GitHub/moto-9609/android_kernel_motorola_exynos9610.git
12 years agodrm/i915: Implement WaSetupGtModeTdRowDispatch
Daniel Vetter [Fri, 14 Dec 2012 22:38:29 +0000 (23:38 +0100)]
drm/i915: Implement WaSetupGtModeTdRowDispatch

I'm not really sure, since the w/a entry is as thin on details as
ever, and Bspec doesn't say anything about it. But I've figured only
dispatching to rows 0&1 instead of all four should be the right thing
for GT1.

Reviewed-by: Rodrigo Vivi <rodrigo.vivi@gmail.com>
[danvet: Add the missing snb server GT1 to the check, spotted by Chris
Wilson.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
12 years agodrm/i915: Implement WaDisableHiZPlanesWhenMSAAEnabled
Daniel Vetter [Fri, 14 Dec 2012 22:38:28 +0000 (23:38 +0100)]
drm/i915: Implement WaDisableHiZPlanesWhenMSAAEnabled

Quoting from Bspec, 3D_CHICKEN1, bit 10

This bit needs to be set always to "1", Project: DevSNB "

Reviewed-by: Rodrigo Vivi <rodrigo.vivi@gmail.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
12 years agodrm/i915: Prefer CRTC 'active' rather than 'enabled' during WM computations
Chris Wilson [Fri, 7 Dec 2012 10:43:25 +0000 (10:43 +0000)]
drm/i915: Prefer CRTC 'active' rather than 'enabled' during WM computations

Only the intel_crtc->active is accurate at the point where we wish to
perform WM computations, so use it instead of crtc->enabled.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
12 years agodrm/i915: Clear self-refresh watermarks when disabled
Chris Wilson [Fri, 7 Dec 2012 10:43:24 +0000 (10:43 +0000)]
drm/i915: Clear self-refresh watermarks when disabled

If we elect to disable self-refresh as they require too many FIFO
entries, clear the values prior to writing them into the registers. If
they are too large they may occupy more bits than available and so
corrupt neighbouring WM values.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
12 years agodrm/i915: Double the cursor self-refresh latency on Valleyview
Chris Wilson [Tue, 11 Dec 2012 12:01:43 +0000 (12:01 +0000)]
drm/i915: Double the cursor self-refresh latency on Valleyview

It operates at twice the declared latency, so double the latency value
used for the cursor watermark calculation.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=50248
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
CC: Jesse Barnes <jbarnes@virtuousgeek.org>
Reviewed-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
12 years agodrm/i915: Fixup cursor latency used for IVB lp3 watermarks
Chris Wilson [Tue, 11 Dec 2012 12:01:42 +0000 (12:01 +0000)]
drm/i915: Fixup cursor latency used for IVB lp3 watermarks

It operates at twice the declared latency, so adjust the computation to
avoid potential flicker at low power.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=50248
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
CC: Jesse Barnes <jbarnes@virtuousgeek.org>
Reviewed-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
12 years agodrm/i915: Fix missed needs_dmar setting
Zhenyu Wang [Thu, 13 Dec 2012 15:47:47 +0000 (23:47 +0800)]
drm/i915: Fix missed needs_dmar setting

From Ben's AGP dependence removal change, "needs_dmar" flag has not
been properly setup for new chips using new GTT init function. This
one adds missed setting of that flag to make sure we do pci mappings
with IOMMU enabled.

Signed-off-by: Zhenyu Wang <zhenyuw@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
12 years agodrm/i915: Fix shifted screen on top of LVDS on IVY laptop
Takashi Iwai [Tue, 11 Dec 2012 10:46:29 +0000 (11:46 +0100)]
drm/i915: Fix shifted screen on top of LVDS on IVY laptop

The commit [23670b322: drm/i915: CPT+ pch transcoder workaround]
caused a regression on some HP laptops with IvyBridge.  The whole
laptop screen is shifted downward for a few pixels constantly.
The problem appears only on LVDS while DP and VGA seem unaffected.
Also, the problem disappears once when go and back from S3.
(S4 resume still shows the same problem.)

This patch revives the minimum part the commit above dropped.
For fixing this regression, only the setup of CHICKEN2 bit in
cpt_init_clock_gating() is needed.

Signed-off-by: Takashi Iwai <tiwai@suse.de>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
12 years agodrm/i915: disable cpt phase pointer fdi rx workaround
Daniel Vetter [Sat, 8 Dec 2012 11:58:33 +0000 (12:58 +0100)]
drm/i915: disable cpt phase pointer fdi rx workaround

We've originally added this in

commit 291427f5fdadec6e4be2924172e83588880e1539
Author: Jesse Barnes <jbarnes@virtuousgeek.org>
Date:   Fri Jul 29 12:42:37 2011 -0700

    drm/i915: apply phase pointer override on SNB+ too

and then copy-pasted it over to ivb/ppt. The w/a was originally added
for ilk/ibx in

commit 5b2adf897146edeac6a1e438fb67b5a53dbbdf34
Author: Jesse Barnes <jbarnes@virtuousgeek.org>
Date:   Thu Oct 7 16:01:15 2010 -0700

    drm/i915: add Ironlake clock gating workaround for FDI link training

and fixed up a bit in

commit 6f06ce184c765fd8d50669a8d12fdd566c920859
Author: Jesse Barnes <jbarnes@virtuousgeek.org>
Date:   Tue Jan 4 15:09:38 2011 -0800

    drm/i915: set phase sync pointer override enable before setting phase sync pointer

It turns out that this w/a isn't actually required on cpt/ppt and
positively harmful on ivb/ppt when using fdi B/C links - it results in
a black screen occasionally, with seemingfully everything working as
it should. The only failure indication I've found in the hw is that
eventually (but not right after the modeset completes) a pipe underrun
is signalled.

Big thanks to Arthur Runyan for all the ideas for registers to check
and changes to test, otherwise I couldn't ever have tracked this down!

Cc: "Runyan, Arthur J" <arthur.j.runyan@intel.com>
Cc: stable@vger.kernel.org
Reviewed-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
12 years agodrm/i915: set the LPT FDI RX polarity reversal bit when needed
Paulo Zanoni [Sat, 1 Dec 2012 14:04:26 +0000 (12:04 -0200)]
drm/i915: set the LPT FDI RX polarity reversal bit when needed

If we fail to set the bit when needed we get some nice FDI link
training failures (AKA "black screen on VGA output").

While we don't really know how to properly choose whether we need to
set the bit or not (VBT?), just read the initial value set by the BIOS
and store it for later usage.

Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
12 years agodrm/i915: add lpt_init_pch_refclk
Paulo Zanoni [Sat, 1 Dec 2012 14:04:25 +0000 (12:04 -0200)]
drm/i915: add lpt_init_pch_refclk

We need this code to init the PCH SSC refclk and the FDI registers.
The BIOS does this too and that's why VGA worked before this patch,
until you tried to suspend the machine...

This patch implements the "Sequence to enable CLKOUT_DP for FDI usage
and configure PCH FDI/IO" from our documentation.

v2:
- Squash Damien Lespiau's reset spelling fix on top.
- Add a comment that we don't need to bother about the ULT special
  case Damien noticed, since ULT won't have VGA.
- Add a comment to rip out the SDV codepaths once haswell ships for
  real.

Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com> (v1)
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
12 years agodrm/i915: add support for mPHY destination on intel_sbi_{read, write}
Paulo Zanoni [Sat, 1 Dec 2012 14:04:24 +0000 (12:04 -0200)]
drm/i915: add support for mPHY destination on intel_sbi_{read, write}

This way we should be able to write mPHY registers using the Sideband
Interface in the next commit. Also fixed some syntax oddities in the
related code.

Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Reviewed-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
12 years agodrm/i915: reject modes the LPT FDI receiver can't handle
Paulo Zanoni [Thu, 29 Nov 2012 13:29:32 +0000 (11:29 -0200)]
drm/i915: reject modes the LPT FDI receiver can't handle

More specifically, the LPT FDI RX only supports 8bpc and a maximum of
2 lanes, so anything above that won't work and should be rejected.

Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
12 years agodrm/i915: fix hsw_fdi_link_train "retry" code
Paulo Zanoni [Thu, 29 Nov 2012 13:29:31 +0000 (11:29 -0200)]
drm/i915: fix hsw_fdi_link_train "retry" code

We were previously doing exactly what the "mode set sequence for CRT"
document mandates, but whenever we failed to train the link in the
first tentative, all the other subsequent retries always failed. In
one of my monitors that has 47 modes, I was usually getting around 3
failures when running "testdisplay -a".

After this patch, even if we fail in the first tentative, we can
succeed in the next ones. So now when running "testdisplay -a" I see
around 3 times the message "FDI link training done on step 1" and no
failures.

Notice that now the "retry" code looks a lot like the DP retry code.

Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
12 years agodrm/i915: Close race between processing unpin task and queueing the flip
Chris Wilson [Mon, 3 Dec 2012 11:36:30 +0000 (11:36 +0000)]
drm/i915: Close race between processing unpin task and queueing the flip

Before queuing the flip but crucially after attaching the unpin-work to
the crtc, we continue to setup the unpin-work. However, should the
hardware fire early, we see the connected unpin-work and queue the task.
The task then promptly runs and unpins the fb before we finish taking
the required references or even pinning it... Havoc.

To close the race, we use the flip-pending atomic to indicate when the
flip is finally setup and enqueued. So during the flip-done processing,
we can check more accurately whether the flip was expected.

v2: Add the appropriate mb() to ensure that the writes to the page-flip
worker are complete prior to marking it active and emitting the MI_FLIP.
On the read side, the mb should be enforced by the spinlocks.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: stable@vger.kernel.org
[danvet: Review the barriers a bit, we need a write barrier both
before and after updating ->pending. Similarly we need a read barrier
in the interrupt handler both before and after reading ->pending. With
well-ordered irqs only one barrier in each place should be required,
but since this patch explicitly sets out to combat spurious interrupts
with is staged activation of the unpin work we need to go full-bore on
the barriers, too. Discussed with Chris Wilson on irc and changes
acked by him.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
12 years agodrm/i915: fixup l3 parity sysfs access check
Daniel Vetter [Wed, 5 Dec 2012 08:52:14 +0000 (09:52 +0100)]
drm/i915: fixup l3 parity sysfs access check

When l3 parity support for Haswell was enabled in

commit f27b92651d72e863c308ea5dca5615fc98e38ca6
Author: Ben Widawsky <benjamin.widawsky@intel.com>
Date:   Tue Jul 24 20:47:32 2012 -0700

    drm/i915: Expand DPF support to Haswell

no one noticed that the patch which introduced this macro

commit e1ef7cc299839e68dae3f1843f62e52acda04538
Author: Ben Widawsky <benjamin.widawsky@intel.com>
Date:   Tue Jul 24 20:47:31 2012 -0700

    drm/i915: Macro to determine DPF support

missed one spot. Fix this.

Cc: Ben Widawsky <benjamin.widawsky@intel.com>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=57441
Reviewed-by: Ben Widawsky <benjamin.widawsky@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
12 years agodrm/i915: Clear the existing watermarks for g4x when modifying the cursor sr
Chris Wilson [Tue, 4 Dec 2012 16:33:19 +0000 (16:33 +0000)]
drm/i915: Clear the existing watermarks for g4x when modifying the cursor sr

In a couple of places we attempt to adjust the existing watermark
registers to update them for the new cursor watermarks. This goes
horribly wrong as instead of clearing the cursor bits prior to or'ing in
the new values, we clear the rest of the register with the result that
the watermark registers contain bogus values.

References: https://bugs.freedesktop.org/show_bug.cgi?id=47034
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
12 years agodrm/i915: do not access BLC_PWM_CTL2 on pre-gen4 hardware
Jani Nikula [Tue, 4 Dec 2012 14:36:28 +0000 (16:36 +0200)]
drm/i915: do not access BLC_PWM_CTL2 on pre-gen4 hardware

The BLC_PWM_CTL2 register does not exist before gen4. While at it, do a
slight drive by cleanup of the code.

Signed-off-by: Jani Nikula <jani.nikula@intel.com>
Reviewed-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
12 years agodrm/i915: Don't allow ring tail to reach the same cacheline as head
Ville Syrjälä [Mon, 3 Dec 2012 16:43:32 +0000 (18:43 +0200)]
drm/i915: Don't allow ring tail to reach the same cacheline as head

From BSpec:
"If the Ring Buffer Head Pointer and the Tail Pointer are on the same
cacheline, the Head Pointer must not be greater than the Tail
Pointer."

The easiest way to enforce this is to reduce the reported ring space.

References:
Gen2 BSpec "1. Programming Environment" / 1.4.4.6 "Ring Buffer Use"
Gen3 BSpec "vol1c Memory Interface Functions" / 2.3.4.5 "Ring Buffer Use"
Gen4+ BSpec "vol1c Memory Interface and Command Stream" / 5.3.4.5 "Ring Buffer Use"

v2: Include the exact BSpec references in the description

v3: s/64/I915_RING_FREE_SPACE, and add the BSpec information to the code

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>
12 years agodrm/i915: Decouple the object from the unbound list before freeing pages
Chris Wilson [Mon, 3 Dec 2012 11:49:00 +0000 (11:49 +0000)]
drm/i915: Decouple the object from the unbound list before freeing pages

As we may actually allocate in order to save the physical swizzling bits
during the free, we have to be careful not to trigger the shrinker on
the same object.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
[danvet: Added a small comment in the code to really drive the
scariness of this patch home.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
12 years agodrm/i915: Set sync_seqno properly after seqno wrap
Mika Kuoppala [Wed, 28 Nov 2012 15:18:45 +0000 (17:18 +0200)]
drm/i915: Set sync_seqno properly after seqno wrap

i915_gem_handle_seqno_wrap() will zero all sync_seqnos but as the
wrap can happen inside ring->sync_to(), pre wrap seqno was
carried over and overwrote the zeroed sync_seqno.

When wrap is handled, all outstanding requests will be retired and
objects moved to inactive queue, causing their last_read_seqno to be zero.
Use this to update the sync_seqno correctly.

RING_SYNC registers after wrap will contain pre wrap values which
are >= seqno. So injecting the semaphore wait into ring completes
immediately.

Original idea for using last_read_seqno from Chris Wilson.

Signed-off-by: Mika Kuoppala <mika.kuoppala@intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
12 years agodrm/i915: Include the last semaphore sync point in the error-state
Chris Wilson [Tue, 27 Nov 2012 17:06:54 +0000 (17:06 +0000)]
drm/i915: Include the last semaphore sync point in the error-state

Should be useful to know what the driver thought the other ring's seqno
was when it last used a semaphore.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
12 years agodrm/i915: Rearrange code to only have a single method for waiting upon the ring
Chris Wilson [Tue, 27 Nov 2012 16:22:54 +0000 (16:22 +0000)]
drm/i915: Rearrange code to only have a single method for waiting upon the ring

Replace the wait for the ring to be clear with the more common wait for
the ring to be idle. The principle advantage is one less exported
intel_ring_wait function, and the removal of a hardcoded value.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
12 years agodrm/i915: Simplify flushing activity on the ring
Chris Wilson [Tue, 27 Nov 2012 16:22:53 +0000 (16:22 +0000)]
drm/i915: Simplify flushing activity on the ring

As we now always preallocate the seqno before writing to the ring, we
can trivially test if we have any pending activity on the ring by
inspecting the olr. This makes it then possible to flush operations that
are not normally associated with a request, like power-management.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
12 years agodrm/i915: Preallocate next seqno before touching the ring
Chris Wilson [Tue, 27 Nov 2012 16:22:52 +0000 (16:22 +0000)]
drm/i915: Preallocate next seqno before touching the ring

Based on the work by Mika Kuoppala, we realised that we need to handle
seqno wraparound prior to committing our changes to the ring. The most
obvious point then is to grab the seqno inside intel_ring_begin(), and
then to reuse that seqno for all ring operations until the next request.
As intel_ring_begin() can fail, the callers must already be prepared to
handle such failure and so we can safely add further checks.

This patch looks like it should be split up into the interface
changes and the tweaks to move seqno wrapping from the execbuffer into
the core seqno increment. However, I found no easy way to break it into
incremental steps without introducing further broken behaviour.

v2: Mika found a silly mistake and a subtle error in the existing code;
inside i915_gem_retire_requests() we were resetting the sync_seqno of
the target ring based on the seqno from this ring - which are only
related by the order of their allocation, not retirement. Hence we were
applying the optimisation that the rings were synchronised too early,
fortunately the only real casualty there is the handling of seqno
wrapping.

v3: Do not forget to reset the sync_seqno upon module reinitialisation,
ala resume.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Mika Kuoppala <mika.kuoppala@intel.com>
Bugzilla: https://bugzilla.redhat.com/show_bug.cgi?id=863861
Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com> [v2]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
12 years agodrm/i915: force restore on lid open
Daniel Vetter [Fri, 23 Nov 2012 17:16:34 +0000 (18:16 +0100)]
drm/i915: force restore on lid open

There seem to be indeed some awkwards machines around, mostly those
without OpRegion support, where the firmware changes the display hw
state behind our backs when closing the lid.

This force-restore logic has been originally introduced in

commit c1c7af60892070e4b82ad63bbfb95ae745056de0
Author: Jesse Barnes <jbarnes@virtuousgeek.org>
Date:   Thu Sep 10 15:28:03 2009 -0700

    drm/i915: force mode set at lid open time

but after the modeset-rework we've disabled it in the vain hope that
it's no longer required:

commit 3b7a89fce3e3dc96b549d6d829387b4439044d0d
Author: Daniel Vetter <daniel.vetter@ffwll.ch>
Date:   Mon Sep 17 22:27:21 2012 +0200

    drm/i915: fix OOPS in lid_notify

Alas, no.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=54677
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=57434
Tested-by: Krzysztof Mazur <krzysiek@podlesie.net>
Reviewed-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
12 years agodrm/i915: Wait upon the last request seqno, rather than a future seqno
Chris Wilson [Thu, 22 Nov 2012 13:07:20 +0000 (13:07 +0000)]
drm/i915: Wait upon the last request seqno, rather than a future seqno

In commit 69c2fc891343cb5217c866d10709343cff190bdc
Author: Chris Wilson <chris@chris-wilson.co.uk>
Date:   Fri Jul 20 12:41:03 2012 +0100

    drm/i915: Remove the per-ring write list

the explicit flush was removed from i915_ring_idle(). However, we
continued to wait upon the next seqno which now did not correspond to
any request (except for the unusual condition of a failure to queue a
request after execbuffer) and so would wait indefinitely.

This has an important side-effect that i915_gpu_idle() does not cause
the seqno to be incremented. This is vital if we are to be able to idle
the GPU to handle seqno wraparound, as in subsequent patches.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
12 years agodrm/i915: fix possible NULL dereference of dev_priv
Mika Kuoppala [Mon, 12 Nov 2012 12:20:19 +0000 (14:20 +0200)]
drm/i915: fix possible NULL dereference of dev_priv

Dereference dev_priv only after we know it is valid.
Found with smatch.

Signed-off-by: Mika Kuoppala <mika.kuoppala@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
12 years agodrm/i915: Increase the response time for slow SDVO devices
Chris Wilson [Fri, 23 Nov 2012 11:57:56 +0000 (11:57 +0000)]
drm/i915: Increase the response time for slow SDVO devices

Some devices may respond very slowly and only flag that the reply is
pending within the first 15us response window. Be kind to such devices
and wait a further 15ms, before checking for the pending reply. This
moves the existing special case delay of 30ms down from the detection
routine into the common path and pretends to explain it...

v2: Simplify the loop constructs as suggested by Jani Nikula.

References: https://bugs.freedesktop.org/show_bug.cgi?id=36997
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Jani Nikula <jani.nikula@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
12 years agodrm/i915: set the VIC of the mode on the AVI InfoFrame
Paulo Zanoni [Fri, 23 Nov 2012 14:09:27 +0000 (12:09 -0200)]
drm/i915: set the VIC of the mode on the AVI InfoFrame

We currently set "0" as the VIC value of the AVI InfoFrames. According
to the specs this should be fine and work for every mode, so to my
point of view we can't consider the current behavior as a bug. The
problem is that  we recently received a bug report (Kernel bug #50371)
from a user that has an AV receiver that gives a black screen for any
mode with VIC set to 0.

So in order to make at least some modes work for him, this patch sets
the correct VIC number when sending AVI InfoFrames.

Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=50371
Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
12 years agodrm: add drm_mode_cea_vic
Paulo Zanoni [Fri, 23 Nov 2012 14:09:26 +0000 (12:09 -0200)]
drm: add drm_mode_cea_vic

This function returns the VIC of the mode. This value can be used when
creating AVI InfoFrames.

Cc: Thierry Reding <thierry.reding@avionic-design.de>
Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=50371
Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Reviewed-by: Thierry Reding <thierry.reding@avionic-design.de>
Acked-by: Dave Airlie <airlied@gmail.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
12 years agodrm/i915: Fix pte updates in ggtt clear range
Ben Widawsky [Tue, 27 Nov 2012 05:52:54 +0000 (21:52 -0800)]
drm/i915: Fix pte updates in ggtt clear range

This bug was introduced by me:
commit e76e9aebcdbfebae8f4cd147e3c0f800d36e97f3
Author: Ben Widawsky <ben@bwidawsk.net>
Date:   Sun Nov 4 09:21:27 2012 -0800

    drm/i915: Stop using AGP layer for GEN6+

The existing code uses memset_io which follows memset semantics in only
guaranteeing a write of individual bytes. Since a PTE entry is 4 bytes,
this can only be correct if the scratch page address is 0.

This caused unsightly errors when we clear the range at load time,
though I'm not really sure what the heck is referencing that memory
anyway. I caught this is because I believe we have some other bug where
the display is doing reads of memory we feel should be cleared (or we
are relying on scratch pages to be a specific value).

Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
12 years agodrm/i915: promote Haswell to full support
Paulo Zanoni [Tue, 20 Nov 2012 15:32:30 +0000 (13:32 -0200)]
drm/i915: promote Haswell to full support

Since it should be working a little bit better now.

Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
12 years agodrm/i915: Report the origin of the LVDS fixed panel mode
Chris Wilson [Wed, 21 Nov 2012 16:14:03 +0000 (16:14 +0000)]
drm/i915: Report the origin of the LVDS fixed panel mode

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Jani Nikula <jani.nikula@intel.com>
[danvet: resolve conflict around the call to intel_crtc_mode_get. And
add the missing NULL check Chris spotted while at it.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
12 years agodrm/i915: LVDS fallback to fixed-mode if EDID not present
Chris Wilson [Wed, 21 Nov 2012 16:14:04 +0000 (16:14 +0000)]
drm/i915: LVDS fallback to fixed-mode if EDID not present

Use the recorded panel fixed-mode to populate the get_modes() request in
the absence of an EDID.

Fixes regression from
commit 9cd300e038d492af4990b04e127e0bd2df64b1ca
Author: Jani Nikula <jani.nikula@intel.com>
Date:   Fri Oct 19 14:51:52 2012 +0300

    drm/i915: Move cached EDID to intel_connector

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Jani Nikula <jani.nikula@intel.com>
Cc: Jesse Barnes <jbarnes@virtuousgeek.org>
Reviewed-by: Jani Nikula <jani.nikula@intel.com>
[danvet: Drop the retval-changing hunk, as suggested by Jani in his
review and acked by Chris.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
12 years agodrm/i915/sdvo: kfree the intel_sdvo_connector, not drm_connector, on destroy
Jani Nikula [Mon, 12 Nov 2012 16:31:36 +0000 (18:31 +0200)]
drm/i915/sdvo: kfree the intel_sdvo_connector, not drm_connector, on destroy

Since the base fields in both struct intel_connector and struct
intel_sdvo_connector are at the beginning of the enclosing struct, the
pointers are essentially the same, but there is no requirement or guarantee
that this is always the case. Kfree the enclosing intel_sdvo_connector
pointer that was originally allocated, not the enclosed drm_connector, in
case someone ever rearranges the structs.

Signed-off-by: Jani Nikula <jani.nikula@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
12 years agodrm/i915: drm_connector_property -> drm_object_property
Rob Clark [Fri, 12 Oct 2012 01:36:04 +0000 (20:36 -0500)]
drm/i915: drm_connector_property -> drm_object_property

v2: Rebased.

Signed-off-by: Rob Clark <rob@ti.com>
Reviewed-by: Jani Nikula <jani.nikula@intel.com> (v1)
[danvet: Pimp commit message a bit.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
12 years agodrm/i915: use drm_send_vblank_event() helper
Rob Clark [Mon, 8 Oct 2012 19:50:40 +0000 (14:50 -0500)]
drm/i915: use drm_send_vblank_event() helper

Signed-off-by: Rob Clark <rob@ti.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
12 years agodrm/i915: Use pci_resource functions for BARs.
Ben Widawsky [Mon, 19 Nov 2012 20:23:44 +0000 (12:23 -0800)]
drm/i915: Use pci_resource functions for BARs.

This was leftover crap from kill-agp. The current code is theoretically
broken for 64b bars. (I resist removing theoretically because I am too
lazy to test).

We still need to ioremap things ourselves because we want to ioremap_wc
the PTEs.

v2: Forgot to kill the tmp variable in v1

CC: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
12 years agodrm/i915: Borrow our struct_mutex for the direct reclaim
Chris Wilson [Wed, 21 Nov 2012 13:04:04 +0000 (13:04 +0000)]
drm/i915: Borrow our struct_mutex for the direct reclaim

If we have hit oom whilst holding our struct_mutex, then currently we
cannot reap our own GPU buffers which likely pin most of memory, making
an outright OOM more likely. So if we are running in direct reclaim and
already hold the mutex, attempt to free buffers knowing that the
original function can not continue until we return.

v2: Add a note explaining that the mutex may be stolen due to
pre-emption, and that is bad.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
12 years agodrm/i915: Defer assignment of obj->gtt_space until after all possible mallocs
Chris Wilson [Wed, 21 Nov 2012 13:04:03 +0000 (13:04 +0000)]
drm/i915: Defer assignment of obj->gtt_space until after all possible mallocs

As we may invoke the shrinker whilst trying to allocate memory to hold
the gtt_space for this object, we need to be careful not to mark the
drm_mm_node as activated (by assigning it to this object) before we
have finished our sequence of allocations.

Note: We also need to move the binding of the object into the actual
pagetables down a bit. The best way seems to be to move it out into
the callsites.

Reported-by: Imre Deak <imre.deak@gmail.com>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
[danvet: Added small note to commit message to summarize review
discussion.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
12 years agodrm/i915: Apply the IBX transcoder A w/a for HDMI to SDVO as well
Chris Wilson [Wed, 21 Nov 2012 10:44:23 +0000 (10:44 +0000)]
drm/i915: Apply the IBX transcoder A w/a for HDMI to SDVO as well

As the SDVO/HDMI registers are multiplex, it is safe to assume that the
w/a required for HDMI on IbexPoint, namely that the SDVO register cannot
both be disabled and have selected transcoder B, is also required for
SDVO. At least the modeset state checker detects that the transcoder
selection is left in the undefined state, and so it appears sensible to
apply the w/a:

[ 1814.480052] WARNING: at drivers/gpu/drm/i915/intel_display.c:1487 assert_pch_hdmi_disabled+0xad/0xb5()
[ 1814.480053] Hardware name: Libretto W100
[ 1814.480054] IBX PCH hdmi port still using transcoder B

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=57066
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
12 years agodrm/i915: implement WaMbcDriverBootEnable on Haswell
Paulo Zanoni [Tue, 20 Nov 2012 15:27:44 +0000 (13:27 -0200)]
drm/i915: implement WaMbcDriverBootEnable on Haswell

Also document the WA name for the previous gens that implement it.

Reviewed-by: Jani Nikula <jani.nikula@intel.com>
Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
12 years agodrm/i915: fix intel_ddi_get_cdclk_freq for ULT machines
Paulo Zanoni [Tue, 20 Nov 2012 15:27:43 +0000 (13:27 -0200)]
drm/i915: fix intel_ddi_get_cdclk_freq for ULT machines

For now, this code is just used by the eDP AUX channel frequency.

Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Reviewed-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
12 years agodrm/i915: make the panel fitter work on pipes B and C on Haswell
Paulo Zanoni [Tue, 20 Nov 2012 15:27:42 +0000 (13:27 -0200)]
drm/i915: make the panel fitter work on pipes B and C on Haswell

This goes on a separate patch since it won't apply on the stable
trees and there's nothing using panel fitter on HSW on the older
Kernels.

Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Reviewed-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
12 years agodrm/i915: make the panel fitter work on pipes B and C on IVB
Paulo Zanoni [Tue, 20 Nov 2012 15:27:41 +0000 (13:27 -0200)]
drm/i915: make the panel fitter work on pipes B and C on IVB

I actually found this problem on Haswell, but then discovered Ivy
Bridge also has it by reading the spec.

I don't have the hardware to test this.

Cc: stable@vger.kernel.org
Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Reviewed-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
12 years agodrm/i915: don't intel_crt_init if DDI A has 4 lanes
Paulo Zanoni [Tue, 20 Nov 2012 15:27:40 +0000 (13:27 -0200)]
drm/i915: don't intel_crt_init if DDI A has 4 lanes

DDI A and E have 4 lanes to share, so if DDI A is using 4 lanes,
there's nothing left for DDI E, which means there's no CRT port on the
machine.

The bit we're checking here is programmed at system boot and it cannot
be changed afterwards, so we cannot change the amount of lanes
reserved for each DDI port.

Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Reviewed-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
12 years agodrm/i915: make DP work on LPT-LP machines
Paulo Zanoni [Tue, 20 Nov 2012 17:12:07 +0000 (15:12 -0200)]
drm/i915: make DP work on LPT-LP machines

We need to enable a special bit, otherwise none of the DP functions
requiring the PCH will work.

Version 2: store the PCH ID inside dev_priv, as suggested by Daniel
Vetter.

Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Reviewed-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
12 years agodrm/i915: fix false positive "Unclaimed write" messages
Paulo Zanoni [Tue, 20 Nov 2012 15:27:38 +0000 (13:27 -0200)]
drm/i915: fix false positive "Unclaimed write" messages

We don't check if the "unclaimed register" bit is set before we call
writel, so if it was already set before, we might print a misleading
message about "unclaimed write" on the wrong register.

This patch makes us check the unclaimed bit before the writel, so we
can print a new "Unknown unclaimed register before writing to %x"
message.

Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Reviewed-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
12 years agodrm/i915: use cpu/pch transcoder on intel_enable_pipe
Paulo Zanoni [Tue, 20 Nov 2012 15:27:37 +0000 (13:27 -0200)]
drm/i915: use cpu/pch transcoder on intel_enable_pipe

This function runs on Haswell, so set the correct pch_transcoder and
cpu_transcoder variables. This fixes an assertion failure on Haswell
VGA.

Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Reviewed-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
12 years agodrm/i915: don't limit Haswell CRT encoder to pipe A
Paulo Zanoni [Tue, 20 Nov 2012 15:27:35 +0000 (13:27 -0200)]
drm/i915: don't limit Haswell CRT encoder to pipe A

This is a full revert of 59c859d6f2e78344945e8a8406a194156176bc4e:
    drm/i915: account for only one PCH receiver on Haswell

Now that the PCH code is fixed to be able use the only PCH transcoder
independently of the pipe and CPU transcoder, we can revert this.

Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Reviewed-by: Damien Lespiau <damien.lespiau@intel.com>
[danvet: Resolve conflict due to the rebasing of dinq on top of
drm-next.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
12 years agodrm/i915: Flush outstanding unpin tasks before pageflipping
Chris Wilson [Thu, 1 Nov 2012 09:26:26 +0000 (09:26 +0000)]
drm/i915: Flush outstanding unpin tasks before pageflipping

If we accumulate unpin tasks because we are pageflipping faster than the
system can schedule its workers, we can effectively create a
pin-leak. The solution taken here is to limit the number of unpin tasks
we have per-crtc and to flush those outstanding tasks if we accumulate
too many. This should prevent any jitter in the normal case, and also
prevent the hang if we should run too fast.

Note: It is important that we switch from the system workqueue to our
own dev_priv->wq since all work items on that queue are guaranteed to
only need the dev->struct_mutex and not any modeset resources. For
otherwise if we have a work item ahead in the queue which needs the
modeset lock (like the output detect work used by both polling or
hpd), this work and so the unpin work will never execute since the
pageflip code already holds that lock. Unfortunately there's no
lockdep support for this scenario in the workqueue code.

Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=46991
Reported-and-tested-by: Tvrtko Ursulin <tvrtko.ursulin@onelan.co.uk>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
[danvet: Added note about workqueu deadlock.]
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=56337
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
12 years agodrm/i915: resurrect panel lid handling
Daniel Vetter [Tue, 20 Nov 2012 13:50:08 +0000 (14:50 +0100)]
drm/i915: resurrect panel lid handling

But disabled by default. This essentially reverts

commit bcd5023c961a44c7149936553b6929b2b233dd27
Author: Dave Airlie <airlied@redhat.com>
Date:   Mon Mar 14 14:17:55 2011 +1000

    drm/i915: disable opregion lid detection for now

but leaves the autodetect mode disabled. There's also the explicit lid
status option added in

commit fca874092597ef946b8f07031d8c31c58b212144
Author: Chris Wilson <chris@chris-wilson.co.uk>
Date:   Thu Feb 17 13:44:48 2011 +0000

    drm/i915: Add a module parameter to ignore lid status

Which overloaded the meaning for the panel_ignore_lid parameter even
more. To fix up this mess, give the non-negative numbers 0,1 the
original meaning back and use negative numbers to force a given state.
So now we have

1  - disable autodetect, return unknown
0  - enable autodetect
-1 - force to disconnected/lid closed
-2 - force to connected/lid open

v2: My C programmer license has been revoked ...

v3: Beautify the code a bit, as suggested by Chris Wilson.

Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=27622
Tested-by: Andreas Sturmlechner <andreas.sturmlechner@gmail.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
12 years agodrm/i915: Enable DP audio for Haswell
Takashi Iwai [Mon, 19 Nov 2012 17:06:51 +0000 (18:06 +0100)]
drm/i915: Enable DP audio for Haswell

This patch adds the missing code to send ELD for Haswell DisplayPort,
based on Xingchao's original patch.

A test was performed with HSW-D machine and NEC EA232Wmi DP monitor.

Cc: Xingchao Wang <xingchao.wang@intel.com>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
Reviewed-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
12 years agodrm/i915: Pin the object whilst faulting it in
Chris Wilson [Tue, 20 Nov 2012 10:45:17 +0000 (10:45 +0000)]
drm/i915: Pin the object whilst faulting it in

In order to prevent reaping of the object whilst setting it up to
handle the pagefault, we need to mark it as pinned. This has the nice
side-effect of eliminating some special cases from the pagefault handler
as well!

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
12 years agodrm/i915: Guard pages being reaped by OOM whilst binding-to-GTT
Chris Wilson [Tue, 20 Nov 2012 10:45:16 +0000 (10:45 +0000)]
drm/i915: Guard pages being reaped by OOM whilst binding-to-GTT

In the circumstances that the shrinker is allowed to steal the mutex
in order to reap pages, we need to be careful to prevent it operating on
the current object and shooting ourselves in the foot.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
12 years agodrm/i915: Remove bogus test for a present execbuffer
Chris Wilson [Mon, 19 Nov 2012 15:30:42 +0000 (15:30 +0000)]
drm/i915: Remove bogus test for a present execbuffer

The intention of checking obj->gtt_offset!=0 is to verify that the
target object was listed in the execbuffer and had been bound into the
GTT. This is guarranteed by the earlier rearrangement to split the
execbuffer operation into reserve and relocation phases and then
verified by the check that the target handle had been processed during
the reservation phase.

However, the actual checking of obj->gtt_offset==0 is bogus as we can
indeed reference an object at offset 0. For instance, the framebuffer
installed by the BIOS often resides at offset 0 - causing EINVAL as we
legimately try to render using the stolen fb.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Eric Anholt <eric@anholt.net>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
12 years agodrm/i915: Remove save/restore of physical HWS_PGA register
Chris Wilson [Fri, 16 Nov 2012 11:43:21 +0000 (11:43 +0000)]
drm/i915: Remove save/restore of physical HWS_PGA register

Now that we always restore the HWS registers (both physical and GTT
virtual addresses) when re-initialising the rings, we can eliminate the
superfluous save/restore of the register across suspend and resume.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
12 years agodrm/i915: Fix warning in i915_gem_chipset_flush
Ben Widawsky [Thu, 15 Nov 2012 20:06:09 +0000 (12:06 -0800)]
drm/i915: Fix warning in i915_gem_chipset_flush

drivers/gpu/drm/i915/i915_drv.h:1545:2: warning: '______f' is static but
declared in inline function 'i915_gem_chipset_flush' which is not static

Reported-by: kbuild test robot <fengguang.wu@intel.com>
dri-devel-Reference: <50a4d41c.586VhmwghPuKZbkB%fengguang.wu@intel.com>
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
12 years agodrm/i915: Only check for valid PP_{ON, OFF}_DELAYS on pre ILK hardware
Damien Lespiau [Wed, 31 Oct 2012 19:23:16 +0000 (19:23 +0000)]
drm/i915: Only check for valid PP_{ON, OFF}_DELAYS on pre ILK hardware

ILK+ have this register on the PCH. This check was triggering unclaimed
writes.

Signed-off-by: Damien Lespiau <damien.lespiau@intel.com>
Reviewed-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
12 years agodrm/i915: drop buggy write to FDI_RX_CHICKEN register
Daniel Vetter [Wed, 14 Nov 2012 16:47:39 +0000 (17:47 +0100)]
drm/i915: drop buggy write to FDI_RX_CHICKEN register

Jani Nikula noticed that the parentheses are wrong and we & the bit
with the register address instead of the read-back value. He sent a
patch to correct that.

On second look, we write the same register in the previous line, and
the w/a seems to be to set FDI_RX_PHASE_SYNC_POINTER_OVR to enable the
logic, then keep always set FDI_RX_PHASE_SYNC_POINTER_OVR and toggle
FDI_RX_PHASE_SYNC_POINTER_EN before/after enabling the pc transcoder.

So the right things seems to be to simply kill the 2nd write.

Cc: Jani Nikula <jani.nikula@intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
[danvet: Dropped a bogus ~ from the commit message that somehow crept
in.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
12 years agodrm/i915: Use LRI to update the semaphore registers
Chris Wilson [Wed, 14 Nov 2012 09:15:14 +0000 (09:15 +0000)]
drm/i915: Use LRI to update the semaphore registers

The bspec was recently updated to remove the ability to update the
semaphore using the MI_SEMAPHORE_BOX command, the ability to wait upon
the semaphore value remained. Instead the advice is to update the
register using the MI_LOAD_REGISTER_IMM command. In cursory testing,
semaphores continue to function - the question is whether this fixes
some of the deadlocks where the semaphore registers contained stale
values?

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Daniel J Blueman <daniel@quora.org>
Reviewed-by: Ben Widawsky <ben@bwidawsk.net>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
12 years agodrm/i915: add LynxPoint-LP PCH ID
Wei Shun Chang [Mon, 12 Nov 2012 20:54:13 +0000 (18:54 -0200)]
drm/i915: add LynxPoint-LP PCH ID

[pzanoni: rebase, print it's an LP PCH]

Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Tested-by: Josh Triplett <josh@joshtriplett.org>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
12 years agodrm/i915: Optimize DIV_ROUND_CLOSEST call
Jean Delvare [Mon, 12 Nov 2012 13:18:02 +0000 (14:18 +0100)]
drm/i915: Optimize DIV_ROUND_CLOSEST call

DIV_ROUND_CLOSEST is faster if the compiler knows it will only be
dealing with unsigned dividends. This optimization rips 32 bytes of
binary code on x86_64.

Signed-off-by: Jean Delvare <khali@linux-fr.org>
Cc: Guenter Roeck <linux@roeck-us.net>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Daniel Vetter <daniel.vetter@ffwll.ch>
Cc: David Airlie <airlied@linux.ie>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
12 years agodrm/vmwgfx: Make vmw_dmabuf_unreference handle NULL objects
Thomas Hellstrom [Fri, 9 Nov 2012 12:26:15 +0000 (12:26 +0000)]
drm/vmwgfx: Make vmw_dmabuf_unreference handle NULL objects

Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Dmitry Torokhov <dtor@vmware.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
12 years agodrm/vmwgfx: Refactor module load to not require fifo unless fbdev is loaded
Thomas Hellstrom [Fri, 9 Nov 2012 12:26:14 +0000 (12:26 +0000)]
drm/vmwgfx: Refactor module load to not require fifo unless fbdev is loaded

This also fixes a bug where the fence manager was left without irq
enabled when waiting for fences, causing various errors at module
load time

Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Dmitry Torokhov <dtor@vmware.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
12 years agodrm/vmwgfx: Make screen object code not require fifo at init time
Thomas Hellstrom [Fri, 9 Nov 2012 12:26:13 +0000 (12:26 +0000)]
drm/vmwgfx: Make screen object code not require fifo at init time

Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Dmitry Torokhov <dtor@vmware.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
12 years agodrm/vmwgfx: Make overlay code not require fifo at init time
Thomas Hellstrom [Fri, 9 Nov 2012 12:26:12 +0000 (12:26 +0000)]
drm/vmwgfx: Make overlay code not require fifo at init time

Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Dmitry Torokhov <dtor@vmware.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
12 years agodrm/vmwgfx: Enable traces *after* we've hidden SVGA
Thomas Hellstrom [Fri, 9 Nov 2012 12:26:11 +0000 (12:26 +0000)]
drm/vmwgfx: Enable traces *after* we've hidden SVGA

Hiding SVGA seems to trigger a VGA screen clear, and with no
traces dirty it doesn't seem to repaint

Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Dmitry Torokhov <dtor@vmware.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
12 years agodrm/ttm: alter cpu_writers to return -EBUSY in ttm_execbuf_util reservations
Maarten Lankhorst [Tue, 6 Nov 2012 13:39:43 +0000 (14:39 +0100)]
drm/ttm: alter cpu_writers to return -EBUSY in ttm_execbuf_util reservations

This is similar to other platforms that don't allow command submission
to buffers locked on the cpu.

Signed-off-by: Maarten Lankhorst <maarten.lankhorst@canonical.com>
Reviewed-by: Thomas Hellstrom <thellstrom@vmware.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
12 years agodrm/ttm: Optimize reservation slightly
Thomas Hellstrom [Tue, 6 Nov 2012 11:31:51 +0000 (11:31 +0000)]
drm/ttm: Optimize reservation slightly

Reservation locking currently always takes place under the LRU spinlock.
Hence, strictly there is no need for an atomic_cmpxchg call; we can use
atomic_read followed by atomic_write since nobody else will ever reserve
without the lru spinlock held.
At least on Intel this should remove a locked bus cycle on successful
reserve.

Note that thit commit may be obsoleted by the cross-device reservation work.

Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
12 years agodrm/ttm, drm/vmwgfx: Use RCU locking for object lookups v3
Thomas Hellstrom [Tue, 6 Nov 2012 11:31:50 +0000 (11:31 +0000)]
drm/ttm, drm/vmwgfx: Use RCU locking for object lookups v3

The mostly used lookup+get put+potential_destroy path of TTM objects
is converted to use RCU locks. This will substantially decrease the amount
of locked bus cycles during normal operation.
Since we use kfree_rcu to free the objects, no rcu synchronization is needed
at module unload time.

v2: Don't touch include/linux/kref.h
v3: Adapt to kref_get_unless_zero return value change

Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
12 years agokref: Implement kref_get_unless_zero v3
Thomas Hellstrom [Tue, 6 Nov 2012 11:31:49 +0000 (11:31 +0000)]
kref: Implement kref_get_unless_zero v3

This function is intended to simplify locking around refcounting for
objects that can be looked up from a lookup structure, and which are
removed from that lookup structure in the object destructor.
Operations on such objects require at least a read lock around
lookup + kref_get, and a write lock around kref_put + remove from lookup
structure. Furthermore, RCU implementations become extremely tricky.
With a lookup followed by a kref_get_unless_zero *with return value check*
locking in the kref_put path can be deferred to the actual removal from
the lookup structure and RCU lookups become trivial.

v2: Formatting fixes.
v3: Invert the return value.

Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
12 years agodrm: Make hashtab rcu-safe
Thomas Hellstrom [Tue, 6 Nov 2012 11:31:48 +0000 (11:31 +0000)]
drm: Make hashtab rcu-safe

TTM base objects will be the first consumer.

Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
12 years agodrm/ttm: remove sync_arg from driver functions
Maarten Lankhorst [Fri, 12 Oct 2012 15:04:00 +0000 (15:04 +0000)]
drm/ttm: remove sync_arg from driver functions

Signed-off-by: Maarten Lankhorst <maarten.lankhorst@canonical.com>
Reviewed-By: Thomas Hellstrom <thellstrom@vmware.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
12 years agodrm/ttm: remove sync_obj_arg from ttm_bo_move_accel_cleanup
Maarten Lankhorst [Fri, 12 Oct 2012 15:03:11 +0000 (15:03 +0000)]
drm/ttm: remove sync_obj_arg from ttm_bo_move_accel_cleanup

Signed-off-by: Maarten Lankhorst <maarten.lankhorst@canonical.com>
Reviewed-By: Thomas Hellstrom <thellstrom@vmware.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
12 years agodrm/ttm: remove sync_obj_arg member
Maarten Lankhorst [Fri, 12 Oct 2012 15:02:19 +0000 (15:02 +0000)]
drm/ttm: remove sync_obj_arg member

vmwgfx was its only user and always sets it to the same..

Signed-off-by: Maarten Lankhorst <maarten.lankhorst@canonical.com>
Reviewed-By: Thomas Hellstrom <thellstrom@vmware.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
12 years agodrm/vmwgfx: remove use of fence_obj_args
Maarten Lankhorst [Fri, 12 Oct 2012 15:01:43 +0000 (15:01 +0000)]
drm/vmwgfx: remove use of fence_obj_args

It's always hardcoded to the same value.

Signed-off-by: Maarten Lankhorst <maarten.lankhorst@canonical.com>
Reviewed-By: Thomas Hellstrom <thellstrom@vmware.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
12 years agoDRM/KMS: Add Bail-Out Conditions for Loop.
Egbert Eich [Sat, 13 Oct 2012 11:36:14 +0000 (11:36 +0000)]
DRM/KMS: Add Bail-Out Conditions for Loop.

When trying to obtain an accurate timestamp for the last vsync interrupt
in vblank_disable_and_save() we loop until the vsync counter after reading
the time stamp is identical to the one before.
In the case where no hardware timestamp can be obtained there is probably
no point in trying to make sure we remain within the same vsync during
the time we obtain the counter.
Furthermore we should make sure there's an 'emergency exit' so that we
don't end up in an endless loop when the driver get_vblank_timestamp()
function doesn't manage to return within the same vsync.
This may happen when this function prints out debugging information over
a slow (ie serial) line.

Signed-off-by: Egbert Eich <eich@suse.de>
Signed-off-by: Dave Airlie <airlied@redhat.com>
12 years agodrm: don't unnecessarily enable the polling work
Daniel Vetter [Wed, 24 Oct 2012 13:35:50 +0000 (13:35 +0000)]
drm: don't unnecessarily enable the polling work

... by properly checking connector->polled. This doesn't matter too
much because the polling work itself gets this slightly more right and
doesn't set repoll if there's nothing to do. But we can do better.

v2: Chris Wilson noticed that I broke polling, since repoll will never
ever be set true. Fix this up, and simplify the logic a bit while at
it.

Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Dave Airlie <airlied@redhat.com>
12 years agovga_switcheroo: Drop unused include and unused variables.
Igor Murzov [Thu, 25 Oct 2012 13:09:01 +0000 (13:09 +0000)]
vga_switcheroo: Drop unused include and unused variables.

Signed-off-by: Igor Murzov <e-mail@date.by>
Signed-off-by: Dave Airlie <airlied@redhat.com>
12 years agodrm/nouveau: free memory allocated with alloc_apertures()
Tommi Rantala [Fri, 9 Nov 2012 09:19:40 +0000 (09:19 +0000)]
drm/nouveau: free memory allocated with alloc_apertures()

Fix a memory leak by deallocating the memory we got from
alloc_apertures().

Signed-off-by: Tommi Rantala <tt.rantala@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
12 years agodrm/radeon: check alloc_apertures() success in radeon_kick_out_firmware_fb()
Tommi Rantala [Fri, 9 Nov 2012 09:19:39 +0000 (09:19 +0000)]
drm/radeon: check alloc_apertures() success in radeon_kick_out_firmware_fb()

Check for alloc_apertures() memory allocation failure, and propagate an
error code in case the allocation failed.

Signed-off-by: Tommi Rantala <tt.rantala@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
12 years agodrm/mgag200: remove unneeded aper->count assignment after alloc_apertures()
Tommi Rantala [Fri, 9 Nov 2012 09:19:38 +0000 (09:19 +0000)]
drm/mgag200: remove unneeded aper->count assignment after alloc_apertures()

alloc_apertures() already does the assignment for us, so assigning the
count member after the alloc_apertures() call is not needed.

Signed-off-by: Tommi Rantala <tt.rantala@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
12 years agodrm/mgag200: free memory allocated with alloc_apertures()
Tommi Rantala [Fri, 9 Nov 2012 09:19:37 +0000 (09:19 +0000)]
drm/mgag200: free memory allocated with alloc_apertures()

Fix a memory leak by deallocating the memory we got from
alloc_apertures().

Signed-off-by: Tommi Rantala <tt.rantala@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
12 years agodrm/mgag200: check alloc_apertures() success in mga_vram_init()
Tommi Rantala [Fri, 9 Nov 2012 09:19:36 +0000 (09:19 +0000)]
drm/mgag200: check alloc_apertures() success in mga_vram_init()

Check for alloc_apertures() memory allocation failure, and propagate an
error code in case the allocation failed.

Signed-off-by: Tommi Rantala <tt.rantala@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
12 years agodrm/cirrus: check alloc_apertures() success in cirrus_kick_out_firmware_fb()
Tommi Rantala [Fri, 9 Nov 2012 09:19:35 +0000 (09:19 +0000)]
drm/cirrus: check alloc_apertures() success in cirrus_kick_out_firmware_fb()

Check for alloc_apertures() memory allocation failure, and propagate an
error code in case the allocation failed.

Signed-off-by: Tommi Rantala <tt.rantala@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
12 years agodrm/ttm: remove ttm_mem_global->queue
Marcin Slusarz [Tue, 6 Nov 2012 21:49:54 +0000 (21:49 +0000)]
drm/ttm: remove ttm_mem_global->queue

It's unused.

Signed-off-by: Marcin Slusarz <marcin.slusarz@gmail.com>
Reviewed-by: Thomas Hellstrom <thellstrom@vmware.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
12 years agodrm/ttm: remove ttm_bo_device->nice_mode
Marcin Slusarz [Tue, 6 Nov 2012 21:49:53 +0000 (21:49 +0000)]
drm/ttm: remove ttm_bo_device->nice_mode

It's unused.

Signed-off-by: Marcin Slusarz <marcin.slusarz@gmail.com>
Reviewed-by: Thomas Hellstrom <thellstrom@vmware.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
12 years agodrm/ttm: remove ttm_buffer_object->buffer_start
Marcin Slusarz [Tue, 6 Nov 2012 21:49:51 +0000 (21:49 +0000)]
drm/ttm: remove ttm_buffer_object->buffer_start

All drivers set it to 0 and nothing uses it.

Signed-off-by: Marcin Slusarz <marcin.slusarz@gmail.com>
Reviewed-by: Thomas Hellstrom <thellstrom@vmware.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
12 years agodrm/radeon: Use hweight32
Akinobu Mita [Fri, 9 Nov 2012 12:10:41 +0000 (12:10 +0000)]
drm/radeon: Use hweight32

Use hweight32 instead of counting for each bit

Signed-off-by: Akinobu Mita <akinobu.mita@gmail.com>
Cc: David Airlie <airlied@linux.ie>
Cc: dri-devel@lists.freedesktop.org
Signed-off-by: Dave Airlie <airlied@redhat.com>
12 years agodrm: use memchr_inv()
Akinobu Mita [Fri, 9 Nov 2012 12:10:42 +0000 (12:10 +0000)]
drm: use memchr_inv()

Use memchr_inv() to check the specified memory region is filled with zero.

Signed-off-by: Akinobu Mita <akinobu.mita@gmail.com>
Cc: David Airlie <airlied@linux.ie>
Cc: dri-devel@lists.freedesktop.org
Signed-off-by: Dave Airlie <airlied@redhat.com>
12 years agodrm: add support for monotonic vblank timestamps
Imre Deak [Tue, 23 Oct 2012 18:53:26 +0000 (18:53 +0000)]
drm: add support for monotonic vblank timestamps

Jumps in the vblank and page flip event timestamps cause trouble for
clients, so we should avoid them. The timestamp we get currently with
gettimeofday can jump, so use instead monotonic timestamps.

For backward compatibility use a module flag to revert back to using
gettimeofday timestamps. Add also a DRM_CAP_TIMESTAMP_MONOTONIC flag
that is simply a read only version of the module flag, so that clients
can query this without depending on sysfs.

Signed-off-by: Imre Deak <imre.deak@intel.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
12 years agodrm: use monotonic time in drm_calc_vbltimestamp_from_scanoutpos
Imre Deak [Tue, 23 Oct 2012 18:53:25 +0000 (18:53 +0000)]
drm: use monotonic time in drm_calc_vbltimestamp_from_scanoutpos

For measuring duration we want to avoid that our start/end timestamps
jump, so use monotonic instead of real time for that.

Signed-off-by: Imre Deak <imre.deak@intel.com>
Reviewed-by: mario.kleiner
Signed-off-by: Dave Airlie <airlied@redhat.com>
12 years agodrm: don't poll forced connectors
Daniel Vetter [Tue, 23 Oct 2012 18:23:38 +0000 (18:23 +0000)]
drm: don't poll forced connectors

Otherwise if the detect callback reports a different state than what
the user forced (rather likely), we continously annoy userspace about
a hotplug uevent.

Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
12 years agodrm: don't start the poll engine in probe_single_connector
Daniel Vetter [Tue, 23 Oct 2012 18:23:36 +0000 (18:23 +0000)]
drm: don't start the poll engine in probe_single_connector

Actually there's a reason this stuff is there, and it's called

commit e58f637bb96d5a0ae0919b9998b891d1ba7e47c9
Author: Chris Wilson <chris@chris-wilson.co.uk>
Date:   Fri Aug 20 09:13:36 2010 +0100

    drm/kms: Add a module parameter to disable polling

The idea has been that users can enable/disable polling at runtime. So
the quick hack has been to just re-enable the output polling if xrandr
asks for the latest state of the connectors.

The problem with that hack is that when we force connectors to another
state than what would be detected, we nicely ping-pong:
- Userspace calls probe, gets the forced state, but polling starts
  again.
- Polling notices that the state is actually different, wakes up
  userspace.
- Repeat.

As that commit already explains, the right fix would be to make the
locking more fine-grained, so that hotplug detection on one output
does not interfere with cursor updates on another crtc.

But that is way too much work. So let's just safe this gross hack by
caching the last-seen state of drm_kms_helper_poll for that driver,
and only fire up the poll engine again if it changed from off to on.

v2: Fixup the edge detection of drm_kms_helper_poll.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=49907
Tested-by: Tvrtko Ursulin <tvrtko.ursulin@onelan.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
12 years agodrm: properly init/reset connector status
Daniel Vetter [Tue, 23 Oct 2012 18:23:35 +0000 (18:23 +0000)]
drm: properly init/reset connector status

This can help drivers to make somewhat intelligent decisions in their
->detect callback: If the connector is hpd capable and in the unknown
state, the driver needs to force a full detect cycle. Otherwise it
could just (if it chooses so) to update the connector state from it's
hpd handler directly, and always return that in the ->detect callback.

Atm only drm/i915 calls drm_mode_config_reset at resume time, so other
drivers would need to add that call first before using this facility.

Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
12 years agodrm: run the hpd irq event code directly
Daniel Vetter [Tue, 23 Oct 2012 18:23:34 +0000 (18:23 +0000)]
drm: run the hpd irq event code directly

All drivers already have a work item to run the hpd code, so we don't
need to launch a new one in the helper code. Dave Airlie mentioned
that the cancel+re-queue might paper over DP related hpd ping-pongs,
hence why this is split out.

Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
12 years agodrm: handle HPD and polled connectors separately
Daniel Vetter [Tue, 23 Oct 2012 18:23:33 +0000 (18:23 +0000)]
drm: handle HPD and polled connectors separately

Instead of reusing the polling code for hpd handling, split them up.
This has a few consequences:
- Don't touch HPD capable connectors in the poll loop.
- Only touch HPD capable connectors in drm_helper_hpd_irq_event.
- We could run the HPD handling directly (because all callers already
  use their own work item), but for easier bisect that happens in it's
  own patch.

The ultimate goal is that drivers grow some smarts about which
connectors have received a hotplug event and only call the detect code
of that connector. But that's a second step.

v2: s/hdp/hpd/, noticed by Adam Jackson. I can't type.

v3: Split out the work item removal as requested by Dave Airlie. This
results in a temporary mode_config.hpd_irq_work item to keep things
the same.

v4: In the hpd_irq_event handler don't bail out if other bits than HPD
are set. This is useful where e.g. hpd is unreliably, but mostly
works. Drivers can then set both HPD and POLL flags, and users get the
best of both worlds: Quick hotplug feedback if the hpd works, but
still reliable detection with the polling. The poll loop already works
the same, and doesn't bail if HPD is set.

Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
12 years agodrm: extract drm_kms_helper_hotplug_event
Daniel Vetter [Tue, 23 Oct 2012 18:23:32 +0000 (18:23 +0000)]
drm: extract drm_kms_helper_hotplug_event

Useful if drivers want to be slightly more clever about hotplug
handling.

Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>