On the commit 3301d409 ("drm/i915: PSR: Fix DP_PSR_NO_TRAIN_ON_EXIT logic")'
we already had identified that DP_PSR_NO_TRAIN_ON_EXIT
doesn't mean we shouldn't send TPS patterns, however we start sending the
minimal TP1 as possible and no TP2.

For most of the panels this is ok, but we found a reported case where
this is not true and panel keeps frozen without updating the screen for a while.

We could just get this case after patch "PSR: Don't Skip aux handshake on
DP_PSR_NO_TRAIN_ON_EXIT." is applied since that one fix the
hard freeze on this kind of panels.

Reference: https://bugs.freedesktop.org/show_bug.cgi?id=91436#c19

Cc: Ivan Mitev <ivan.mitev@gmail.com>
Signed-off-by: NRodrigo Vivi <rodrigo.vivi@intel.com>
Reviewed-by: NDurgadoss R <durgadoss.r@intel.com>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

Since the beginning there is a confusion on the meaning of this bit.

A previous patch had identified this already and fixed it partially:
'commit 3301d409 ("drm/i915: PSR: Fix DP_PSR_NO_TRAIN_ON_EXIT logic")

DP_PSR_NO_TRAIN_ON_EXIT means the source doesn't need to do the
training, but it doesn't tell to avoid TP patterns or to skip
aux handshake.

This patch fixes the hard freeze reported.

Reference: https://bugs.freedesktop.org/show_bug.cgi?id=91436
Reference: https://bugs.freedesktop.org/show_bug.cgi?id=91437

Cc: Ivan Mitev <ivan.mitev@gmail.com>
Signed-off-by: NRodrigo Vivi <rodrigo.vivi@intel.com>
Reviewed-by: NDurgadoss R <durgadoss.r@intel.com>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

With 'commit 30886c5a ("drm/i915: VLV/CHV PSR: Increase wait delay
 time before active PSR.")' we fixed a blank screen when first
activation was happening immediately after PSR being enabled.
There we gave more time for idleness by increasing the delay
between re-activating sequences.

However, commit "drm/i915: Delay first PSR activation."
delay the first activation in a better way keeping a good PSR
residency. So, we can now reduce the delay on re-enable.
Signed-off-by: NRodrigo Vivi <rodrigo.vivi@intel.com>
Reviewed-by: NDurgadoss R <durgadoss.r@intel.com>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

When debuging the frozen screen caused by HW tracking with low
power state I noticed that if we keep moving the mouse non stop
you will miss the screen updates for a while. At least
until we stop moving the mouse for a small time and move again.

The actual enabling should happen immediately after
Display Port enabling sequence finished with links trained and
everything enabled. However we face many issues when enabling PSR
right after a modeset.

On VLV/CHV we face blank screens on this scenario and on HSW+
we face a recoverable frozen screen, at least until next
exit-activate sequence.

Another workaround for the same issue here would be to increase
re-enable idle time from 100 to 500 as we did for VLV/CHV.
However this patch workaround this issue in a better
way since it doesn't reduce PSR residency and also
allow us to reduce the delay time between re-enables at least
on VLV/CHV.

This is also important to make the sysfs toggle working properly.
Signed-off-by: NRodrigo Vivi <rodrigo.vivi@intel.com>
Reviewed-by: NDurgadoss R <durgadoss.r@intel.com>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

Make I915_READ and I915_WRITE more type safe by wrapping the register
offset in a struct. This should eliminate most of the fumbles we've had
with misplaced parens.

This only takes care of normal mmio registers. We could extend the idea
to other register types and define each with its own struct. That way
you wouldn't be able to accidentally pass the wrong thing to a specific
register access function.

The gpio_reg setup is probably the ugliest thing left. But I figure I'd
just leave it for now, and wait for some divine inspiration to strike
before making it nice.

As for the generated code, it's actually a bit better sometimes. Eg.
looking at i915_irq_handler(), we can see the following change:
  lea    0x70024(%rdx,%rax,1),%r9d
  mov    $0x1,%edx
- movslq %r9d,%r9
- mov    %r9,%rsi
- mov    %r9,-0x58(%rbp)
- callq  *0xd8(%rbx)
+ mov    %r9d,%esi
+ mov    %r9d,-0x48(%rbp)
 callq  *0xd8(%rbx)

So previously gcc thought the register offset might be signed and
decided to sign extend it, just in case. The rest appears to be
mostly just minor shuffling of instructions.

v2: i915_mmio_reg_{offset,equal,valid}() helpers added
    s/_REG/_MMIO/ in the register defines
    mo more switch statements left to worry about
    ring_emit stuff got sorted in a prep patch
    cmd parser, lrc context and w/a batch buildup also in prep patch
    vgpu stuff cleaned up and moved to a prep patch
    all other unrelated changes split out
v3: Rebased due to BXT DSI/BLC, MOCS, etc.
v4: Rebased due to churn, s/i915_mmio_reg_t/i915_reg_t/
Signed-off-by: NVille Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: NChris Wilson <chris@chris-wilson.co.uk>
Link: http://patchwork.freedesktop.org/patch/msgid/1447853606-2751-1-git-send-email-ville.syrjala@linux.intel.com

Signed-off-by: NVille Syrjälä <ville.syrjala@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1446672017-24497-29-git-send-email-ville.syrjala@linux.intel.comReviewed-by: NChris Wilson <chris@chris-wilson.co.uk>

Add 'u32 offset' to the uncore register access functions. For now
it's the same as 'reg', but once type safety gets added 'reg' will be
the type safe register variable and 'offset' the raw offset.

v2: s/uint32_t/u32/ (Chris)

Cc: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: NVille Syrjälä <ville.syrjala@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1446839236-20035-1-git-send-email-ville.syrjala@linux.intel.comReviewed-by: NChris Wilson <chris@chris-wilson.co.uk>

I need to add a new variable into GEN6_{READ,WRITE}_HEADER, but the vgpu
won't need it, so let's avoid an unused variable warning by splitting
the vgpu stuff to use its own macros.

Cc: Eddie Dong <eddie.dong@intel.com>
Cc: Jike Song <jike.song@intel.com>
Cc: Kevin Tian <kevin.tian@intel.com>
Cc: Yu Zhang <yu.c.zhang@linux.intel.com>
Cc: Zhi Wang <zhi.a.wang@intel.com>
Cc: Zhiyuan Lv <zhiyuan.lv@intel.com>
Signed-off-by: NVille Syrjälä <ville.syrjala@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1446672017-24497-26-git-send-email-ville.syrjala@linux.intel.comReviewed-by: NZhiyuan Lv <zhiyuan.lv@intel.com>

We'll want to avoid performing arithmetic with register offsets, so
instead calculating the vgpu PDP as pdp0_lo+offset, make the PDPs
into an array. This way we can simply loop through them.

Cc: Eddie Dong <eddie.dong@intel.com>
Cc: Jike Song <jike.song@intel.com>
Cc: Kevin Tian <kevin.tian@intel.com>
Cc: Yu Zhang <yu.c.zhang@linux.intel.com>
Cc: Zhi Wang <zhi.a.wang@intel.com>
Cc: Zhiyuan Lv <zhiyuan.lv@intel.com>
Signed-off-by: NVille Syrjälä <ville.syrjala@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1446672017-24497-25-git-send-email-ville.syrjala@linux.intel.comReviewed-by: NZhiyuan Lv <zhiyuan.lv@intel.com>

We set up a load of LRIs in the logical ring context. Wrap that stuff
in a macro to avoid typos with position of each reg/value pair in the
context. This also makes it easier to make the register defines type
safe.
Signed-off-by: NVille Syrjälä <ville.syrjala@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1446672017-24497-24-git-send-email-ville.syrjala@linux.intel.comReviewed-by: NChris Wilson <chris@chris-wilson.co.uk>

The logical render context population has a bunch of raw ring register
offsets. Use the names we have for them, and in cases where we we don't,
give them names.
Signed-off-by: NVille Syrjälä <ville.syrjala@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1446672017-24497-23-git-send-email-ville.syrjala@linux.intel.comReviewed-by: NChris Wilson <chris@chris-wilson.co.uk>

Signed-off-by: NVille Syrjälä <ville.syrjala@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1446672017-24497-22-git-send-email-ville.syrjala@linux.intel.comReviewed-by: NChris Wilson <chris@chris-wilson.co.uk>

Add a helper for emitting register offsets (for LRI/SRM) into the w/a
batch buffer.
Signed-off-by: NVille Syrjälä <ville.syrjala@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1446672017-24497-21-git-send-email-ville.syrjala@linux.intel.comReviewed-by: NChris Wilson <chris@chris-wilson.co.uk>

When register type safety happens, we can't just try to emit the
register itself to the ring. Instead we'll need to extract the
offset from it first. Add some convenience functions that will do
that.

v2: Convert MOCS setup too
Signed-off-by: NVille Syrjälä <ville.syrjala@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1446672017-24497-20-git-send-email-ville.syrjala@linux.intel.comReviewed-by: NChris Wilson <chris@chris-wilson.co.uk>

Add defines for the upper halves of the registers used by the cmd
parser. Getting rid of the arithmetic with the register offset
will help in making registers type safe.

v2: s/_HI/_UDW/ (Chris)
Signed-off-by: NVille Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: NChris Wilson <chris@chris-wilson.co.uk>
Link: http://patchwork.freedesktop.org/patch/msgid/1446839080-18732-1-git-send-email-ville.syrjala@linux.intel.com

Store the upper dword of the register offset in the whitelist as well.
This would allow it to read register where the two halves aren't sitting
right next to each other, and it'll make it easier to make register
access type safe.

While at it change the register offsets to u32 from u64. Our register
space isn't quite that big, yet :)

v2: Use ldw/udw as the suffixes, and add a note about
    64bit wide split regs (Chris)

Cc: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: NVille Syrjälä <ville.syrjala@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1446839021-18599-1-git-send-email-ville.syrjala@linux.intel.comReviewed-by: NChris Wilson <chris@chris-wilson.co.uk>

Use the RING_PSMI_CTL define insted of hand rolling the register offset.
Signed-off-by: NVille Syrjälä <ville.syrjala@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1446672017-24497-17-git-send-email-ville.syrjala@linux.intel.comReviewed-by: NChris Wilson <chris@chris-wilson.co.uk>

v2: Use for_each_ring() (Chris)

Cc: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Francisco Jerez <currojerez@riseup.net>
Signed-off-by: NVille Syrjälä <ville.syrjala@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1446725633-6419-1-git-send-email-ville.syrjala@linux.intel.comReviewed-by: NChris Wilson <chris@chris-wilson.co.uk>

Signed-off-by: NVille Syrjälä <ville.syrjala@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1446672017-24497-15-git-send-email-ville.syrjala@linux.intel.comReviewed-by: NChris Wilson <chris@chris-wilson.co.uk>

Most of our register defines follow the convention that if there's a
need for the raw register offset, that one has an underscore sa a
prefix. The define (possibly parametrized) without the underscore is
the one people should normally use, since it will take into account
all the parameters and other potential offsets that are needed.

Fix up the few stragglers that don't follow this convention.
Signed-off-by: NVille Syrjälä <ville.syrjala@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1446672017-24497-14-git-send-email-ville.syrjala@linux.intel.comReviewed-by: NChris Wilson <chris@chris-wilson.co.uk>

If we ignore the BXT situation, we can observe that the only variables
affecting gpio_mmio_base is IS_VALLEVIEW and HAS_GMCH_DISPLAY. The BXT
situation we can fit into the same pattern if we change gmbus_pins_bxt[]
to house the GMCH GPIO register offsets (like we do for all other
platfotms). So let's do that.

We could even simplify the VLV situation more by including the
display_mmio_offset in the GPIO register defines, but let's leave it be
for now.
Signed-off-by: NVille Syrjälä <ville.syrjala@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1446672017-24497-13-git-send-email-ville.syrjala@linux.intel.comReviewed-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

Store the DVO SRCDIM register offset alongside the DVO control register
offset in intel_dvo_device. This gets rid of the switch statement whose
case values are the DVO control register offsets. Such a construct would
cause problems for register type safety.
Signed-off-by: NVille Syrjälä <ville.syrjala@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1446672017-24497-12-git-send-email-ville.syrjala@linux.intel.comReviewed-by: NChris Wilson <chris@chris-wilson.co.uk>

Replace the is_sdvob bool and some sdvo_reg checks with enum port. This
makes the SDVO code look more modern, and gets rid of explicit register
offset checks in the code which will hamper register type checking.

v2: Add assert_sdvo_port_valid() (Chris)
Signed-off-by: NVille Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: NChris Wilson <chris@chris-wilson.co.uk>
Link: http://patchwork.freedesktop.org/patch/msgid/1446838199-3666-1-git-send-email-ville.syrjala@linux.intel.com

Signed-off-by: NVille Syrjälä <ville.syrjala@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1446672017-24497-9-git-send-email-ville.syrjala@linux.intel.comReviewed-by: NChris Wilson <chris@chris-wilson.co.uk>

i915 register defines are going to become type safe, so going forward
the register defines can't be used as straight numbers. Since quirks.c
needs just a few extra register defines from i915_reg.h, decouple the
two by defining the required registers locally in quirks.c. This was
already done for a few other igpu related registers.

Cc: Bjorn Helgaas <bhelgaas@google.com>
Cc: linux-pci@vger.kernel.org
Cc: linux-kernel@vger.kernel.org
Signed-off-by: NVille Syrjälä <ville.syrjala@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1446672017-24497-2-git-send-email-ville.syrjala@linux.intel.comReviewed-by: NChris Wilson <chris@chris-wilson.co.uk>

Signed-off-by: NRodrigo Vivi <rodrigo.vivi@intel.com>
Reviewed-by: NPaulo Zanoni <paulo.r.zanoni@intel.com>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

It was created at 'commit aabc95dc (drm/i915: Dont -ETIMEDOUT
on identical new and previous (count, crc).")' becase the counter
wasn't reliable.

Now that we properly wait for the counter to be reset we can rely
a bit more in the counter.

Also that patch stopped to return -ETIMEDOUT so the test case is
unable to skip when it is unreliable and end up in many fails
that should be skip instead.

So, with the counter more reliable we can remove
this hack that just makes things more confusing when test cases
are really expecting the same CRC and let test case skip if that's
not the case.
Signed-off-by: NRodrigo Vivi <rodrigo.vivi@intel.com>
Reviewed-by: NPaulo Zanoni <paulo.r.zanoni@intel.com>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

According to VESA DP spec TEST_CRC_COUNT (Bits 3:0) at
TEST_SINK_MISC (00246h) is "Reset to 0 when TEST_SINK bit 0 = 0;

So let's give few vblanks so we are really sure that this counter
is really zeroed on the next sink_crc read.
Signed-off-by: NRodrigo Vivi <rodrigo.vivi@intel.com>
Reviewed-by: NPaulo Zanoni <paulo.r.zanoni@intel.com>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

According to VESA DP Spec, setting TEST_SINK_START (bit 0)
of TEST_SINK (00270h) "Stop/Start calculating CRC on the next frame"

So let's wait at least 1 vblank to really say the calculation
stopped or started.
Signed-off-by: NRodrigo Vivi <rodrigo.vivi@intel.com>
Reviewed-by: NPaulo Zanoni <paulo.r.zanoni@intel.com>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

Signed-off-by: NPatrik Jakobsson <patrik.jakobsson@linux.intel.com>
Reviewed-by: NImre Deak <imre.deak@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1447084107-8521-13-git-send-email-patrik.jakobsson@linux.intel.com

v2: Use _unsafe (Jani)
v3: Allow specifying specific DC-states instead of just DC6 (Imre)
Signed-off-by: NPatrik Jakobsson <patrik.jakobsson@linux.intel.com>
Reviewed-by: NImre Deak <imre.deak@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1447682467-6237-3-git-send-email-patrik.jakobsson@linux.intel.com

Handle DC off as a power well where enabling the power well will prevent
the DMC to enter selected DC states (required around modesets and Aux
A). Disabling the power well will allow DC states again. For now the
highest DC state is DC6 for Skylake and DC5 for Broxton but will be
configurable for Skylake in a later patch.

v2: Check both DC5 and DC6 bits in power well enabled function (Ville)
v3:
- Remove unneeded DC_OFF case in skl_set_power_well() (Imre)
- Add PW2 dependency to DC_OFF (Imre)
v4: Put DC_OFF before PW2 in BXT power well array
Signed-off-by: NPatrik Jakobsson <patrik.jakobsson@linux.intel.com>
Reviewed-by: NImre Deak <imre.deak@intel.com>
[fixed line over 80 and parenthesis alignment checkpatch warns (imre)]
Link: http://patchwork.freedesktop.org/patch/msgid/1447687201-24759-1-git-send-email-patrik.jakobsson@linux.intel.com

v2: Add explanation of the fixed power well bits (Imre)
Signed-off-by: NPatrik Jakobsson <patrik.jakobsson@linux.intel.com>
Reviewed-by: NImre Deak <imre.deak@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1447682467-6237-2-git-send-email-patrik.jakobsson@linux.intel.com

PG2 enabled is not a requirement for disabling DC5. It's just one
of the reasons why the DMC wouldn't enter DC5. During modeset we don't
care about PG2 from a DC perspective, only the fact that DC5/DC6 is not
allowed.
Signed-off-by: NPatrik Jakobsson <patrik.jakobsson@linux.intel.com>
Reviewed-by: NVille Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: NImre Deak <imre.deak@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1447084107-8521-9-git-send-email-patrik.jakobsson@linux.intel.com

We need a power domain for disabling DC5/DC6 around modesets to prevent
confusing the DMC.
Signed-off-by: NPatrik Jakobsson <patrik.jakobsson@linux.intel.com>
Reviewed-by: NImre Deak <imre.deak@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1447084107-8521-8-git-send-email-patrik.jakobsson@linux.intel.com

We never make use of the distinction between 2 vs 4 lanes so combine
them into a per port domain instead. This saves us a few bits in the
power domain mask. Change suggested by Ville.
Signed-off-by: NPatrik Jakobsson <patrik.jakobsson@linux.intel.com>
Reviewed-by: NImre Deak <imre.deak@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1447084107-8521-7-git-send-email-patrik.jakobsson@linux.intel.com

All the DDI power domains are already excluded from
SKL_DISPLAY_ALWAYS_ON_POWER_DOMAINS on account of
excluding SKL_DISPLAY_POWERWELL_1_POWER_DOMAINS and
SKL_DISPLAY_POWERWELL_2_POWER_DOMAINS, no need to spell them out again.
Signed-off-by: NVille Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: NPatrik Jakobsson <patrik.jakobsson@linux.intel.com>
Signed-off-by: NImre Deak <imre.deak@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1447084107-8521-6-git-send-email-patrik.jakobsson@linux.intel.com

Currently the gmbus code uses intel_aux_display_runtime_get/put in an
effort to make sure the hardware is powered up sufficiently for gmbus.
That function only takes the runtime PM reference which on VLV/CHV/BXT
is not enough. We need the disp2d/pipe-a well on VLV/CHV and power well
2 on BXT. So add a new power domnain for gmbus and kill off the now
unused intel_aux_display_runtime_get/put. And change
intel_hdmi_set_edid() to use the gmbus power domain too since that's all
we need there.

Also toss in a BUILD_BUG_ON() to catch problems if we run out of
bits for power domains. We're already really close to the limit...

[Patrik: Add gmbus string to debugfs output]
Signed-off-by: NVille Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: NPatrik Jakobsson <patrik.jakobsson@linux.intel.com>
Signed-off-by: NImre Deak <imre.deak@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1447084107-8521-5-git-send-email-patrik.jakobsson@linux.intel.com

Introduce intel_display_port_aux_power_domain() which simply returns
the appropriate AUX power domain for a specific port, and then replace
the intel_display_port_power_domain() with calls to the new function
in the DP code. As long as we're not actually enabling the port we don't
need the lane power domains, and those are handled now purely from
modeset_update_crtc_power_domains().

My initial motivation for this was to see if I could keep the DPIO power
wells powered down while doing AUX on CHV, but turns out I can't so this
doesn't change anything for CHV at least. But I think it's still a
worthwile change.

v2: Add case for PORT E. Default to POWER_DOMAIN_AUX_D for now. (Ville)
Signed-off-by: NVille Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: NPatrik Jakobsson <patrik.jakobsson@linux.intel.com>
Signed-off-by: NImre Deak <imre.deak@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1447682467-6237-1-git-send-email-patrik.jakobsson@linux.intel.com

Move call to gen9_set_dc_state_debugmask_memory_up() into
gen9_set_dc_state() to prevent us missing it somewhere.
Signed-off-by: NPatrik Jakobsson <patrik.jakobsson@linux.intel.com>
Reviewed-by: NImre Deak <imre.deak@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1447084107-8521-3-git-send-email-patrik.jakobsson@linux.intel.com