On pre-ddi platforms we don't shut down the link when changing link
training parameters. Except when clock recovery fails too hard and we
restart with channel eq training. Which doesn't make a lot of sense
really, since just stopping/restarting the DP port at this point
violates the modeset sequence documented in the Bspec.

So let's tempt fate and try this.

This patch is motivated by a WARN_ON triggered by

commit bc76e320
Author: Daniel Vetter <daniel.vetter@ffwll.ch>
Date:   Tue May 20 22:46:50 2014 +0200

    drm/i915: Drop now misleading DDI comment from dp_link_down

References: https://bugs.freedesktop.org/show_bug.cgi?id=85670Signed-off-by: NDaniel Vetter <daniel.vetter@intel.com>
Acked-by: NJani Nikula <jani.nikula@linux.intel.com>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

Chris removed the code using it in:

  commit be2d599b
  Author: Chris Wilson <chris@chris-wilson.co.uk>
  Date:   Wed Sep 10 19:52:18 2014 +0100

      drm/i915: Remove dead code, i915_gem_verify_gtt
Signed-off-by: NDamien Lespiau <damien.lespiau@intel.com>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

As Paulo said when introducing the enum, having more types is really
good to document what should go where (int foo(int, int, bool, bool).

Cc: Paulo Zanoni <paulo.r.zanoni@intel.com>
Signed-off-by: NDamien Lespiau <damien.lespiau@intel.com>
Reviewed-by: NJani Nikula <jani.nikula@intel.com>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

It's really part of the "push all new_* state into current state
pointers" done in that function. So let's move it there to make this
clear.

Also, with the conversion done the num_shared_dpll check the function
does in it's loop is enough, so we can drop the check for the dpll
compute callback, too.

Cc: Ander Conselvan de Oliveira <ander.conselvan.de.oliveira@intel.com>
Signed-off-by: NDaniel Vetter <daniel.vetter@intel.com>

Now that shared DPLLs configuration is staged, there's no need to track
the current ones in the new pipe_config since those are released before
making the new pipe_config effective.
Signed-off-by: NAnder Conselvan de Oliveira <ander.conselvan.de.oliveira@intel.com>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

There's no users left after the conversion to calculate clocks before
disabling crtcs during mode set.
Signed-off-by: NAnder Conselvan de Oliveira <ander.conselvan.de.oliveira@intel.com>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

Use the infrastructure added in a previous patch to choose shared DPLLs
and calculate clocks before touching the hardware.

v2: Don't set mode_set hooks since dev_priv is kzalloc()'d (Ville)
Signed-off-by: NAnder Conselvan de Oliveira <ander.conselvan.de.oliveira@intel.com>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

Use the infrastructure added in a previous patch to choose shared DPLLs
and calculate clocks before touching the hardware.

v2: Don't set mode_set hooks since dev_priv is kzalloc()'d (Ville)
Signed-off-by: NAnder Conselvan de Oliveira <ander.conselvan.de.oliveira@intel.com>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

Use the infrastructure added in a previous patch to choose shared DPLLs
and calculate clocks before touching the hardware.

v2: Don't set mode_set hooks since dev_priv is kzalloc()'d (Ville)
Signed-off-by: NAnder Conselvan de Oliveira <ander.conselvan.de.oliveira@intel.com>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

It is possible for a mode set to fail if there aren't shared DPLLS that
match the new configuration requirement or other errors in clock
computation. If that step is executed after disabling crtcs, in the
failure case the hardware configuration is changed and needs to be
restored. Doing those things early will allow the mode set to fail
before actually touching the hardware.

Follow up patches will convert different platforms to use the new
infrastructure.

v2: Keep pll->new_config valid only during mode set (Ville)
    Use kmemdup() in i915_shared_dpll_start_config() (Ville)
    Restore old pll config if something fails before commit (Ville)
    Don't set compute_clock hooks since dev_priv is kzalloc()'d (Ville)
Signed-off-by: NAnder Conselvan de Oliveira <ander.conselvan.de.oliveira@intel.com>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

The new struct will be used in a follow up patch to allow a current and
a staged config to exist for the same shared DPLL.

v2: Rebase on by mask_to_refcount()->hweight32() change. (Damien)
Signed-off-by: NAnder Conselvan de Oliveira <ander.conselvan.de.oliveira@intel.com>
Reviewed-by: NDamien Lespiau <damien.lespiau@intel.com>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

This will be used in a follow up patch to properly release shared DPLLs
without relying on the shared_dpll field in pipe_config.

v2: Fix white space error (Ville)
    Use hweight32() (Ville)
Signed-off-by: NAnder Conselvan de Oliveira <ander.conselvan.de.oliveira@intel.com>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

More concise. Noticed while reviewing Ander's patch which touched a
lot of the pipe_has_type checks.

v2: Use new_config in one place Ander spotted.

Cc: Ander Conselvan de Oliveira <conselvan2@gmail.com>
Signed-off-by: NDaniel Vetter <daniel.vetter@intel.com>

This shouldn't change the behavior of those functions, since they are
called after the new_config is made effective and that points to the
current config. In a follow up patch, the mode set sequence will be
changed so this is called before disabling crtcs, and in that case
those functions should work on the staged config.
Signed-off-by: NAnder Conselvan de Oliveira <ander.conselvan.de.oliveira@intel.com>
[danvet: Flatten if by moving the check into the WARN.]
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

execlists_submit_context() always returns 0, which is redundant.
And its name is inaccurate, since it actually submits (up to)
TWO contextS. So we rename it, change it to "void", and remove
the WARN_ON() testing its return value.

Change-Id: Ie225b0eca7754c6093c8b8bd15550b251b6feb82
Signed-off-by: NDave Gordon <david.s.gordon@intel.com>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

If a ring failed to initialise for any reason then the error path would try to
clean up all rings including those that had not yet been allocated. The ring
clean up code did a check that the ring was valid before starting its work.
Unfortunately, that was after it had already dereferenced the ring to obtain a
dev_private pointer.
Signed-off-by: NJohn Harrison <John.C.Harrison@Intel.com>
Reviewed-by: NDamien Lespiau <damien.lespiau@intel.com>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

When looking at the bug report logs with triggered
WARN_ON, the person doing bug triaging will have to
find exact kernel source and match file/line.

Attach the condition that triggered the WARN_ON
to kernel log. In most cases the context is self
evident and this way we can save developer time.

The drawback is ~16kbytes bigger i915.ko
Signed-off-by: NMika Kuoppala <miku@iki.fi>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

When we suspend we turn everything off so the pps should be idle, and we
also (or at least should) disable all power wells which will reset the
power sequencer port assignment. So when we resume all power sequencers
should be in their reset state. However it's at least theoretically
possible that the BIOS would touch the power seuqencer(s), so to be safe
we ought to read out the current port assignment like we do at driver
init time.

To do that we can simply call vlv_initial_power_sequencer_setup() from
the encoder ->reset() hook before calling intel_edp_panel_vdd_sanitize().
There's no danger or clobbering the pps delays since we now have those
stored within intel_dp and we don't change them once initialized.

This will make sure that the vdd state gets correctly tracked post-resume
in case the BIOS enabled it.

We need to shuffle things around a bit to get the locking right, and
while at it, make intel_edp_panel_vdd_sanitize() static and move it
around a bit to avoid a forward declaration.

Cc: Imre Deak <imre.deak@intel.com>
Signed-off-by: NVille Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: NImre Deak <imre.deak@intel.com>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

The pps timestamp initialization was accidentally lost on vlv/chv in

 commit a4a5d2f8
 Author: Ville Syrjälä <ville.syrjala@linux.intel.com>
 Date:   Thu Sep 4 14:54:20 2014 +0300

    drm/i915: Track which port is using which pipe's power sequencer

Restore it so that we avoid introducing random delays into the pps operations
during/after driver init time.

Cc: Imre Deak <imre.deak@intel.com>
Reported-by: NImre Deak <imre.deak@intel.com>
Signed-off-by: NVille Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: NImre Deak <imre.deak@intel.com>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

We'll never end up in the hooks with eld[0] unset, as that's checked by
drm_select_eld().
Signed-off-by: NJani Nikula <jani.nikula@intel.com>
Reviewed-by: NRodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

Introduce functions to enable/disable the audio codec, incorporating the
ELD setup within enable. The disable is initially limited to HSW,
covering exactly what was done previously.

The only functional difference is that ELD valid is no longer set if
there is no connector with ELD, which should be the right thing to do
anyway. Otherwise the sequence remains the same, with warts and all, in
preparation for applying more sanity.

v2: add kernel doc.
Signed-off-by: NJani Nikula <jani.nikula@intel.com>
Reviewed-by: NRodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

The audio programming sequence states that the ELD must be written and
enabled after the pipe is ready. Indeed, this should clarify the
situation with

commit c7905792
Author: Daniel Vetter <daniel.vetter@ffwll.ch>
Date:   Wed Apr 16 16:56:09 2014 +0200

    drm/i915: Remove vblank wait from haswell_write_eld

and Ville's review of it [1].

Moreover, we should not touch the relevant registers before we get the
audio power domain.

[1] http://mid.gmane.org/20140416155309.GK18465@intel.comSigned-off-by: NJani Nikula <jani.nikula@intel.com>
Reviewed-by: NRodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

Keep the driver modifications to ELD together. This also sets the
Conn_Type for G4X DP which wasn't done before.

Clean up the debugs while at it; this is all obvious from the connector
name.

v3: add missing ~ (Rodrigo)
Signed-off-by: NJani Nikula <jani.nikula@intel.com>
Reviewed-by: NRodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

It seems that the pipe-a power well has replaced the disp2d power well
on chv. At least that's the case with the current punit firmware. So
enable the pipe-a power and expand its domains to cover everything the
disp2d well ought to cover.

The other power wells (apart from the cmnlane wells) still seem awol
in the current punit firmware. So leave them disabled in the code.

This fixes a hilarious oops during resume on bsw where
intel_hdmi_get_config() would read the port register and get back
0xffffffff and thus think the port is enabled on pipe D. It would then
go and index the pipe_to_crtc_mapping[] array with PIPE_D and blow up
when intel_hdmi_get_config() tries to write to crtc->config. Someone
really ought to replace all naked pipe_to_crtc_mapping[] uses with the
appropriate function call so we could add a warning there if the pipe
doesn't actually exist...

We must also call the power seqeuencer state reset function from
the pipe-a well disable just like we do from disp2d on vlv. Otherwise
the eDP panel won't recover at resume time since the PPS has lost its
hold on the port.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=84903Signed-off-by: NVille Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: NJesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

CHV has a programmable CSC unit on the pipe B sprites. Program the unit
appropriately for BT.601 limited range YCbCr to full range RGB color
conversion. This matches the programming we currently do for sprites
on the other pipes and on other platforms.

It seems the CSC only works when the input data is YCbCr. For RGB
pixel formats it doesn't matter what we program into the CSC registers.
Doesn't make much sense to me especially since the register names give
the impression that RGB input data would also work. But that's how
it behaves here.

In the review discussions there's been some nice math to explain the
values obtained here. First about the YCbCr->RGB matrix:

"I had the RGB->YCbCr matrix, inverted it and the values came out. But they
should match the wikipedia article. Also keep in mind that the coefficients
are in .12 in fixed point format, hence we need a 1<<12 factor. So let's
try it:

Kb=.114
Kr=.299
(1<<12) * 255/219 ~= 4769
-(1<<12) * 255/112*(1-Kb)*Kb/(1-Kb-Kr) ~= -1605
-(1<<12) * 255/112*(1-Kr)*Kr/(1-Kb-Kr) ~= -3330
(1<<12) * 255/112*(1-Kr) ~= 6537
(1<<12) * 255/112*(1-Kb) ~= 8263

"Looks like the same values to me."

And then about the limits used for clamping:

"> where did you get these min/max?

"The hardware apparently deals in 10bit values, so we need to multiply everything
by 4 when we start with the 8bit min/max values.

Y = [16:235] * 4 = [64:940]
CbCr = ([16:240] - 128) * 4 = [-112:112] * 4 = [-448:448]

"The -128 being the -0.5 bias that the hardware already applied before
the data entered the CSC unit."

Raw data is also supplied in 10bpc in the registers.
Signed-off-by: NVille Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by Rodrigo Vivi <rodrigo.vivi@intel.com>
[danvet: Copypaste explanations&math from the review discussion.]
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

We want to run intel_uncore_early_sanitize() before we touch any
registers, because on BDW, when we resume, the FPGA_DBG_RM_NOCLAIM bit
is set, so we need to clear it - through intel_uncore_early_sanitize()
- before we do anything else. With the current code, we don't clear
the bit before our first register access, so we print a WARN
complaining about an unclaimed register error.

v1: Was called "drm/i915: run intel_uncore_early_sanitize earlier on
resume"
v2: Was called "drm/i915: run intel_uncore_early_sanitize earlier on
resume on non-VLV"
v3: This one, on top of the intel_resume_prepare() rework.
v4: Rebase.
Signed-off-by: NPaulo Zanoni <paulo.r.zanoni@intel.com>
Reviewed-by: NImre Deak <imre.deak@intel.com>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

Because, really, the abstraction is not working for us. It is nice for
VLV, but doesn't add anything useful on SNB/HSW/BDW. We want to change
this code due to a recently-discovered bug, but we can't seem to find
a nice solution that repects the current abstraction. So let's kill
intel_resume_prepare() and its friends, and add an equivalent
implementation to both its callers.

Also, look at the diffstat!

v2: - Rebase.
Signed-off-by: NPaulo Zanoni <paulo.r.zanoni@intel.com>
Reviewed-by: NImre Deak <imre.deak@intel.com>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

CHV adds a bunch of new registers for primary plane size/position and
pipe blender setup. Initialize all those registers to avoid nasty
surprises. PRIMSIZE is especially important as without programming it
the outout will be garbled whenever the primary plane size would not
match what the BIOS set up.

Also program the sprite constant alpha register to disable the constant
alpha blending factor. This applies to vlv as well as chv.
Signed-off-by: NVille Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: NRodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

In case the cmnlane power well is down but cmnreset isn't asserted we
would currently skip the off+on toggle for the power well. That could
leave cmnreset deasserted while cmnlane is powered down which might
lead to problems with the PHY.

To avoid such issues skip the cmnlane toggle only if both cmnlane and
disp2d wells are up and cmnreset is already deasserted. In all other
cases power down the cmnlane well which will also make sure cmnreset
gets asserted correctly while cmnlane is powered down.
Signed-off-by: NVille Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: NJesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

Use the macros makes the code cleaner and it also checks for a NULL fb.
Signed-off-by: NGustavo Padovan <gustavo.padovan@collabora.co.uk>
Reviewed-by: NVille Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

take out pin_fb code so the commit phase can't fail anymore.
Signed-off-by: NGustavo Padovan <gustavo.padovan@collabora.co.uk>
Reviewed-by: NVille Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

Take out the pin_fb code so commit phase can't fail anymore.
Signed-off-by: NGustavo Padovan <gustavo.padovan@collabora.co.uk>
Reviewed-by: NVille Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

The power seqeuencer kick procedure requires the DPLL to be running
in order to complete successfully. In case the DPLL isn't currently
running when we need to kick the power seqeuncer enable it
temporarily. This can happen eg. during ->detect() when the pipe is
not already active.

To avoid needlessly duplicating the DPLL programming re-use the already
existing functions by passing a temporary pipe config to them instead
of having them consult the current pipe config at crtc->config.

v2: Introduce vlv_force_pll_{on,off}() (Daniel)
v3: Rebase due to drm_crtc vs. intel_crtc changes
    Fix a typo in commit msg (checkpatch)
Signed-off-by: NVille Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Imre Deak <imre.deak@intel.com> (v1)
[danvet: Appease checkpatch.]
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

eDP ports need the power seqeuncer whenever the port is active. Warn if
we accidentally steal the power sequener from an active eDP port. This
should not happen unless there's a bug somewhere else, but it's best to
scream loudly if it happens to help with debugging.

Note that this only checks for active pipes and not for enabled pipes
which are turned off with dpms. Which means we might run the risk that
the pps might get stolen and we can't reacquire one when enabling the
pipe again with dpms on. But on current platforms that's impossible
since we only support two edp ports with just two panel power
sequencers. So a more elaborate scheme which reserves the pps even
when the pipe is inactive isn't required.
Signed-off-by: NVille Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: NImre Deak <imre.deak@intel.com>
[danvet: Summarize my discussion with Ville about dpms on/off issues.]
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

We should never enable the panel power twice. That would indicate a bug
somewhere else as we would need to enable the port twice without
disabling it in between. Also print the port name.
Signed-off-by: NVille Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: NImre Deak <imre.deak@intel.com>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

Print the port name in the VDD/PPS debugs messages.
Signed-off-by: NVille Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: NImre Deak <imre.deak@intel.com>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

In case we fumble something and end up picking an already used power
seqeuencer in vlv_power_sequencer_pipe() at least try to steal it
gracefully. In theory this should never happen though.
Signed-off-by: NVille Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: NImre Deak <imre.deak@intel.com>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

There's no power sequencer on pipe C on VLV/CHV so scream a bit if we
try to steal one from pipes other than A and B.
Signed-off-by: NVille Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: NImre Deak <imre.deak@intel.com>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

VLV gets confused if two power sequencers have the same port selected.
It would seem the port doesn't start up properly in the is case and
vlv_wait_port_ready() will fail as will the link training. Clearing the
port select in the PP_ON_DELAYS register fixes this problem.

CHV doesn't seem to need this, but it doesn't seem to hurt either so
let's just do it for both to keep the code between the platforms as
uniform as possible.
Signed-off-by: NVille Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: NImre Deak <imre.deak@intel.com>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

If there's no power sequencer assigned to the port currently we can't
very well have vdd or panel power enabled either. If we would try to
check that from the pps registers we'd need to pick a power seqeuncer
and kick it. So let's skip the register read and the kick.

Note that there's still a bit an issue about correctly recovering pps
state from resume if the bios is nasty: With this check we'll always
assume that the pps is off. But that's better done in a follow-up
patch and it shouldn't be too harmful - at most we waste time enabling
the pps if it's on already.
Signed-off-by: NVille Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: NImre Deak <imre.deak@intel.com>
[danvet: Add note about resume issues Imre spotted.]
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>