openwrt/staging/blogic.git
10 years agodrm/i915: preserve other DP_TEST_SINK bits.
Rodrigo Vivi [Mon, 29 Sep 2014 22:29:52 +0000 (18:29 -0400)]
drm/i915: preserve other DP_TEST_SINK bits.

Sink crc was implemented based on dp 1.1 spec that had all TEST_SINK bits
reserved reading all 0s. But when reviewing my latest changes on sink crc
Todd warned me that on new specs we have other valid bits on this reg that we
might want to preserve.

Cc: Todd Previte <tprevite@gmail.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Reviewed-by: Todd Previte <tprevite@gmail.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915/bdw: WaDisableFenceDestinationToSLM
Rodrigo Vivi [Sat, 20 Sep 2014 00:16:27 +0000 (20:16 -0400)]
drm/i915/bdw: WaDisableFenceDestinationToSLM

This WA affect BDW GT3 pre-production steppings.

Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com>
[danvet: Don't mention steppings ...]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Add IS_BDW_GT3 macro.
Rodrigo Vivi [Sat, 20 Sep 2014 00:16:26 +0000 (20:16 -0400)]
drm/i915: Add IS_BDW_GT3 macro.

It will be usefull to specify w/a that affects only BDW GT3.

Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Reviewed-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Fix Sink CRC
Rodrigo Vivi [Tue, 16 Sep 2014 23:18:12 +0000 (19:18 -0400)]
drm/i915: Fix Sink CRC

In some cases like when PSR just got enabled the panel need more vblank
times to calculate CRC. I figured that out with the new PSR test cases
facing some cases that I had a green screen but a blank CRC. Even with
2 vblank waits on kernel + 2 vblank waits on test case.

So let's give up to 6 vblank wait time. However we now check for
TEST_CRC_COUNT that shows when panel finished to calculate CRC and
has it ready.

v2: Jani pointed out attempts decrements was wrong and should never reach
the error condition. And Daniel pointed out that EIO is more appropriated than
EGAIN. Also I realized that I have to read test_crc_count after setting
test_sink

v3: Rebase and adding error message

Cc: Todd Previte <tprevite@gmail.com>
Cc: Daniel Vetter <daniel.vetter@ffwll.ch>
Cc: Jani Nikula <jani.nikula@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Reviewed-by: Todd Previte <tprevite@gmail.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Broadwell DDI Buffer translation - more tuning
Rodrigo Vivi [Thu, 25 Sep 2014 16:28:32 +0000 (12:28 -0400)]
drm/i915: Broadwell DDI Buffer translation - more tuning

BDW display - DP buffer translation values changed to give better margin.

Further change to entry 6; set dword 0 bit 31=1.

Both changes were approved already but this one didn't landed BSpec yet
this is why it is in a separated patch. Making reviewer's life easier.
Also alowing separated tests and any future bisect that might be needed.

Reference: Predator r74080 / HSD 4394389

v2: Arthur noticed I was changing the wrong bit.

Cc: Arthur Runyan <arthur.j.runyan@intel.com>
Cc: Paulo Zanoni <paulo.r.zanoni@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Reviewed-by: Arthur Runyan <arthur.j.runyan@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Broadwell DDI Buffer translation changed to give better margin.
Rodrigo Vivi [Thu, 25 Sep 2014 00:32:43 +0000 (20:32 -0400)]
drm/i915: Broadwell DDI Buffer translation changed to give better margin.

Reference: Predator r73977 / HSD 4394389

Cc: Arthur Runyan <arthur.j.runyan@intel.com>
Cc: Paulo Zanoni <paulo.r.zanoni@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Reviewed-by: Arthur Runyan <arthur.j.runyan@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Make sure PSR is ready for been re-enabled.
Rodrigo Vivi [Wed, 24 Sep 2014 22:16:58 +0000 (18:16 -0400)]
drm/i915: Make sure PSR is ready for been re-enabled.

Let's make sure PSR is propperly disabled before to re-enabled it.

According to Spec, after disabled PSR CTL, the Idle state might occur
up to 24ms, that is one full frame time (1/refresh rate),
plus SRD exit training time (max of 6ms),
plus SRD aux channel handshake (max of 1.5ms).

So if something went wrong PSR will be disabled until next full
enable/disable setup.

v2: The 24ms above takes in account 16ms for refresh rate on 60Hz mode. However
on low frequency modes this can take longer. So let's use 50ms for safeness.

v3: Move wait out of psr.lock critical area.

Cc: Daniel Vetter <daniel.vetter@ffwll.ch>
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: Minimize the huge amount of unecessary fbc sw cache clean.
Rodrigo Vivi [Wed, 24 Sep 2014 23:50:59 +0000 (19:50 -0400)]
drm/i915: Minimize the huge amount of unecessary fbc sw cache clean.

The sw cache clean on BDW is a tempoorary workaround because we cannot
set cache clean on blt ring with risk of hungs. So we are doing the cache clean on sw.
However we are doing much more than needed. Not only when using blt ring.
So, with this extra w/a we minimize the ammount of cache cleans and call it only
on same cases that it was being called on gen7.

The traditional FBC Cache clean happens over LRI on BLT ring when there is a
frontbuffer touch happening. frontbuffer tracking set fbc_dirty variable
to let BLT flush that it must clean FBC cache.

fbc.need_sw_cache_clean works in the opposite information direction
of ring->fbc_dirty telling software on frontbuffer tracking to perform
the cache clean on sw side.

v2: Clean it a little bit and fully check for Broadwell instead of gen8.

v3: Rebase after frontbuffer organization.

v4: Wiggle confused me. So fixing v3!

Cc: Daniel Vetter <daniel.vetter@ffwll.ch>
Cc: Paulo Zanoni <paulo.r.zanoni@intel.com>
Reviewed-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Avoid re-configure panel on every PSR re-enable.
Rodrigo Vivi [Tue, 16 Sep 2014 23:19:07 +0000 (19:19 -0400)]
drm/i915: Avoid re-configure panel on every PSR re-enable.

The panel has to be reconfigured only when it really loose the power.
The traditional enable/disable sequence already take care of this so we can
minimize the time spend on every re-enable.

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: PSR: Organize PSR enable function
Rodrigo Vivi [Tue, 16 Sep 2014 23:19:06 +0000 (19:19 -0400)]
drm/i915: PSR: Organize PSR enable function

We don't need to setup everything else if it doesn't match all conditions.

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: PSR: organize setup function.
Rodrigo Vivi [Tue, 16 Sep 2014 23:19:05 +0000 (19:19 -0400)]
drm/i915: PSR: organize setup function.

psr_enabled is already by itself a setup once so let's put the W/As there and
rename old setup once to setup_vsc.

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: add SW tracking to FBC enabling
Paulo Zanoni [Fri, 19 Sep 2014 19:04:55 +0000 (16:04 -0300)]
drm/i915: add SW tracking to FBC enabling

Currently, calling intel_fbc_enabled() will trigger a register read.
And we call it a lot of times, even when FBC is disabled, so saving a
few cycles would be a good thing.

Another reason for this patch is because we currently call
intel_fbc_enabled() while the HW is runtime suspended, so the read
makes no sense and triggers a WARN. This happens even if FBC is
disabled by default. Of course one could argue that we just shouldn't
be calling intel_fbc_enabled() while the driver is runtime suspended,
and I agree that's a good argument, but I still think that the reason
explained in the first paragraph already justifies the patch.
This problem can easily be reproduced with many subtests of
igt/pm_rpm, and it is a regression introduced by:

    commit c5ad011d7d256ecbe173324029e992817194d2b0
    Author: Rodrigo Vivi <rodrigo.vivi@intel.com>
    Date:   Mon Aug 4 03:51:38 2014 -0700
        drm/i915: FBC flush nuke for BDW

Testcase: igt/pm_rpm/cursor (and others)
Cc: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: extract intel_init_fbc()
Paulo Zanoni [Fri, 19 Sep 2014 19:04:54 +0000 (16:04 -0300)]
drm/i915: extract intel_init_fbc()

Because I plan to expand it a little bit.

Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: DocBook integration for frontbuffer tracking
Daniel Vetter [Fri, 19 Sep 2014 16:27:27 +0000 (18:27 +0200)]
drm/i915: DocBook integration for frontbuffer tracking

I shouldn't ask everyone to do this and fail myself ...

This extracts all the frontbuffer tracking functions into
intel_frontbuffer.c, adds a DOC overview section and also adds the
missing kerneldoc for i915_gem_track_fb and also pulls it into the
same section for convenience.

v2: Don't forget about the header files.

v3: Oops, might check compilation next time around. To make my life
easier drop the increase_pllclock from set_base_atomic since really,
it doesn't matter if you see your Oops or kgdb with a tiny bit of lag.

v4: Try to better explain how to actually use this, requested by Paulo
on irc.

v5: Explain invalidate/flush a bit clearer.

v6: s/business/busyness/

Acked-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Cc: Paulo Zanoni <paulo.r.zanoni@intel.com>
Cc: Vandana Kannan <vandana.kannan@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@intel.com>
10 years agodrm/i915: Inline feature detection into sanitize_enable_ppgtt
Chris Wilson [Fri, 19 Sep 2014 10:56:27 +0000 (11:56 +0100)]
drm/i915: Inline feature detection into sanitize_enable_ppgtt

Rather than splitting and hiding away critical parts of
sanitize_enable_ppgtt() into single use macros in the headers, inline
them into the function for clarity.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Log a message when rejecting LRM to OACONTROL
Brad Volkin [Thu, 18 Sep 2014 23:26:27 +0000 (16:26 -0700)]
drm/i915: Log a message when rejecting LRM to OACONTROL

The other paths in the command parser that reject a batch all
log a message indicating the reason. We simply missed this one.

Signed-off-by: Brad Volkin <bradley.d.volkin@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Re-enable the command parser when using PPGTT
Brad Volkin [Thu, 18 Sep 2014 23:26:26 +0000 (16:26 -0700)]
drm/i915: Re-enable the command parser when using PPGTT

In commit

commit 896ab1a5d54269b463a24194c2e4a369103b46d8
Author: Daniel Vetter <daniel.vetter@ffwll.ch>
Date:   Wed Aug 6 15:04:51 2014 +0200

    drm/i915: Fix up checks for aliasing ppgtt

it looks like we accidentally inverted the check that the command
parser should only run when the driver enables some form of PPGTT.

Testcase: igt/gem_exec_parse
Cc: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Brad Volkin <bradley.d.volkin@intel.com>
[danvet: Also drop the comment right above, all production vlv now
have hw ppgtt enabled.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Update DRIVER_DATE to 20140919
Daniel Vetter [Fri, 19 Sep 2014 15:07:10 +0000 (17:07 +0200)]
drm/i915: Update DRIVER_DATE to 20140919

Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915/vlv: Remove check for Old Ack during forcewake
Deepak S [Thu, 18 Sep 2014 13:21:50 +0000 (18:51 +0530)]
drm/i915/vlv: Remove check for Old Ack during forcewake

Based on the HW team inputs. We can should not wait for the old ack,
Waiting for old ack might fail, when other forcewake came before the
present one is desserted.

for example, if forcewake bit 0 was set and before it could get cleared
forcewake bit 1 got set, HW eventually clear bit 0, when the bit 1
is cleared. i.e, bit 1 is still sent then forcewake bit 0 will still be
set.

Signed-off-by: Deepak S <deepak.s@linux.intel.com>
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
[danvet: Add comment Ville requested.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Avoid reading fbc registers in vain when fbc was never enabled.
Rodrigo Vivi [Wed, 17 Sep 2014 20:59:20 +0000 (16:59 -0400)]
drm/i915: Avoid reading fbc registers in vain when fbc was never enabled.

If it wasn't never enabled by kernel parameter or platform default
we can avoid reading registers so many times in vain

Cc: Paulo Zanoni <paulo.r.zanoni@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Only flush fbc on sw when fbc is enabled.
Rodrigo Vivi [Fri, 5 Sep 2014 20:57:20 +0000 (16:57 -0400)]
drm/i915: Only flush fbc on sw when fbc is enabled.

Avoid touching fbc register when fbc is disabled.

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: Clarify mmio_flip_lock locking
Daniel Vetter [Mon, 15 Sep 2014 12:55:32 +0000 (14:55 +0200)]
drm/i915: Clarify mmio_flip_lock locking

The ->queue_flip callback is always called from process context, so
plain _irq spinlock variants are enough.

Reviewed-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Clarify uncore.lock locking
Daniel Vetter [Mon, 15 Sep 2014 12:55:31 +0000 (14:55 +0200)]
drm/i915: Clarify uncore.lock locking

Only one place looked in need of a bit of polish: hsw_restore_lcpll.
It's used by the runtime pm code and hence is always called from
process context. No irq flag saving required.

Another thing I've stumbled over is that we might need to add a
raw forcewake_get/put helpers which don't grab a runtime pm reference
but just check that the device isn't suspended - we have this duplicated
in the execlist code, too.

Reviewed-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Clarify irq_lock locking, special cases
Daniel Vetter [Mon, 15 Sep 2014 12:55:29 +0000 (14:55 +0200)]
drm/i915: Clarify irq_lock locking, special cases

Grab bag for all the special cases:
- i9xx_check_fifo_underruns is only called from crtc_enable hooks,
  i.e. process context.
- i915_enable_asle_pipestat is only called from interrupt postinstall
  hooks. So again process context.
- gen8_irq_power_well_post_enable is called from the runtime pm code,
  which again means process context.
- The open-coded hpd_irq_setup loop in _thaw is also running in process
  context.

So for all of them the plain _irq variant is sufficient.

Reviewed-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Clarify irq_lock locking, irq handlers
Daniel Vetter [Mon, 15 Sep 2014 12:55:28 +0000 (14:55 +0200)]
drm/i915: Clarify irq_lock locking, irq handlers

irq handlers always run with interrupts locally disabled, so
plain spinlocks is all we need. I've also reviewed again that they
all follow the _irq_handler postfix convention.

Reviewed-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Clarify irq_lock locking, interrupt install/uninstall
Daniel Vetter [Mon, 15 Sep 2014 12:55:27 +0000 (14:55 +0200)]
drm/i915: Clarify irq_lock locking, interrupt install/uninstall

All the interrupt setup/teardown hooks are always run from plain
process context. So again just the _irq variant is good enough.

Reviewed-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Clarify irq_lock locking, work functions
Daniel Vetter [Mon, 15 Sep 2014 12:55:26 +0000 (14:55 +0200)]
drm/i915: Clarify irq_lock locking, work functions

Work functions are in process context, so plain _irq spinlock variants
is all we need.

The hpd reenable work didn't follow the _work/_work_func postfix
naming scheme, so adjust that while at it.

Reviewed-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Clarify irq_lock locking, intel_tv_detect
Daniel Vetter [Mon, 15 Sep 2014 12:55:25 +0000 (14:55 +0200)]
drm/i915: Clarify irq_lock locking, intel_tv_detect

->detect callbacks are only ever called from process context, and
there's no fancy nesting going on here. So plain _irq spinlock
variants is what we want.

Reviewed-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Clarify gpu_error.lock locking
Daniel Vetter [Mon, 15 Sep 2014 12:55:24 +0000 (14:55 +0200)]
drm/i915: Clarify gpu_error.lock locking

i915_capture_error_state can be called from all kinds of contexts, so
needs the full irqsave dance. But the other two places to grab and
release the error state are only called from process context. So
simplify them to the plaine _irq spinlock versions to clarify the
locking semantics.

Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com
Reviewed-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Clarify event_lock locking, irq&mixed context
Daniel Vetter [Mon, 15 Sep 2014 12:55:23 +0000 (14:55 +0200)]
drm/i915: Clarify event_lock locking, irq&mixed context

Now we tackle the functions also called from interrupt handlers.

- intel_check_page_flip is exclusively called from irq handlers, so a
  plain spin_lock is all we need. In i915_irq.c we have the convention
  to give all such functions an _irq_handler postfix, but that would
  look strange and als be a bit a misleading name. I've opted for a
  WARN_ON(!in_irq()) instead.

- The other two places left are called both from interrupt handlers
  and from our reset work, so need the full irqsave dance. Annotate
  them with a short comment.

Reviewed-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Clarify event_lock locking, process context
Daniel Vetter [Mon, 15 Sep 2014 12:55:22 +0000 (14:55 +0200)]
drm/i915: Clarify event_lock locking, process context

It's good practice to use the more specific versions for irq save
spinlocks both as executable documentation and to enforce saner
design. The _irqsave version really should only be used if the calling
context is unknown and there's a good reason to call a function from
all kinds of places.

This is the first step whice replaces all occurances of _irqsave in
process context with the simpler irq disable/enable variants. We don't
have any funky spinlock nesting going on, especially since the
event_lock is the outermost of the irq/vblank related spinlocks.

Reviewed-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Move vblank enable earlier and disable later
Ville Syrjälä [Thu, 14 Aug 2014 19:04:37 +0000 (22:04 +0300)]
drm/i915: Move vblank enable earlier and disable later

We changed to an interrupt based vblank wait (as opposed to polling)
in:
 commit 44bd93a3d367913d883be6abba9a6e51a53c4e90
 Author: Daniel Vetter <daniel.vetter@ffwll.ch>
 Date:   Fri Jul 25 23:36:44 2014 +0200

    drm/i915: Use generic vblank wait

However we already had vblank waits on the wrong side of
drm_vblank_{on,off}() calls due to various workarounds, so now we get
a warning more or less every time we do a modeset, and we fail to
wait for the vblank like we should.

Move the drm_vblank_{on,off}() calls back out from
intel_crtc_{enable,disable}_planes() so that all of these vblank waits
return to proper operation. Also move the cxsr wait a bit earlier so
that we can keep the encoder disable after we've turned off vblanks.
Moving stuff out from the plane enable/disable functions seems
preferrable to moving the workaround stuff in since the workarounds are
required only on specific platforms.

While at it switch over to the drm_crtc_ variants of the vblank on/off
functions.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=82525
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=82490
Cc: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: static inline for intel_wait_for_vblank
Daniel Vetter [Mon, 15 Sep 2014 12:12:21 +0000 (14:12 +0200)]
drm/i915: static inline for intel_wait_for_vblank

Requested by Chris, and also requested to keep it since it's a
more accurate name in his opinion.

Cc: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Use generic vblank wait
Daniel Vetter [Mon, 15 Sep 2014 10:36:02 +0000 (12:36 +0200)]
drm/i915: Use generic vblank wait

This has the upside that it will no longer steal interrupts from the
interrupt handler on pre-g4x. Furthermore this will now scream properly
on all platforms if we don't have hw counters enabled.

v2: Adjust to the new names.

Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Matt Roper <matthew.d.roper@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Convert backlight_lock to a mutex
Daniel Vetter [Mon, 15 Sep 2014 12:35:09 +0000 (14:35 +0200)]
drm/i915: Convert backlight_lock to a mutex

Originally the irq safe spinlock was required because of asle
interrupts. But since

commit 91a60f20712179e56b7a6c3d332a5f6f9a54aa11
Author: Jani Nikula <jani.nikula@intel.com>
Date:   Thu Oct 31 18:55:48 2013 +0200

    drm/i915: move opregion asle request handling to a work queue

there's no need for this any more. So switch to the simpler mutex.

v2: Cite the right commit, spotted by Jani.

Cc: Jani Nikula <jani.nikula@intel.com>
Reviewed-by: Jani Nikula <jani.nikula@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915/bios: add missing __packed to structs used for reading vbt
Jani Nikula [Mon, 15 Sep 2014 13:59:28 +0000 (16:59 +0300)]
drm/i915/bios: add missing __packed to structs used for reading vbt

This does not seem to make a difference for the structs in question, but
document the intent.

v2: also pack union child_device_config (Daniel)

Signed-off-by: Jani Nikula <jani.nikula@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Fix irq checks in ring->irq_get/put functions
Daniel Vetter [Mon, 15 Sep 2014 09:38:57 +0000 (11:38 +0200)]
drm/i915: Fix irq checks in ring->irq_get/put functions

Yet another place that wasn't properly transformed when implementing
SOix. While at it convert the checks to WARN_ON on gen5+ (since we
don't have UMS potentially doing stupid things on those platforms).
And also add the corresponding checks to the put functions (again with
a WARN_ON) for gen5+.

v2: Drop the WARNINGS in the irq_put functions (including the existing
one for vebox), Chris convinced me that they're not that terribly
useful.

v3: Don't forget about execlist code.

Cc: Imre Deak <imre.deak@intel.com>
Cc: Jesse Barnes <jbarnes@virtuousgeek.org>
Cc: "Volkin, Bradley D" <bradley.d.volkin@intel.com>
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@intel.com>
10 years agodrm/i915: vlv: fix display IRQ enable/disable
Imre Deak [Mon, 8 Sep 2014 12:21:09 +0000 (15:21 +0300)]
drm/i915: vlv: fix display IRQ enable/disable

We want to enable/disable display IRQs only if global i915 IRQs are
enabled. To check the latter it's not enough to consult the DRM
dev->irq_enabled flag, since runtime PM can disable/enable IRQs
and it won't adjust this flag only the i915 specific
dev_priv->pm._irqs_disabled flag. Fix this by using the proper
intel_irqs_enabled() helper instead.

Fortunately this didn't cause an actual problem since even if we enabled
display IRQs too early (before enabling global i915 IRQs) the
VLV_MASTER_IER would still be clear masking all IRQs.

This issue was caught by

commit 920dd15a2b2fc60d054646a8a1ffd6aeb6090e05
Author: Daniel Vetter <daniel.vetter@ffwll.ch>
Date:   Wed Aug 27 10:43:37 2014 +0200

    drm/i915: WARN if interrupts aren't on in en/disable_pipestat

Signed-off-by: Imre Deak <imre.deak@intel.com>
Cc: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Only set CURSOR_PIPE_CSC_ENABLE when cursor is enabled
Ville Syrjälä [Fri, 12 Sep 2014 17:53:33 +0000 (20:53 +0300)]
drm/i915: Only set CURSOR_PIPE_CSC_ENABLE when cursor is enabled

It seems cleaner if we keep CURCNTR at 0 when the cursor is disabled,
so don't set the CURSOR_PIPE_CSC_ENABLE bit unless the cursor is
enabled.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Move the cursor_base setup to i{845, 9xx}_update_cursor()
Ville Syrjälä [Fri, 12 Sep 2014 17:53:32 +0000 (20:53 +0300)]
drm/i915: Move the cursor_base setup to i{845, 9xx}_update_cursor()

To make the code a bit more undestandable move the
intel_crtc->cursor_base assignment into the low level update cursor
routines. That's were we compare the current value with the new one
so immediately seeing that it gets assigned only afterwards helps
one to understand that it gets assigned only after the comparison.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agoagp/intel-gtt: Remove get/put_pages
Daniel Vetter [Fri, 12 Sep 2014 13:20:24 +0000 (15:20 +0200)]
agp/intel-gtt: Remove get/put_pages

If a page isn't allocated as __GFP_MOVEABLE it won't move around, so
no need to grab a reference to lock it into place.

Discovered while reviewing page allocation handling in i915 gem.

Acked-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@intel.com>
10 years agodrm/i915: Drop get/put_pages for scratch page
Daniel Vetter [Fri, 12 Sep 2014 13:18:13 +0000 (15:18 +0200)]
drm/i915: Drop get/put_pages for scratch page

While discussing/reviewing __GFP_MOVEABLE behaviour and interactions
with our various page allocations on irc Chris brought up that the
scratch page isn't allocated as moveable, but we still grab/put a
reference to lock it in place. Which is unecessary.

So drop that.

Acked-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@intel.com>
10 years agodrm/i915: Fix regression in the sprite plane update split
Gustavo Padovan [Thu, 11 Sep 2014 20:42:15 +0000 (17:42 -0300)]
drm/i915: Fix regression in the sprite plane update split

7e4bf45dbd99a965c7b5d5944c6dc4246f171eb5 introduced the regression.
We fix it by doing the right assignment of crtc_y

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=83747
Signed-off-by: Gustavo Padovan <gustavo.padovan@collabora.co.uk>
Reviewed-by: Jani Nikula <jani.nikula@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Remove dead code, i915_gem_verify_gtt
Chris Wilson [Wed, 10 Sep 2014 18:52:18 +0000 (19:52 +0100)]
drm/i915: Remove dead code, i915_gem_verify_gtt

The data structure it was supposed to be sanity checking has long gone.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: create intel_update_pipe_size()
Gustavo Padovan [Wed, 10 Sep 2014 15:04:17 +0000 (12:04 -0300)]
drm/i915: create intel_update_pipe_size()

Factor out a piece of code from intel_pipe_set_base() that updates
the pipe size and adjust fitter.

This will help refactor the update primary plane path.

v2: use struct intel_crtc as argument to intel_update_pipe_size()

v3: use 'crtc' as argument name

Signed-off-by: Gustavo Padovan <gustavo.padovan@collabora.co.uk>
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: pin sprite fb only if it changed
Gustavo Padovan [Wed, 10 Sep 2014 15:03:17 +0000 (12:03 -0300)]
drm/i915: pin sprite fb only if it changed

Optimize code avoiding helding dev mutex if old fb and current fb
are the same.

v2: take Ville's comments
- move comment along with the pin_and_fence call
- check for error before calling i915_gem_track_fb
- move old_obj != obj to an upper if condition

Signed-off-by: Gustavo Padovan <gustavo.padovan@collabora.co.uk>
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: remove !enabled handling from commit primary plane step
Gustavo Padovan [Tue, 9 Sep 2014 14:43:19 +0000 (11:43 -0300)]
drm/i915: remove !enabled handling from commit primary plane step

The !crtc->enabled case will now be handled by the !visible code,
since the handling is basically the same.

Signed-off-by: Gustavo Padovan <gustavo.padovan@collabora.co.uk>
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Restore resume irq ordering comment
Daniel Vetter [Mon, 8 Sep 2014 16:28:20 +0000 (18:28 +0200)]
drm/i915: Restore resume irq ordering comment

This was lost in

commit e11aa362308f5de467ce355a2a2471321b15a35c
Author: Jesse Barnes <jbarnes@virtuousgeek.org>
Date:   Wed Jun 18 09:52:55 2014 -0700

    drm/i915: use runtime irq suspend/resume in freeze/thaw

which makes the second part of this commen a bit nonsense. Both were
originally added in

commit 15239099d7a7a9ecdc1ccb5b187ae4cda5488ff9
Author: Daniel Vetter <daniel.vetter@ffwll.ch>
Date:   Tue Mar 5 09:50:58 2013 +0100

    drm/i915: enable irqs earlier when resuming

Cc: Jesse Barnes <jbarnes@virtuousgeek.org>
Cc: Imre Deak <imre.deak@intel.com>
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Mika Kuoppala <mika.kuoppala@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@intel.com>
Reviewed-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: split intel_primary_plane_setplane() into check() and commit()
Gustavo Padovan [Fri, 5 Sep 2014 20:04:49 +0000 (17:04 -0300)]
drm/i915: split intel_primary_plane_setplane() into check() and commit()

As a preparation for atomic updates we need to split the code to check
everything we are going to commit first. This patch starts the work to
split intel_primary_plane_setplane() into check() and commit() parts.

More work is expected on this to get a better split of the two steps.
Ideally the commit() step should never fail.

Signed-off-by: Gustavo Padovan <gustavo.padovan@collabora.co.uk>
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: split intel_cursor_plane_update() into check() and commit()
Gustavo Padovan [Fri, 5 Sep 2014 20:22:31 +0000 (17:22 -0300)]
drm/i915: split intel_cursor_plane_update() into check() and commit()

Due to the upcoming atomic modesetting feature we need to separate
some update functions into a check step that can fail and a commit
step that should, ideally, never fail.

The commit part can still fail, but that should be solved in another
upcoming patch.

Signed-off-by: Gustavo Padovan <gustavo.padovan@collabora.co.uk>
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: split intel_update_plane into check() and commit()
Gustavo Padovan [Fri, 5 Sep 2014 20:04:47 +0000 (17:04 -0300)]
drm/i915: split intel_update_plane into check() and commit()

Due to the upcoming atomic modesetting feature we need to separate
some update functions into a check step that can fail and a commit
step that should, ideally, never fail.

This commit splits intel_update_plane() and its commit part can still
fail due to the fb pinning procedure.

Signed-off-by: Gustavo Padovan <gustavo.padovan@collabora.co.uk>
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: create struct intel_plane_state
Gustavo Padovan [Fri, 5 Sep 2014 20:04:46 +0000 (17:04 -0300)]
drm/i915: create struct intel_plane_state

This new struct will be the storage of src and dst coordinates
between the check and commit stages of a plane update.

Signed-off-by: Gustavo Padovan <gustavo.padovan@collabora.co.uk>
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: WARN if interrupts aren't on in en/disable_pipestat
Daniel Vetter [Wed, 27 Aug 2014 08:43:37 +0000 (10:43 +0200)]
drm/i915: WARN if interrupts aren't on in en/disable_pipestat

Now that vlv has runtime pm we kinda should check for that like on the
pch split platforms. Looks like this was simply lost in the vlv rpm
enabling.

Cc: Paulo Zanoni <paulo.r.zanoni@intel.com>
Cc: Imre Deak <imre.deak@intel.com>
Cc: Jesse Barnes <jbarnes@virtuousgeek.org>
Reviewed-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Enable full PPGTT on gen7
Michel Thierry [Fri, 5 Sep 2014 13:13:16 +0000 (14:13 +0100)]
drm/i915: Enable full PPGTT on gen7

Use full PPGTT as the default option in gen7.

Note that aliasing PPGTT is the default option for gen8 (see
HAS_PPGTT) since we're still fighting troubles around context
switching and execlists.

This may well come back to bite me later.

Signed-off-by: Michel Thierry <michel.thierry@intel.com>
[danvet: Explain that gen8 full ppgtt is blocked on execlists for
now.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Use EIO instead of EAGAIN for sink CRC error.
Rodrigo Vivi [Mon, 15 Sep 2014 23:24:03 +0000 (19:24 -0400)]
drm/i915: Use EIO instead of EAGAIN for sink CRC error.

If something while getting panel CRC this means that probably hw I/O error
so hw is busted and try again shouldn't help much.

Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Extend BIOS stolen mem handling to all platform
Daniel Vetter [Thu, 11 Sep 2014 11:28:08 +0000 (13:28 +0200)]
drm/i915: Extend BIOS stolen mem handling to all platform

Based upon a patch from Deepak, but reworked to only apply on gen7+
and with the logic a bit clarified.

v2: Fix s/SHIFT/MASK/ fumble that Ville spotted.

Cc: Deepak S <deepak.s@intel.com>
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Match GTT space sanity checker with implementation
Chris Wilson [Thu, 11 Sep 2014 07:43:48 +0000 (08:43 +0100)]
drm/i915: Match GTT space sanity checker with implementation

If we believe that the device can cross cache domains in its prefetcher
(i.e. we allow neighbouring pages in different domains), we don't supply
a color_adjust callback. Use the presence of this callback to better
determine when we should be verifying that the GTT space we just
used is valid.

v2: Remove the superfluous struct drm_device function param as well.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
[danvet: Also adjust the comment per irc discussion with Chris.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: HSW always use GGTT selector for secure batches
Chris Wilson [Wed, 10 Sep 2014 11:18:27 +0000 (12:18 +0100)]
drm/i915: HSW always use GGTT selector for secure batches

gen6 and earlier conflate address space selection (ppgtt vs ggtt) with
the security bit (i.e. only privileged batches were allowed to run from
ggtt). From Haswell only, you are able to select the security bit
separate from the address space - and we always requested to use ppgtt.
This breaks the golden render state batch execution with full-ppgtt as
that is only present in the global GTT and more generally any secure
batch that is not colocated in the ppgtt and ggtt. So we need to
disable the use of the ppgtt selector bit for secure batches, or else we
hang immediately upon boot and thence after every GPU reset...

v2: Only HSW differentiates between secure dispatch and ggtt, so simply
ignore the differentiation and always use secure==ggtt.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
[danvet: Rectify commit message as noted by Chris.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: add cherryview specfic forcewake in execlists_elsp_write
Deepak S [Tue, 9 Sep 2014 13:44:16 +0000 (19:14 +0530)]
drm/i915: add cherryview specfic forcewake in execlists_elsp_write

In chv, we have two power wells Render & Media. We need to use
corresponsing forcewake count. If we dont follow this we are getting
error "*ERROR*: Timed out waiting for forcewake old ack to clear" due to
multiple entry into __vlv_force_wake_get.

Signed-off-by: Deepak S <deepak.s@linux.intel.com>
Requested-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: fix another use-after-free in i915_gem_evict_everything
Michel Thierry [Tue, 9 Sep 2014 12:04:43 +0000 (13:04 +0100)]
drm/i915: fix another use-after-free in i915_gem_evict_everything

Also here, i915_gem_evict_vm causes an unbind, which can end up dropping
the last ref to the ppgtt.

Triggered by igt gem_evict_everything test.

Testcase: igt/gem_evict_everything
Signed-off-by: Michel Thierry <michel.thierry@intel.com>
Reviewed-by: Chris Wilson <chris@cris-wilsonc.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Don't reinit hpd interrupts after gpu reset
Daniel Vetter [Mon, 8 Sep 2014 16:17:18 +0000 (18:17 +0200)]
drm/i915: Don't reinit hpd interrupts after gpu reset

Somehow I've overlooked this when simplifying the irq reinit
scheme on gen4.5+ in

commit 78ad455fd229c6f6cc2f390ccbe0d8f1a62d55a9
Author: Daniel Vetter <daniel.vetter@ffwll.ch>
Date:   Thu May 22 22:18:21 2014 +0200

    drm/i915: Improve irq handling after gpu resets

Since display interrups in general survive a gpu reset on those
platforms there's also no need to reinit the hotplug settings.

Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Wrap -EIO send-vblank event for failed pageflip in spinlock
Chris Wilson [Sun, 7 Sep 2014 15:51:12 +0000 (16:51 +0100)]
drm/i915: Wrap -EIO send-vblank event for failed pageflip in spinlock

drm_send_vblank_event() demands that we hold the event spinlock whilst
calling it, so do so.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
[danvet: Fix the double lock as requested by Chris.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Drop any active reference before unbinding
Chris Wilson [Tue, 9 Sep 2014 06:02:43 +0000 (07:02 +0100)]
drm/i915: Drop any active reference before unbinding

Before we process the final unbind on an object and move it to the
unbound list, it is semantically cleaner if there are no more active
references to the object. (An active reference would imply that it was
still being accessed by the GPU after it became inaccessible.) The
caveat is that all callsites must be prepared for the object to
disappeared during the unbind - i.e. they must hold their own reference.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Objects on the unbound list may still have an active reference
Chris Wilson [Tue, 9 Sep 2014 10:16:08 +0000 (11:16 +0100)]
drm/i915: Objects on the unbound list may still have an active reference

Due to the lazy retirement semantics, even though we have unbound an
object, it may still hold onto an active reference. So in the debug code,
play safe.

v2: Export i915_gem_shrink() rather than opencoding it.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915/edp: use lane count and link rate from DPCD for eDP
Jani Nikula [Tue, 9 Sep 2014 08:25:13 +0000 (11:25 +0300)]
drm/i915/edp: use lane count and link rate from DPCD for eDP

eDP panels are generally designed to support only a single clock and
lane configuration.

commit 56071a207602a451f0c46d3dcc8379b59ef576e2
Author: Jani Nikula <jani.nikula@intel.com>
Date:   Tue May 6 14:56:52 2014 +0300

    drm/i915: use lane count and link rate from VBT as minimums for eDP

should have started using the optimal link parameters for eDP
panels. Turns out a certain other OS uses DPCD instead of VBT, which
means trusting VBT on this may not be so reliable after all. Follow
suit.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=81647
Tested-by: Adam Jirasek <libm3l@gmail.com>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=79386
Tested-by: Narthana Epa <narthana.epa+freedesktop@gmail.com>
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915/dp: add missing \n in the TPS3 debug message
Jani Nikula [Fri, 5 Sep 2014 13:19:18 +0000 (16:19 +0300)]
drm/i915/dp: add missing \n in the TPS3 debug message

This goes back to

commit 06ea66b6bb445043dc25a9626254d5c130093199
Author: Todd Previte <tprevite@gmail.com>
Date:   Mon Jan 20 10:19:39 2014 -0700

    drm/i915: Enable 5.4Ghz (HBR2) link rate for Displayport 1.2-capable devices

Cc: Todd Previte <tprevite@gmail.com>
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
[danvet: Pimp commit message a bit.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915/hdmi, dp: Do not dereference the encoder in the connector destroy
Chris Wilson [Thu, 4 Sep 2014 20:43:45 +0000 (21:43 +0100)]
drm/i915/hdmi, dp: Do not dereference the encoder in the connector destroy

Oops, apparently intel_hdmi/intel_dp is the encoder - an object with a
distinct lifetime to the connector, and so we cannot simply reuse the
common function to unset and free the edid.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Limit the watermark to at least 8 entries on gen2/3
Ville Syrjälä [Fri, 5 Sep 2014 18:54:13 +0000 (21:54 +0300)]
drm/i915: Limit the watermark to at least 8 entries on gen2/3

830 is very unhappy of the watermark value is too low (indicating a very
high watermark in fact, ie. memory fetch will occur with an almost full
FIFO). Limit the watermark value to at least 8 cache lines.

That also matches the burst size we use on most platforms. BSpec seems
to indicate we should limit the watermark to 'burst size + 1'. But on
gen4 we already use a hardcoded 8 as the watermark value (as the spec
says we should), so just use 8 as the limit on gen2/3 as well.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Fix DVO 2x clock enable on 830M
Ville Syrjälä [Fri, 5 Sep 2014 18:52:42 +0000 (21:52 +0300)]
drm/i915: Fix DVO 2x clock enable on 830M

The spec says:
"For the correct operation of the muxed DVO pins (GDEVSELB/ I2Cdata,
GIRDBY/I2CClk) and (GFRAMEB/DVI_Data, GTRDYB/DVI_Clk): Bit 31
(DPLL VCO Enable) and Bit 30 (2X Clock Enable) must be set to “1” in
both the DPLL A Control Register (06014h-06017h) and DPLL B Control
Register (06018h-0601Bh)."

The pipe A and B force quirks take care of DPLL_VCO_ENABLE, so we
just need a bit of special care to handle DPLL_DVO_2X_MODE.

v2: Recompute num_dvo_pipes on the spot, use PIPE_A/PIPE_B instead
    of pipe/!pipe for the register offsets in disable (Daniel)
    Add a comment about the ordering in enable and another one
    about filtering out the DVO 2x bit in state readout

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Tested-by: Thomas Richter <richter@rus.uni-stuttgart.de> (v1)
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agoMerge branch 'drm/next/du' of git://linuxtv.org/pinchartl/fbdev into drm-next
Dave Airlie [Wed, 17 Sep 2014 09:55:19 +0000 (19:55 +1000)]
Merge branch 'drm/next/du' of git://linuxtv.org/pinchartl/fbdev into drm-next

Commit "drm/rcar-du: Use struct videomode in platform data" touches board code
in arch/arm/mach-shmobile. There is, to the best of my knowledge, no risk of
conflict for v3.18. Simon, are you fine with getting those changes merged
through Dave's tree (and could you confirm that no conflict should occur) ?

Simon acked the merge:
Acked-by: Simon Horman <horms+renesas@verge.net.au>
* 'drm/next/du' of git://linuxtv.org/pinchartl/fbdev:
  drm/rcar-du: Add OF support
  drm/rcar-du: Use struct videomode in platform data
  video: Add DT bindings for the R-Car Display Unit
  video: Add THC63LVDM83D DT bindings documentation
  video: Add ADV7123 DT bindings documentation
  video: Add DT binding documentation for VGA connector
  devicetree: Add vendor prefix "thine" to vendor-prefixes.txt
  devicetree: Add vendor prefix "mitsubishi" to vendor-prefixes.txt
  drm/shmob: Update copyright notice
  drm/rcar-du: Update copyright notice

10 years agodrm/ttm: make sure format string cannot leak in
Kees Cook [Thu, 11 Sep 2014 20:53:54 +0000 (13:53 -0700)]
drm/ttm: make sure format string cannot leak in

While zone->name is currently hard coded, the call to kobject_init_and_add()
should follow the more defensive argument list usage (as already done in
other places in ttm_memory.c) where "%s" is used instead of directly passing
in a variable as a format string.

Signed-off-by: Kees Cook <keescook@chromium.org>
Signed-off-by: Dave Airlie <airlied@redhat.com>
10 years agodrm/tilcdc: panel: Add support for enable GPIO
Ezequiel Garcia [Tue, 2 Sep 2014 12:51:22 +0000 (09:51 -0300)]
drm/tilcdc: panel: Add support for enable GPIO

In order to support the "enable GPIO" available in many panel devices,
this commit adds a proper devicetree binding.

By providing an enable GPIO in the devicetree, the driver can now turn
off and on the panel device, and/or the backlight device. Both the
backlight and the GPIO are optional properties.

Tested-by: Darren Etheridge <detheridge@ti.com>
Tested-by: Johannes Pointner <johannes.pointner@br-automation.com>
Signed-off-by: Ezequiel Garcia <ezequiel@vanguardiasur.com.ar>
Signed-off-by: Dave Airlie <airlied@redhat.com>
10 years agodrm/tilcdc: panel: Set return value explicitly
Ezequiel Garcia [Tue, 2 Sep 2014 12:51:21 +0000 (09:51 -0300)]
drm/tilcdc: panel: Set return value explicitly

Instead of setting an initial value for the return code, set it explicitly
on each error path. This is just a cosmetic cleanup, as preparation for the
enable GPIO support.

Tested-by: Darren Etheridge <detheridge@ti.com>
Tested-by: Johannes Pointner <johannes.pointner@br-automation.com>
Signed-off-by: Ezequiel Garcia <ezequiel@vanguardiasur.com.ar>
Signed-off-by: Dave Airlie <airlied@redhat.com>
10 years agodrm/tilcdc: panel: Fix backlight devicetree support
Ezequiel Garcia [Tue, 2 Sep 2014 12:51:20 +0000 (09:51 -0300)]
drm/tilcdc: panel: Fix backlight devicetree support

The current backlight support is broken; the driver expects a backlight-class
in the panel devicetree node. Fix this by implementing it properly, getting
an optional backlight from a phandle.

This shouldn't cause any backward-compatibility DT issue because the current
implementation doesn't work and is not even documented.

Tested-by: Darren Etheridge <detheridge@ti.com>
Tested-by: Johannes Pointner <johannes.pointner@br-automation.com>
Signed-off-by: Ezequiel Garcia <ezequiel@vanguardiasur.com.ar>
Signed-off-by: Dave Airlie <airlied@redhat.com>
10 years agodrm/tilcdc: panel: Use devm_kzalloc to simplify the error path
Ezequiel Garcia [Tue, 2 Sep 2014 12:51:19 +0000 (09:51 -0300)]
drm/tilcdc: panel: Use devm_kzalloc to simplify the error path

Using the managed variant to allocate the resource makes the code simpler
and less error-prone.

Tested-by: Darren Etheridge <detheridge@ti.com>
Tested-by: Johannes Pointner <johannes.pointner@br-automation.com>
Signed-off-by: Ezequiel Garcia <ezequiel@vanguardiasur.com.ar>
Signed-off-by: Dave Airlie <airlied@redhat.com>
10 years agodrm/tilcdc: panel: Spurious whitespace removal
Ezequiel Garcia [Tue, 2 Sep 2014 12:51:18 +0000 (09:51 -0300)]
drm/tilcdc: panel: Spurious whitespace removal

Just a cosmetic cleanup.

Tested-by: Darren Etheridge <detheridge@ti.com>
Tested-by: Johannes Pointner <johannes.pointner@br-automation.com>
Signed-off-by: Ezequiel Garcia <ezequiel@vanguardiasur.com.ar>
Signed-off-by: Dave Airlie <airlied@redhat.com>
10 years agodrm/tilcdc: panel: Remove unused variable
Ezequiel Garcia [Tue, 2 Sep 2014 12:51:17 +0000 (09:51 -0300)]
drm/tilcdc: panel: Remove unused variable

Just a trivial cleanup to remove the variable.

Tested-by: Darren Etheridge <detheridge@ti.com>
Tested-by: Johannes Pointner <johannes.pointner@br-automation.com>
Signed-off-by: Ezequiel Garcia <ezequiel@vanguardiasur.com.ar>
Signed-off-by: Dave Airlie <airlied@redhat.com>
10 years agodrm/tilcdc: panel: Add missing of_node_put
Ezequiel Garcia [Tue, 2 Sep 2014 12:51:16 +0000 (09:51 -0300)]
drm/tilcdc: panel: Add missing of_node_put

This commit adds the missing calls to of_node_put to release the node
that's currently held by the of_get_child_by_name() call in the panel
info parsing code.

Tested-by: Darren Etheridge <detheridge@ti.com>
Tested-by: Johannes Pointner <johannes.pointner@br-automation.com>
Signed-off-by: Ezequiel Garcia <ezequiel@vanguardiasur.com.ar>
Signed-off-by: Dave Airlie <airlied@redhat.com>
10 years agodrm/tilcdc: Fix the error path in tilcdc_load()
Ezequiel Garcia [Tue, 2 Sep 2014 12:51:15 +0000 (09:51 -0300)]
drm/tilcdc: Fix the error path in tilcdc_load()

The current error path calls tilcdc_unload() in case of an error to release
the resources. However, this is wrong because not all resources have been
allocated by the time an error occurs in tilcdc_load().

To fix it, this commit adds proper labels to bail out at the different
stages in the load function, and release only the resources actually allocated.

Tested-by: Darren Etheridge <detheridge@ti.com>
Tested-by: Johannes Pointner <johannes.pointner@br-automation.com>
Signed-off-by: Ezequiel Garcia <ezequiel@vanguardiasur.com.ar>
Signed-off-by: Dave Airlie <airlied@redhat.com>
10 years agoMerge tag 'drm-intel-next-2014-09-05' of git://anongit.freedesktop.org/drm-intel...
Dave Airlie [Tue, 16 Sep 2014 06:02:09 +0000 (16:02 +1000)]
Merge tag 'drm-intel-next-2014-09-05' of git://anongit.freedesktop.org/drm-intel into drm-next

- final bits (again) for the rotation support (Sonika Jindal)
- support bl_power in the intel backlight (Jani)
- vdd handling improvements from Ville
- i830M fixes from Ville
- piles of prep work all over to make skl enabling just plug in (Damien, Sonika)
- rename DP training defines to reflect latest edp standards, this touches all
  drm drivers supporting DP (Sonika Jindal)
- cache edids during single detect cycle to avoid re-reading it for e.g. audio,
  from Chris
- move w/a for registers which are stored in the hw context to the context init
  code (Arun&Damien)
- edp panel power sequencer fixes, helps chv a lot (Ville)
- piles of other chv fixes all over
- much more paranoid pageflip handling with stall detection and better recovery
  from Chris
- small things all over, as usual

* tag 'drm-intel-next-2014-09-05' of git://anongit.freedesktop.org/drm-intel: (114 commits)
  drm/i915: Update DRIVER_DATE to 20140905
  drm/i915: Decouple the stuck pageflip on modeset
  drm/i915: Check for a stalled page flip after each vblank
  drm/i915: Introduce a for_each_plane() macro
  drm/i915: Rewrite ABS_DIFF() in a safer manner
  drm/i915: Add comments explaining the vdd on/off functions
  drm/i915: Move DP port disable to post_disable for pch platforms
  drm/i915: Enable DP port earlier
  drm/i915: Turn on panel power before doing aux transfers
  drm/i915: Be more careful when picking the initial power sequencer pipe
  drm/i915: Reset power sequencer pipe tracking when disp2d is off
  drm/i915: Track which port is using which pipe's power sequencer
  drm/i915: Fix edp vdd locking
  drm/i915: Reset the HEAD pointer for the ring after writing START
  drm/i915: Fix unsafe vma iteration in i915_drop_caches
  drm/i915: init sprites with univeral plane init function
  drm/i915: Check of !HAS_PCH_SPLIT() in PCH transcoder funcs
  drm/i915: Use HAS_GMCH_DISPLAY un underrun reporting code
  drm/i915: Use IS_BROADWELL() instead of IS_GEN8() in forcewake code
  drm/i915: Don't call gen8_fbc_sw_flush() on chv
  ...

10 years agoMerge branch 'drm-next-ast-fixes' of ssh://people.freedesktop.org/~/linux into drm...
Dave Airlie [Tue, 16 Sep 2014 04:59:16 +0000 (14:59 +1000)]
Merge branch 'drm-next-ast-fixes' of ssh://people.freedesktop.org/~/linux into drm-next

Pull in first set of changes from Ben for ast on ppc.

I've done a quick boot test on x86 and it still seems to boot.

* 'drm-next-ast-fixes' of ssh://people.freedesktop.org/~/linux:
  drm/ast: Cleanup analog init code path
  drm/ast: Don't assume DVO enabled means SIL164 on uninitialized chips
  drm/ast: Properly initialize P2A base before using it in ast_init_3rdtx()
  drm/ast: POST chip at probe time if VGA not enabled
  drm/ast: Try to use MMIO registers when PIO isn't supported

10 years agodrm/ast: Add reduced blanking modes for wide screen mode
Y.C. Chen [Thu, 28 Aug 2014 09:11:04 +0000 (17:11 +0800)]
drm/ast: Add reduced blanking modes for wide screen mode

Signed-off-by: Egbert Eich <eich@suse.com>
Tested-by: Steven You2 Liang <liangyou2@lenovo.com>
Signed-off-by: Y.C. Chen <yc_chen@aspeedtech.com>
v3: based on [PATCH 1/2] drm/ast: Add missing entry to dclk_table[].
    Add reduced blanking modes, improve mode matching to
    identify these modes by thier sync polarities.

[airlied: argh whitespace damage]
Signed-off-by: Dave Airlie <airlied@redhat.com>
10 years agodrm: backmerge tag 'v3.17-rc5' into drm-next
Dave Airlie [Tue, 16 Sep 2014 01:28:52 +0000 (11:28 +1000)]
drm: backmerge tag 'v3.17-rc5' into drm-next

This is requested to get the fixes for intel and radeon into the
same tree for future development work.

i915_display.c: fix missing dev_priv conflict.

10 years agoMerge branch 'linux-3.18' of git://anongit.freedesktop.org/git/nouveau/linux-2.6...
Dave Airlie [Mon, 15 Sep 2014 20:20:53 +0000 (06:20 +1000)]
Merge branch 'linux-3.18' of git://anongit.freedesktop.org/git/nouveau/linux-2.6 into drm-next

This is the main merge request for Nouveau 3.18, overview:
- various bits of roy's gt21x clock work
- various bits of kepler memory clock work (don't get too excited, there's at least one more major bit left that's busting higher freqs)
- misc fan control improvements
- kepler hdmi infoframe fixes
- dp audio
- l2 cache + cbc improvements

* 'linux-3.18' of git://anongit.freedesktop.org/git/nouveau/linux-2.6: (68 commits)
  drm/gt214-/disp: enable dp audio
  drm/gt214-/kms: fix hda eld regression
  drm/g94-/disp: calculate some dp audio constants
  drm/gt214-/kms: perform hda codec setup on displayport too
  drm/gk104-/disp: infoframe registers moved yet again on kepler
  drm/nouveau/bios: parse older ramcfg/timing data like we do newer ones
  drm/nva3/fb/ram: Per-partition regs
  drm/nouveau/fb/ram: Support strided regs
  drm/nv50/fb/ram: Store the number of partitions in the designated fields
  drm/nv50/kms: Set VBLANK time in modeset script
  drm/nouveau/bios: Add rammap support for version 1.0
  drm/gf100-/pwr/memx: block host and fifo around reclock
  drm/nouveau/pwr/memx: fix command ordering around block/unblock
  drm/nouveau/pwr/memx: rename fb off/on to block/unblock
  drm/nva3/clk: Pause the GPU before reclocking
  drm/nouveau/gpio: rename g92 class to g94
  drm/gk104-/fb/ram: move fb enable/disable to same place as nvidia
  drm/gk104/fb/ram: twiddle some more bits when reclocking
  drm/nouveau/bios: parse another large chunk of random memory config data
  drm/gk104-/fb/ram: perform certain steps only when bios data differs
  ...

10 years agodrm/gt214-/disp: enable dp audio
Ben Skeggs [Mon, 15 Sep 2014 11:29:05 +0000 (21:29 +1000)]
drm/gt214-/disp: enable dp audio

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
10 years agodrm/gt214-/kms: fix hda eld regression
Ben Skeggs [Mon, 15 Sep 2014 11:11:51 +0000 (21:11 +1000)]
drm/gt214-/kms: fix hda eld regression

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
10 years agodrm/g94-/disp: calculate some dp audio constants
Ben Skeggs [Mon, 15 Sep 2014 05:55:56 +0000 (15:55 +1000)]
drm/g94-/disp: calculate some dp audio constants

NVIDIA appear to have tweaked the algorithm from GF110, this implements
the previous algorithm for them still.

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
10 years agodrm/gt214-/kms: perform hda codec setup on displayport too
Ben Skeggs [Mon, 15 Sep 2014 05:20:47 +0000 (15:20 +1000)]
drm/gt214-/kms: perform hda codec setup on displayport too

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
10 years agodrm/gk104-/disp: infoframe registers moved yet again on kepler
Ben Skeggs [Mon, 15 Sep 2014 05:15:09 +0000 (15:15 +1000)]
drm/gk104-/disp: infoframe registers moved yet again on kepler

Thanks to Vincent Pelletier for pointing this out and providing a proof of
concept patch on the list.

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
10 years agodrm/nouveau/bios: parse older ramcfg/timing data like we do newer ones
Ben Skeggs [Mon, 15 Sep 2014 02:30:08 +0000 (12:30 +1000)]
drm/nouveau/bios: parse older ramcfg/timing data like we do newer ones

Done after discussion with Roy.

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
10 years agodrm/nva3/fb/ram: Per-partition regs
Roy Spliet [Fri, 12 Sep 2014 16:00:16 +0000 (18:00 +0200)]
drm/nva3/fb/ram: Per-partition regs

Signed-off-by: Roy Spliet <rspliet@eclipso.eu>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
10 years agodrm/nouveau/fb/ram: Support strided regs
Roy Spliet [Fri, 12 Sep 2014 16:00:15 +0000 (18:00 +0200)]
drm/nouveau/fb/ram: Support strided regs

Signed-off-by: Roy Spliet <rspliet@eclipso.eu>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
10 years agodrm/nv50/fb/ram: Store the number of partitions in the designated fields
Roy Spliet [Fri, 12 Sep 2014 16:00:14 +0000 (18:00 +0200)]
drm/nv50/fb/ram: Store the number of partitions in the designated fields

Signed-off-by: Roy Spliet <rspliet@eclipso.eu>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
10 years agodrm/nv50/kms: Set VBLANK time in modeset script
Roy Spliet [Fri, 12 Sep 2014 16:00:13 +0000 (18:00 +0200)]
drm/nv50/kms: Set VBLANK time in modeset script

Solves blinking on reclocking memory. The value set is an underestimate, but
with non-reduced vblanking this should give us plenty of time

Signed-off-by: Roy Spliet <rspliet@eclipso.eu>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
10 years agodrm/nouveau/bios: Add rammap support for version 1.0
Roy Spliet [Fri, 12 Sep 2014 16:00:12 +0000 (18:00 +0200)]
drm/nouveau/bios: Add rammap support for version 1.0

Signed-off-by: Roy Spliet <rspliet@eclipso.eu>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
10 years agodrm/gf100-/pwr/memx: block host and fifo around reclock
Ben Skeggs [Thu, 11 Sep 2014 13:32:20 +0000 (23:32 +1000)]
drm/gf100-/pwr/memx: block host and fifo around reclock

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
10 years agodrm/nouveau/pwr/memx: fix command ordering around block/unblock
Ben Skeggs [Thu, 11 Sep 2014 13:04:22 +0000 (23:04 +1000)]
drm/nouveau/pwr/memx: fix command ordering around block/unblock

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
10 years agodrm/nouveau/pwr/memx: rename fb off/on to block/unblock
Ben Skeggs [Thu, 11 Sep 2014 12:59:13 +0000 (22:59 +1000)]
drm/nouveau/pwr/memx: rename fb off/on to block/unblock

More accurate as to the function of the opcodes.  Not only is FB disabled,
but the host is prevented from touching the GPU.  An upcoming patch for
Kepler will also halt PFIFO (as NVIDIA does).

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
10 years agodrm/nva3/clk: Pause the GPU before reclocking
Roy Spliet [Fri, 29 Aug 2014 10:27:42 +0000 (12:27 +0200)]
drm/nva3/clk: Pause the GPU before reclocking

V2: always call post correctly even if pre fails
V3: move function prototype to nva3.h

Signed-off-by: Roy Spliet <rspliet@eclipso.eu>
10 years agodrm/nouveau/gpio: rename g92 class to g94
Emil Velikov [Mon, 8 Sep 2014 19:27:57 +0000 (20:27 +0100)]
drm/nouveau/gpio: rename g92 class to g94

nv92 hardware has only 16 interrupt lines, while nv94 and later
has 32. Accessing 0xe0c{0,4} registers on nv92 can lead to incorrect
PDISP setup. This is a regression introduced with

commit 9d0f5ec9ee0fd5dc5fc1cc2cf559286431e406e3
Author: Ben Skeggs <bskeggs@redhat.com>
Date:   Mon May 12 15:22:42 2014 +1000

    gpio: split g92 class from nv50

Reported-by: estece on #nouveau
Cc: stable@vger.kernel.org # 3.16+
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>