Jul 2014 09:40:30 +0000 (10:40 +0100)]
drm/i915: Use mmio flips to change tiling mode on Baytrail

For whatever reason, MI_DISPLAY_FLIP fails to change tiling mode on
Baytrail, so just use CPU driven mmio flips instead.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=76176
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Disable RCS flips on Ivybridge
Chris Wilson [Tue, 8 Jul 2014 09:40:29 +0000 (10:40 +0100)]
drm/i915: Disable RCS flips on Ivybridge

We currently see random GPU hangs when using RCS flips with multiple
pipes on Ivybridge. Now that we have mmio flips, we can fairly cheaply
fallback to using CPU driven flips instead.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=77104
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Extract the actual workload submission mechanism from execbuffer
Oscar Mateo [Thu, 3 Jul 2014 15:28:05 +0000 (16:28 +0100)]
drm/i915: Extract the actual workload submission mechanism from execbuffer

So that we isolate the legacy ringbuffer submission mechanism, which becomes
a good candidate to be abstracted away. This is prep-work for Execlists (which
will its own workload submission mechanism).

No functional changes.

Reviewed-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Oscar Mateo <oscar.mateo@intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Generalize intel_ring_get_tail to take a ringbuf
Oscar Mateo [Thu, 3 Jul 2014 15:28:04 +0000 (16:28 +0100)]
drm/i915: Generalize intel_ring_get_tail to take a ringbuf

Again, it's low-level enough to simply take a ringbuf and nothing
else.

Trivial change.

Reviewed-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Oscar Mateo <oscar.mateo@intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Generalize ring_space to take a ringbuf
Oscar Mateo [Thu, 3 Jul 2014 15:28:03 +0000 (16:28 +0100)]
drm/i915: Generalize ring_space to take a ringbuf

It's simple enough that it doesn't need to know anything about the
engine.

Trivial change.

Reviewed-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Oscar Mateo <oscar.mateo@intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Extract ringbuffer destroy & generalize alloc to take a ringbuf
Oscar Mateo [Thu, 3 Jul 2014 15:28:02 +0000 (16:28 +0100)]
drm/i915: Extract ringbuffer destroy & generalize alloc to take a ringbuf

More prep work: with Execlists, we are going to start creating a lot
of extra ringbuffers soon, so these functions are handy.

No functional changes.

v2: rename allocate/destroy_ring_buffer to alloc/destroy_ringbuffer_obj
because the name is more meaningful and to mirror a similar function in
the context world: i915_gem_alloc_context_obj(). Change suggested by Brad
Volkin.

Reviewed-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Oscar Mateo <oscar.mateo@intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Add kerneldoc comments to the intel_context struct
Oscar Mateo [Thu, 3 Jul 2014 15:28:01 +0000 (16:28 +0100)]
drm/i915: Add kerneldoc comments to the intel_context struct

A bit of background on the context elements.

Cc: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Oscar Mateo <oscar.mateo@intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
[danvet: Appease checkpatch.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Emphasize that ctx->id is merely a user handle
Oscar Mateo [Thu, 3 Jul 2014 15:28:00 +0000 (16:28 +0100)]
drm/i915: Emphasize that ctx->id is merely a user handle

This is an Execlists preparatory patch, since they make context ID become an
overloaded term:

- In the software, it was used to distinguish which context userspace was
  trying to use.
- In the BSpec, the term is used to describe the 20-bits long field the
  hardware uses to it to discriminate the contexts that are submitted to
  the ELSP and inform the driver about their current status (via Context
  Switch Interrupts and Context Status Buffers).

Initially, I tried to make the different meanings converge, but it proved
impossible:

- The software ctx->id is per-filp, while the hardware one needs to be
  globally unique.
- Also, we multiplex several backing states objects per intel_context,
  and all of them need unique HW IDs.
- I tried adding a per-filp ID and then composing the HW context ID as:
  ctx->id + file_priv->id + ring->id, but the fact that the hardware only
  uses 20-bits means we have to artificially limit the number of filps or
  contexts the userspace can create.

The ctx->user_handle renaming bits are done with this Cocci patch (plus
manual frobbing of the struct declaration):

    @@
    struct intel_context c;
    @@
    - (c).id
    + c.user_handle

    @@
    struct intel_context *c;
    @@
    - (c)->id
    + c->user_handle

Also, while we are at it, s/DEFAULT_CONTEXT_ID/DEFAULT_CONTEXT_HANDLE and
change the type to unsigned 32 bits.

v2: s/handle/user_handle and change the type to uint32_t as suggested by
Chris Wilson.

Reviewed-by: Jesse Barnes <jbarnes@virtuousgeek.org> (v1)
Signed-off-by: Oscar Mateo <oscar.mateo@intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Emphasize that ctx->obj & ctx->is_initialized refer to the legacy rcs ctx
Oscar Mateo [Thu, 3 Jul 2014 15:27:59 +0000 (16:27 +0100)]
drm/i915: Emphasize that ctx->obj & ctx->is_initialized refer to the legacy rcs ctx

We have already advanced that Logical Ring Contexts have their own kind
of backing objects, but everything will be better explained in the Execlists
series. For now, suffice it to say that the current backing object is only
ever used with the render ring, so we're making this fact more explicit
(which is a good reason on its own).

As for the is_initialized flag, we only use to signify that the render state
has been initialized (a.k.a. golden context, a.k.a. null context). It doesn't
mean anything for the other engines, so make that distinction obvious.

Done with the following Coccinelle patch (plus manual frobbing of the struct):

    @@
    struct intel_context c;
    @@
    - (c).obj
    + c.legacy_hw_ctx.rcs_state

    @@
    struct intel_context *c;
    @@
    - (c)->obj
    + c->legacy_hw_ctx.rcs_state

    @@
    struct intel_context c;
    @@
    - (c).is_initialized
    + c.legacy_hw_ctx.initialized

    @@
    struct intel_context *c;
    @@
    - (c)->is_initialized
    + c->legacy_hw_ctx.initialized

This Execlists prep-work patch has been suggested by Chris Wilson and Daniel
Vetter separately.

Initially, it was two separate patches:
drm/i915: Rename ctx->obj to ctx->rcs_state
drm/i915: Make it obvious that ctx->id is merely a user handle

Signed-off-by: Oscar Mateo <oscar.mateo@intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
[danvet: s/id/is_initialized/ to fix the subject and resolve a
conflict in i915_gem_context_reset. Also introduce a new lctx local
variable to avoid overtly long lines.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Extract context backing object allocation
Oscar Mateo [Thu, 3 Jul 2014 15:27:58 +0000 (16:27 +0100)]
drm/i915: Extract context backing object allocation

This is preparatory work for Execlists: we plan to use it later to
allocate our own context objects (since Logical Ring Contexts do
not have the same kind of backing objects).

No functional changes.

Reviewed-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Oscar Mateo <oscar.mateo@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 system freeze support depend on CONFIG_ACPI_SLEEP
Imre Deak [Mon, 23 Jun 2014 12:46:02 +0000 (15:46 +0300)]
drm/i915: make system freeze support depend on CONFIG_ACPI_SLEEP

To achieve further power savings during system freeze (aka connected
standby, or s0ix) we have to send a PCI_D1 opregion notification. As
the information about the state we're entering (system freeze,
suspend to ram or suspend to disk) is only available through the ACPI
subsystem, make this support depend on the relevant kconfig option.
Things will still work if this option isn't set, albeit with less than
optimial power saving.

This also fixes a compile breakage when the option is not set introduced
in

commit e5747e3adcd67ae27105003ec99fb58cba180105
Author: Jesse Barnes <jbarnes@virtuousgeek.org>
Date:   Thu Jun 12 08:35:47 2014 -0700

    drm/i915: send proper opregion notifications on suspend/resume

Reported-by: Randy Dunlap <rdunlap@infradead.org>
Signed-off-by: Imre Deak <imre.deak@intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Restrict GPU boost to the RCS engine
Chris Wilson [Thu, 12 Jun 2014 09:28:55 +0000 (10:28 +0100)]
drm/i915: Restrict GPU boost to the RCS engine

Make the assumption that media workloads are not as latency sensitive
for __wait_seqno, and that upclocking the GPU does not affect the BLT
engine. Under that assumption, we only wait to forcibly upclock the GPU
when we are stalling for results from the render pipeline.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Deepak S<deepak.s@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: correct BLC vs PWM enable/disable ordering
Jesse Barnes [Mon, 31 Mar 2014 18:13:56 +0000 (11:13 -0700)]
drm/i915: correct BLC vs PWM enable/disable ordering

With the new checks in place, we can see we're doing things backwards,
so fix them up per the spec.

Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Enable semaphores on BDW
Rodrigo Vivi [Mon, 30 Jun 2014 16:53:44 +0000 (09:53 -0700)]
drm/i915: Enable semaphores on BDW

Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915/bdw: poll semaphores
Ben Widawsky [Mon, 30 Jun 2014 16:53:43 +0000 (09:53 -0700)]
drm/i915/bdw: poll semaphores

As Ville points out, it's possible/probable we don't actually need this.
Potentially, this validates the letter of the spec, and not the spirit.

Ville:
> I discussed this on irc w/ Ben, and I was suggesting we don't need to
> poll. Polling apparently can be used as a workaround for certain
> hardware issues, but it looks like those issues shouldn't affect us,
> for the momemnt at least. So my suggestion was to try w/o polling
> first (since there could be some power cost to polling) and add the
> poll bit if problems arise.

Rodrigo: Spec suggests this as an W/A for GT3. However semaphores didn't
worked in my BDW GT2 on Signal Mode. So pool mode is definitely needed.

Reviewed-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Tested-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: semaphore debugfs
Ben Widawsky [Mon, 30 Jun 2014 16:53:42 +0000 (09:53 -0700)]
drm/i915: semaphore debugfs

Simple debugfs file to display the current state of semaphores. This is
useful if you want to see the state without hanging the GPU.

NOTE: This patch is optional to the series.

NOTE2: Like the GPU error state collection, the reads are currently
incoherent.

v2 (Rodrigo): * Iterate only on active rings.
          * s/ring_buffer/engine_cs.

Reviewed-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915/bdw: collect semaphore error state
Ben Widawsky [Mon, 30 Jun 2014 16:53:41 +0000 (09:53 -0700)]
drm/i915/bdw: collect semaphore error state

Since the semaphore information is in an object, just dump it, and let
the user parse it later.

NOTE: The page being used for the semaphores are incoherent with the
CPU. No matter what I do, I cannot figure out a way to read anything but
0s. Note that the semaphore waits are indeed working.

v2: Don't print signal, and wait (they should be the same). Instead,
print sync_seqno (Chris)

v3: Free the semaphore error object (Chris)

v4: Fix semaphore offset calculation during error state collection
(Ville)

v5: VCS2 rebase
Make semaphore object error capture coding style consistent (Ville)
Do the proper math for the signal offset (Ville)

v6: Fix small conflicts on rebase and s/ring_buffer/engine_cs (Rodrigo)

Reviewed-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Extract semaphore error collection
Ben Widawsky [Mon, 30 Jun 2014 16:53:40 +0000 (09:53 -0700)]
drm/i915: Extract semaphore error collection

v2: s/ring_buffer/engine_cs (Rodrigo)

Reviewed-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Implement MI decode for gen8
Rodrigo Vivi [Mon, 30 Jun 2014 16:53:39 +0000 (09:53 -0700)]
drm/i915: Implement MI decode for gen8

Ipehr just carries Dword 0 and on Gen 8, offsets are located
on Dword 2 and 3 of MI_SEMAPHORE_WAIT.

This implementation was based on Ben's work and on Ville's suggestion for Ben

Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
Cc: Ben Widawsky <ben@bwidawsk.net>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Reviewed-by: Ben Widawsky <ben@bwidawsk.net>
[danvet: Fixup format string.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915/bdw: implement semaphore wait
Ben Widawsky [Mon, 30 Jun 2014 16:53:38 +0000 (09:53 -0700)]
drm/i915/bdw: implement semaphore wait

Semaphore waits use a new instruction, MI_SEMAPHORE_WAIT. The seqno to
wait on is all well defined by the table in the previous patch. There is
nothing else different from previous GEN's semaphore synchronization
code.

v2: Update macros to not require the other ring's ring->id (Chris)

v3: Add missing VCS2 gen8_ring_wait init besides
    s/ring_buffer/engine_cs (Rodrigo)

Reviewed-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915/bdw: implement semaphore signal
Ben Widawsky [Mon, 30 Jun 2014 16:53:37 +0000 (09:53 -0700)]
drm/i915/bdw: implement semaphore signal

Semaphore signalling works similarly to previous GENs with the exception
that the per ring mailboxes no longer exist. Instead you must define
your own space, somewhere in the GTT.

The comments in the code define the layout I've opted for, which should
be fairly future proof. Ie. I tried to define offsets in abstract terms
(NUM_RINGS, seqno size, etc).

NOTE: If one wanted to move this to the HWSP they could. I've decided
one 4k object would be easier to deal with, and provide potential wins
with cache locality, but that's all speculative.

v2: Update the macro to not need the other ring's ring->id (Chris)
Update the comment to use the correct formula (Chris)

v3: Move the macros the ringbuffer.h to prevent churn in next patch
(Ville)

v4: Fixed compilation rebase conflict
commit 1ec9e26ddab06459e89a890431b2de064c5d1056
Author: Daniel Vetter <daniel.vetter@ffwll.ch>
Date:   Fri Feb 14 14:01:11 2014 +0100

    drm/i915: Consolidate binding parameters into flags

v5: VCS2 rebase
Replace hweight_long with hweight32

v6 (Rodrigo): * Add missed VC2 gen8 ring signal init
          * fixing conflicst on rebase
           * minor fixes on address table
      * remove WARN_ON

Reviewed-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
[danvet: s/BUG_ON/WARN_ON/]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Make semaphore updates more precise
Ben Widawsky [Mon, 30 Jun 2014 16:53:35 +0000 (09:53 -0700)]
drm/i915: Make semaphore updates more precise

With the ring mask we now have an easy way to know the number of rings
in the system, and therefore can accurately predict the number of dwords
to emit for semaphore signalling. This was not possible (easily)
previously.

There should be no functional impact, simply fewer instructions emitted.

While we're here, simply do the round up to 2 instead of the fancier
rounding we did before, which rounding up per mbox, ie 4. This also
allows us to drop the unnecessary MI_NOOP, so not really 4, 3.

v2: Use 3 dwords instead of 4 (Ville)
Do the proper calculation to get the number of dwords to emit (Ville)
Conditionally set .sync_to when semaphores are enabled (Ville)

v3: Rebased on VCS2
Replace hweight_long with hweight32 (Ville)

v4: Pull out the accidentally squashed hunk from the next patch after
rebase (Daniel).

v5: Fix conflict after rebase (Rodrigo)

Reviewed-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com> (v1)
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: gen specific ring init
Ben Widawsky [Mon, 30 Jun 2014 16:53:36 +0000 (09:53 -0700)]
drm/i915: gen specific ring init

Gen8 has already had some differentiation with how it handles rings.
Semaphores bring yet more differences, and now is as good a time as any
to do the split.

Also, since gen8 doesn't actually use semaphores up until this point,
put the proper "NULL" values in for the mbox info.

v2: v1 had a stale commit message

v3: Move everything in the is_semaphore_enabled() check

v4: VCS2 rebase
Remove double assignment of signal in render ring (Ville)

v5: Adding missed VCS2 signal init on gen8+ (Rodrigo)

Reviewed-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Updating comments.
Rodrigo Vivi [Mon, 30 Jun 2014 16:51:11 +0000 (09:51 -0700)]
drm/i915: Updating comments.

ring index calculation table was out of date after other rings were added,
although the formula is flexible and scale when adding new rings.

So this patch just update the comments and add a brief explanation
why to use sync_seqno[ring index].

Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Reviewed-by: Ben Widawsky <ben@bwidawsk.net>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Fix VCS2's ring name.
Rodrigo Vivi [Tue, 1 Jul 2014 09:41:36 +0000 (02:41 -0700)]
drm/i915: Fix VCS2's ring name.

It just fix a typo.

v2: removing underscore to let this like all other ring names (Oscar)

Cc: Oscar Mateo <oscar.mateo@intel.com>
Reviewed-by (v1): Ben Widawsky <benjamin.widawsky@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Corrected 'file_priv' to 'file' in 'i915_driver_preclose()'
John Harrison [Thu, 26 Jun 2014 17:23:52 +0000 (18:23 +0100)]
drm/i915: Corrected 'file_priv' to 'file' in 'i915_driver_preclose()'

The 'i915_driver_preclose()' function has a parameter called 'file_priv'.
However, this is misleading as the structure it points to is a 'drm_file' not a
'drm_i915_file_private'. It should be named just 'file' to avoid confusion.

Signed-off-by: John Harrison <John.C.Harrison@Intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Support pf CRC source on haswell transcoder edp
Daniel Vetter [Thu, 29 May 2014 12:10:22 +0000 (14:10 +0200)]
drm/i915: Support pf CRC source on haswell transcoder edp

The always-on power well pixel path on haswell is routed such that it
bypasses the panel fitter when we use is. Which means the pfit CRC
source won't work in that configuration.

Add a new disallow-bypass flags to the pfit pipe config state and set
it when we want to use the pf CRC. Results in a bit of flicker, but
should get the job done. We'll also undo do it afterwards to make sure
other tests arent' negatively affected.

Totally untested due to lack of hsw laptops around here.

v2: s/disallow_bypass/force_power_well_on/ to avoid a double negative
(Damien).

v3: force_thru because roadsigns.

v4: Don't forget the power wells! Also note that until the runtime pm
for DPMS series is fully merged the simple disable/enable trick won't
work since the ->crtc_mode_set callback is still required to do nasty
things. This stuff is tricky, but I think by both fixing up
get_crtc_power_domains and the debugfs wa code we should always
grab/drop the additional power well correctly.

v5: Wrap in () as suggested by Damien to avoid setting reserved values
for the edp transcoder path on bdw+

References: https://bugs.freedesktop.org/show_bug.cgi?id=72864
Cc: Damien Lespiau <damien.lespiau@intel.com>
Reviewed-by: Damien Lespiau <damien.lespiau@intel.com>
Tested-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915/bdw: 3D_CHICKEN3 has write mask bits
Michel Thierry [Mon, 7 Jul 2014 11:40:17 +0000 (12:40 +0100)]
drm/i915/bdw: 3D_CHICKEN3 has write mask bits

The workaround to limit SDE poly depth FIFO to 2 is not applied because
3D Chicken-3 mask bit is not set.

WaLimitSizeOfSDEPolyFifo is only for BDW-A and could be removed.

Signed-off-by: Michel Thierry <michel.thierry@intel.com>
Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agoDRM/i915: Remove magic to prevent blank screen on gen4 chipsets
Egbert Eich [Mon, 7 Jul 2014 16:20:34 +0000 (18:20 +0200)]
DRM/i915: Remove magic to prevent blank screen on gen4 chipsets

Since the root cause is understood now and with the fix

   commit 564ed191f5d816d24501664296991ec70327e2bc
   Author: Imre Deak <imre.deak@intel.com>
   Date:   Fri Jun 13 14:54:21 2014 +0300

       drm/i915: gmch: fix stuck primary plane due to memory self-refresh mode

in place the magic for G4x chipsets introduced with commit

   commit 61bc95c1fbbb6a08b55bbe161fdf1ea5493fc595
   Author: Egbert Eich <eich@suse.com>
   Date:   Mon Mar 4 09:24:38 2013 -0500

       DRM/i915: On G45 enable cursor plane briefly after enabling the display plane.

to avoided occasional screen blanking on mode changes can finally
be removed.
It's been verified that Imre's fix also resolves the said issue.

Signed-off-by: Egbert Eich <eich@suse.de>
Tested-by: Stefan Dirsch <sndirsch@suse.de>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Don't pretend ips is always enabled on BDW.
Rodrigo Vivi [Mon, 30 Jun 2014 11:45:01 +0000 (04:45 -0700)]
drm/i915: Don't pretend ips is always enabled on BDW.

As pointed out before we don't have a reliable way to read back ips
status on BDW without the risk to disable it when reading.
However now we are pretending that IPS on BDW is always on and getting
people confused about it.

So this patch allows people to know if ips was ever attempted to be enabled.
Even if the current status is impossible to be ascertain.

v2: (spotted by Paulo):
     * A version that at least compiles
     * with more clear messages
     * let Cheryview on the safe side until we aren't sure that checking ips
       state on ips won't disable it.

Cc: Paulo Zanoni <paulo.r.zanoni@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Reviewed-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Unpin last_context at reset
Ville Syrjälä [Wed, 18 Jun 2014 19:04:48 +0000 (22:04 +0300)]
drm/i915: Unpin last_context at reset

We're forgetting to unpin the last_context from the ggtt at GPU reset
time. This leads to the vma pin_count leaking at every reset if the
last context wasn't the ring default context. Further use of the same
context will trigger the pin_count check in i915_gem_object_pin() and
userspace will be faced with EBUSY as a result.

This plaques kms_flip rather badly since it performs lots of resets,
and every fd has its own default context these days.

Fix the problem by properly unpinning the last context at reset.

This regression seems to back to

commit acce9ffa4807027965ebd948456fa8385bbee32e
Author: Ben Widawsky <ben@bwidawsk.net>
Date:   Fri Dec 6 14:11:03 2013 -0800

   drm/i915: Better reset handling for contexts

Testcase: igt/gem_ctx_exec/reset-pin-leak
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: rework digital port IRQ handling (v2)
Dave Airlie [Wed, 18 Jun 2014 01:29:35 +0000 (11:29 +1000)]
drm/i915: rework digital port IRQ handling (v2)

The digital ports from Ironlake and up have the ability to distinguish
between long and short HPD pulses. Displayport 1.1 only uses the short
form to request link retraining usually, so we haven't really needed
support for it until now.

However with DP 1.2 MST we need to handle the short irqs on their
own outside the modesetting locking the long hpd's involve. This
patch adds the framework to distinguish between short/long to the
current code base, to lay the basis for future DP 1.2 MST work.

This should mean we get better bisectability in case of regression
due to the new irq handling.

v2: add GM45 support (untested, due to lack of hw)

Signed-off-by: Dave Airlie <airlied@redhat.com>
Reviewed-by: Todd Previte <tprevite@gmail.com>
[danvet: Fix conflicts in i915_irq.c with Oscar Mateo's irq handling
race fixes and a trivial one in intel_drv.h with the psr code.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: gmch: fix stuck primary plane due to memory self-refresh mode
Imre Deak [Fri, 13 Jun 2014 11:54:21 +0000 (14:54 +0300)]
drm/i915: gmch: fix stuck primary plane due to memory self-refresh mode

Blanking/unblanking the console in a loop on an Asus T100 sometimes
leaves the console blank. After some digging I found that applying

commit 61bc95c1fbbb6a08b55bbe161fdf1ea5493fc595
Author: Egbert Eich <eich@suse.com>
Date:   Mon Mar 4 09:24:38 2013 -0500

    DRM/i915: On G45 enable cursor plane briefly after enabling the display plane.

fixed VLV too.

In my case the problem seemed to happen already during the previous crtc
disabling and went away if I disabled self-refresh mode before disabling
the primary plane.

The root cause for this is that updates from the shadow to live plane
control register are blocked at vblank time if the memory self-refresh
mode (aka max-fifo mode on VLV) is active at that moment. The controller
checks at frame start time if the CPU is in C0 and the self-refresh mode
enable bit is set and if so activates self-reresh mode, otherwise
deactivates it. So to make sure that the plane truly gets disabled before
pipe-off we have to:

1. disable memory self-refresh mode
2. disable plane
3. wait for vblank
4. disable pipe
5. wait for pipe-off

v2:
- add explanation for the root cause from HW team (Cesar Mancini et al)
- remove note about the CPU C7S state, in my latest tests disabling it
  alone didn't make a difference
- add vblank between disabling plane and pipe (Ville)
- apply the same workaround for all gmch platforms (Ville)

Signed-off-by: Imre Deak <imre.deak@intel.com>
Reviewed-by: Vijay Purushothaman <vijay.a.purushothaman@intel.com>
Reviewed-by: Deepak S<deepak.s@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: gmch: set SR WMs to valid values before enabling them
Imre Deak [Fri, 13 Jun 2014 11:54:20 +0000 (14:54 +0300)]
drm/i915: gmch: set SR WMs to valid values before enabling them

Atm it's possible that we enable the memory self-refresh mode before the
watermark levels used by this mode are programmed with valid values. So
move the enabling after we programmed the WM levels.

Signed-off-by: Imre Deak <imre.deak@intel.com>
Reviewed-by: Vijay Purushothaman <vijay.a.purushothaman@intel.com>
Reviewed-by: Deepak S<deepak.s@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: gmch: factor out intel_set_memory_cxsr
Imre Deak [Tue, 1 Jul 2014 09:36:17 +0000 (12:36 +0300)]
drm/i915: gmch: factor out intel_set_memory_cxsr

This functionality will be also needed by an upcoming patch, so factor
it out. As a bonus this also makes things a bit more uniform across
platforms. Note that this also changes the register read-modify-write
to a simple write during disabling. This is what we do during enabling
anyway and according to the spec all the relevant bits are reserved-MBZ
or reserved with a 0 default value.

v2:
- unchanged
v3:
- fix missing cxsr disabling on pineview (Deepak)

Signed-off-by: Imre Deak <imre.deak@intel.com>
Reviewed-by: Deepak S <deepak.s@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Move VLV cmnlane workaround to intel_power_domains_init_hw()
Ville Syrjälä [Fri, 13 Jun 2014 10:37:56 +0000 (13:37 +0300)]
drm/i915: Move VLV cmnlane workaround to intel_power_domains_init_hw()

Now that the CMNRESET deassert is part of the cmnlane power well,
intel_reset_dpio() is called too late to make any difference. We've
deasserted CMNRESET by that time, and so the off+on toggle w/a will
never kick in.

Move the workaround to intel_power_domains_init_hw() where it gets
called before we enable the init power domain.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Pull the cmnlane tricks into its own power well ops
Ville Syrjälä [Fri, 13 Jun 2014 10:37:55 +0000 (13:37 +0300)]
drm/i915: Pull the cmnlane tricks into its own power well ops

Remove the clutter in __vlv_set_power_well() by moving the cmnlane
handling into custom enable/disable hooks for the cmnlane.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Kill duplicated cdclk readout code from i2c
Ville Syrjälä [Fri, 13 Jun 2014 10:37:54 +0000 (13:37 +0300)]
drm/i915: Kill duplicated cdclk readout code from i2c

We have a slightly different way of readoing out the cdclk in
gmbus_set_freq(). Kill that and just call .get_display_clock_speed().

Also need to remove the GMBUSFREQ update from intel_i2c_reset() since
that gets called way too early. Let's do it in intel_modeset_init_hw()
instead, and also pull the initial vlv_cdclk_freq update there from
init_clock gating.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Warn if there's a cdclk change in progess
Ville Syrjälä [Fri, 13 Jun 2014 10:37:53 +0000 (13:37 +0300)]
drm/i915: Warn if there's a cdclk change in progess

If someone is interested in the current cdclk frquency it should
be stable and not in process of changing frquency. Warn if the current
and requested cdclk don't match in .get_display_clock_spee() on vlv.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>