drm/exynos: use ERR_PTR instead of NULL in exynos_drm_gem_init

For more correct error information.

Signed-off-by: Joonyoung Shim <jy0922.shim@samsung.com>
Signed-off-by: Inki Dae <inki.dae@samsung.com>

drm/exynos: remove unused fields of struct exynos_drm_gem_buf

The userptr, write and pfnmap of struct exynos_drm_gem_buf are not used
anywhere.

Signed-off-by: Joonyoung Shim <jy0922.shim@samsung.com>
Signed-off-by: Inki Dae <inki.dae@samsung.com>

drm/exynos: stop copying sg table

Already struct exynos_drm_gem_buf has pages of the buffer, so we don't
need to copy from sg table of the buffer to sg table of dma-buf
attachment, just can make sg table from pages of the buffer.

Signed-off-by: Joonyoung Shim <jy0922.shim@samsung.com>
Signed-off-by: Inki Dae <inki.dae@samsung.com>

drm/exynos: remove function exynos_drm_gem_map_buf

The exynos_drm_gem_map_buf can be merged in exynos_drm_gem_fault.

Signed-off-by: Joonyoung Shim <jy0922.shim@samsung.com>
Signed-off-by: Inki Dae <inki.dae@samsung.com>

drm/exynos: remove mutex locking in pagefault handler

There is no reason to use mutex locking in pagefault handler.

Signed-off-by: Joonyoung Shim <jy0922.shim@samsung.com>
Signed-off-by: Inki Dae <inki.dae@samsung.com>

drm/exynos: remove function convert_to_vm_err_msg

The convert_to_vm_err_msg is called just once by exynos_drm_gem_fault,
so it's simple not to use the function.

Signed-off-by: Joonyoung Shim <jy0922.shim@samsung.com>
Signed-off-by: Inki Dae <inki.dae@samsung.com>

drm/exynos: stop using sgtable in page fault handler

Already struct exynos_drm_gem_buf has pages of the buffer when buffer is
created, so just can use pages in page fault handler, we don't have to
make sgtable of the buffer. But this needs to construct pages of the
buffer that is imported from dma-buf prime.

Signed-off-by: Joonyoung Shim <jy0922.shim@samsung.com>
Signed-off-by: Inki Dae <inki.dae@samsung.com>

drm/exynos: remove struct exynos_drm_encoder layer

struct exynos_drm_encoder was justing wrapping struct drm_encoder, it had
only a drm_encoder member and the internal exynos_drm_encoders ops that
was directly mapped to the drm_encoder helper funcs.

So now exynos DRM uses struct drm_encoder directly, this removes
completely the struct exynos_drm_encoder.

v2: add empty .mode_fixup() and .mode_set() to DSI and DPI to avoid null
pointer.

Signed-off-by: Gustavo Padovan <gustavo.padovan@collabora.co.uk>
Signed-off-by: Inki Dae <inki.dae@samsung.com>

drm/exynos: fold encoder setup into exynos_drm_load()

As we are removing the exynos encoder move the encoder setup operation
directly inside the exynos_drm_load()

Signed-off-by: Gustavo Padovan <gustavo.padovan@collabora.co.uk>
Signed-off-by: Inki Dae <inki.dae@samsung.com>

drm/exynos: remove exynos_drm_create_enc_conn()

This functions was just hiding the encoder and connector creation in
a way that was less clean than if we get rid of it. For example,
exynos_encoder ops had .create_connector() defined only because we were
handing off the encoder and connector creation to
exynos_drm_create_enc_conn(). Without this function we can directly call
the create_connector function internally in the code, without the need of
any vtable access.

It also does some refactoring in the code like creating a bind function
for dpi devices.

Signed-off-by: Gustavo Padovan <gustavo.padovan@collabora.co.uk>
Signed-off-by: Inki Dae <inki.dae@samsung.com>

drm/exynos: remove exynos_encoder's .commit() op

.commit() is not used anymore, Exynos encoders now follow the
.enable()/.disable() semantics from drm atomic core, so remove this
callback.

Signed-off-by: Gustavo Padovan <gustavo.padovan@collabora.co.uk>
Signed-off-by: Inki Dae <inki.dae@samsung.com>

drm/exynos: remove extra call to exynos_dp_commit()

exynos_dp_commit() was getting called twice by exynos encoder core, once
inside the .enable() call and another time by .commit() itself.

The remove of the second call caused the wake of a bug, the operations
orders inside exynos_dp_commit was wrong and we had to move
exynos_dp_start_video() to be the last operation in there.

Signed-off-by: Gustavo Padovan <gustavo.padovan@collabora.co.uk>
Signed-off-by: Inki Dae <inki.dae@samsung.com>

drm/exynos: remove extra call to hdmi_commit()

hdmi_commit() was getting called twice by exynos encoder core, once inside
the .enable() call and another time by .commit() itself.

Signed-off-by: Gustavo Padovan <gustavo.padovan@collabora.co.uk>
Signed-off-by: Inki Dae <inki.dae@samsung.com>

drm/exynos: remove struct exynos_drm_display

This struct was just representing encoder information, it was a member of
struct exynos_drm_encoder, so any code trying to access encoder data would
have to go through the encoder struct, get the display struct and then get
the data it want.

During this patchset we also realized that the only data
exynos_drm_encoder needs to store is the drm_encoder parent and the
exynos_drm_encoder_ops.

Signed-off-by: Gustavo Padovan <gustavo.padovan@collabora.co.uk>
Signed-off-by: Inki Dae <inki.dae@samsung.com>

drm/exynos: simplify calculation of possible CRTCs

All CRTCs can only be LCD, HDMI or VIDI, so basically all CRTCs will be a
possible CRTCs. This patch removes an extra function with switch that was
only checking if the CRTC type was one of those three above.

Signed-off-by: Gustavo Padovan <gustavo.padovan@collabora.co.uk>
Signed-off-by: Inki Dae <inki.dae@samsung.com>

drm/exynos: remove unused .remove() and .check_mode() ops from display

These two display_ops are not used anywhere, remove them.

Signed-off-by: Gustavo Padovan <gustavo.padovan@collabora.co.uk>
Signed-off-by: Inki Dae <inki.dae@samsung.com>

drm/exynos: remove wrappers for phy_power_{on,off}

phy_power_on() and phy_power_off() already checks for NULL pointer.
This patch removes the wrappers exynos_dp_phy_init() and
exynos_dp_phy_exit() since the only think they were doing was a check for
NULL phy.

Signed-off-by: Gustavo Padovan <gustavo.padovan@collabora.co.uk>
Signed-off-by: Inki Dae <inki.dae@samsung.com>

drm/exynos: split display's .dpms() into .enable() and .disable()

The DRM Core doesn't have a dpms() operation anymore, everything
now is enable() or disable().

Signed-off-by: Gustavo Padovan <gustavo.padovan@collabora.co.uk>
Signed-off-by: Inki Dae <inki.dae@samsung.com>

drm/exynos: return return value of exynos_crtc->enable_vblank

Instead of blindly ignore the return value of enable_vblank return it
to the upper DRM layer for error handling.

Suggested-by: Joonyoung Shim <jy0922.shim@samsung.com>
Signed-off-by: Gustavo Padovan <gustavo.padovan@collabora.co.uk>
Reviewed-by: Joonyoung Shim <jy0922.shim@samsung.com>
Signed-off-by: Inki Dae <inki.dae@samsung.com>

drm/exynos: unify exynos_drm_plane names with drm core

Rename crtc_{widht,height} to crtc_{w,h} and src_{width,height} to
src_{w,h} to make it similar to the atomic state names.

Signed-off-by: Gustavo Padovan <gustavo.padovan@collabora.co.uk>
Reviewed-by: Joonyoung Shim <jy0922.shim@samsung.com>
Signed-off-by: Inki Dae <inki.dae@samsung.com>

drm/exynos: remove unused fields from struct exynos_drm_plane

Now after the move to use drm_plane_state directly struct drm_plane_state
has many unused fields, along with others that weren't used before the
plane state change. Thus remove them all.

Signed-off-by: Gustavo Padovan <gustavo.padovan@collabora.co.uk>
Reviewed-by: Joonyoung Shim <jy0922.shim@samsung.com>
Signed-off-by: Inki Dae <inki.dae@samsung.com>

drm/exynos: use drm atomic state directly

For some fields the use of struct exynos_drm_plane filled with data from
the plane state just creates a source of duplicated information and
overhead.  Here we change the crtc drivers to access the plane state
directly simplifying the code by not relying on a exynos internal struct.

Signed-off-by: Gustavo Padovan <gustavo.padovan@collabora.co.uk>
Reviewed-by: Joonyoung Shim <jy0922.shim@samsung.com>
Signed-off-by: Inki Dae <inki.dae@samsung.com>

drm/exynos: pass struct exynos_drm_plane in update/enable

We already have the plane pointer in before calling .update_plane() or
disable_plane() so pass it directly to those calls avoiding a new
conversion from zpos to struct exynos_drm_plane.

v2: don't remove check for suspended in FIMD (comment by Joonyoung)

Signed-off-by: Gustavo Padovan <gustavo.padovan@collabora.co.uk>
Reviewed-by: Joonyoung Shim <jy0922.shim@samsung.com>
Signed-off-by: Inki Dae <inki.dae@samsung.com>

drm/exynos: rename win_commit/disable to atomic-like names

Rename win_commit() helper to update_plane() and win_disable() to
disable_plane().

Signed-off-by: Gustavo Padovan <gustavo.padovan@collabora.co.uk>
Reviewed-by: Joonyoung Shim <jy0922.shim@samsung.com>
Signed-off-by: Inki Dae <inki.dae@samsung.com>

drm/exynos: remove duplicated check for suspend

The same check is placed twice in fimd/decon_update_plane(), remove
one of them.

Signed-off-by: Gustavo Padovan <gustavo.padovan@collabora.co.uk>
Reviewed-by: Joonyoung Shim <jy0922.shim@samsung.com>
Signed-off-by: Inki Dae <inki.dae@samsung.com>

drm/exynos: use KMS version of DRM vblanks functions

Get rid of legacy DRM vblank function that are less clear to use.
The new ones basically requires only the crtc as parameters.

It also clean ups exynos_drm_crtc_finish_pageflip() parameters as a
consequence.

Signed-off-by: Gustavo Padovan <gustavo.padovan@collabora.co.uk>
Reviewed-by: Joonyoung Shim <jy0922.shim@samsung.com>
Signed-off-by: Inki Dae <inki.dae@samsung.com>

drm/exynos: pass the correct pipe number

Instead of giving -1 to as arg to  drm_send_vblank_event() pass the
correct pipe number to it.

Signed-off-by: Gustavo Padovan <gustavo.padovan@collabora.co.uk>
Reviewed-by: Joonyoung Shim <jy0922.shim@samsung.com>
Signed-off-by: Inki Dae <inki.dae@samsung.com>

drm/exynos: clear channels only when iommu is enabled

This is simplest solution about reported problem[1]. It's no problem to
clear channel only when iommu is enabled, if we consider that we cannot
recognize iommu errors when iommu is disabled and it have been valid
until now. But this cannot be nice solution.

[1] https://lkml.org/lkml/2015/7/21/404

Reported-by: Krzysztof Kozlowski <k.kozlowski@samsung.com>
Signed-off-by: Joonyoung Shim <jy0922.shim@samsung.com>
Signed-off-by: Inki Dae <inki.dae@samsung.com>

drm/exynos: remove drm_iommu_attach_device_if_possible

Already drm_iommu_attach_device checks whether support iommu internally.
It should clear channels always regardless iommu support. We didn't know
because we can detect the problem when iommu is enabled, so we don't
have to use drm_iommu_attach_device_if_possible and then we can remove
drm_iommu_attach_device_if_possible and clear_channels function pointer.

Signed-off-by: Joonyoung Shim <jy0922.shim@samsung.com>
Tested-by: Marek Szyprowski <m.szyprowski@samsung.com>
Signed-off-by: Inki Dae <inki.dae@samsung.com>

drm/exynos: move order to register vidi kms driver

The vidi is virtual kms driver and now it is registered earlier than
actual hw kms drivers, so it will occupy crtc index 0. Some users
assume the condition yet that actual hw kms driver has crtc index 0.
It may or may not be matter but let's arrange register order.

Signed-off-by: Joonyoung Shim <jy0922.shim@samsung.com>
Signed-off-by: Inki Dae <inki.dae@samsung.com>

drm/exynos/mixer: replace MXR_INT_EN register cache with flag

Driver uses only VSYNC interrupts, so we need to cache VSYNC bit state only.

Signed-off-by: Andrzej Hajda <a.hajda@samsung.com>
Reviewed-by: Joonyoung Shim <jy0922.shim@samsung.com>
Signed-off-by: Inki Dae <inki.dae@samsung.com>

drm/exynos/mixer: simplify poweron flag

The driver uses bool protected by mutex to track power state.
The patch replaces this combo with single bit and atomic bitops.

Signed-off-by: Andrzej Hajda <a.hajda@samsung.com>
Reviewed-by: Joonyoung Shim <jy0922.shim@samsung.com>
Signed-off-by: Inki Dae <inki.dae@samsung.com>

drm/exynos: remove unnecessary checking to support iommu

Already drm_iommu_attach_device and drm_iommu_detach_device check
whether support iommu internally, so we don't have to call
is_drm_iommu_supported before call them.

Signed-off-by: Joonyoung Shim <jy0922.shim@samsung.com>
Tested-by: Marek Szyprowski <m.szyprowski@samsung.com>
Signed-off-by: Inki Dae <inki.dae@samsung.com>

drm/exynos: remove to use ifdef CONFIG_ARM_DMA_USE_IOMMU

If CONFIG_ARM_DMA_USE_IOMMU is disable, CONFIG_DRM_EXYNOS_IOMMU also is
disable. When CONFIG_DRM_EXYNOS_IOMMU is disable,
is_drm_iommu_supported() returns always false, so we can remove to use
ifdef CONFIG_ARM_DMA_USE_IOMMU in is_drm_iommu_supported().

Signed-off-by: Joonyoung Shim <jy0922.shim@samsung.com>
Tested-by: Marek Szyprowski <m.szyprowski@samsung.com>
Signed-off-by: Inki Dae <inki.dae@samsung.com>

drm/exynos: gsc: Handles the combination of rotation and flip

The unique results of all the combination of rotation and flip can
be represented by just 8 states. This patch handles all the combination
correctly.

Signed-off-by: Hyungwon Hwang <human.hwang@samsung.com>
Signed-off-by: Inki Dae <inki.dae@samsung.com>

drm/exynos/hdmi: remove hdmi_v14_conf struct

The patch removes intermediate struct for HDMIv14 register configuration,
instead registry values are calculated on the fly.

Signed-off-by: Andrzej Hajda <a.hajda@samsung.com>
Reviewed-by: Joonyoung Shim <jy0922.shim@samsung.com>
Signed-off-by: Inki Dae <inki.dae@samsung.com>

drm/exynos/hdmi: remove hdmi_v13_conf struct

The patch removes intermediate struct for HDMIv13 register configuration,
instead registry values are calculated on the fly.

Signed-off-by: Andrzej Hajda <a.hajda@samsung.com>
Reviewed-by: Joonyoung Shim <jy0922.shim@samsung.com>
Signed-off-by: Inki Dae <inki.dae@samsung.com>

drm/exynos/hdmi: remove redundant configuration fields

The patch removes redundant fields from hdmi_conf_regs. Their values
can be calculated from current_mode. This patch is the first step to remove
whole hdmi_conf_regs structure.

Signed-off-by: Andrzej Hajda <a.hajda@samsung.com>
Reviewed-by: Joonyoung Shim <jy0922.shim@samsung.com>
Signed-off-by: Inki Dae <inki.dae@samsung.com>

drm/exynos/hdmi: add driver data pointer to private context

The patch replaces duplicated driver data fields in private context with
pointer to driver data. It also simplifies driver data lookup code.

Signed-off-by: Andrzej Hajda <a.hajda@samsung.com>
Reviewed-by: Joonyoung Shim <jy0922.shim@samsung.com>
Signed-off-by: Inki Dae <inki.dae@samsung.com>

drm/exynos/hdmi: remove private lock code

Most of the code is called by drm core framework, so it is already synchronized.
The only async function is irq routine which only calls drm framework so it
does not need to be synchronized.

Signed-off-by: Andrzej Hajda <a.hajda@samsung.com>
Reviewed-by: Joonyoung Shim <jy0922.shim@samsung.com>
Signed-off-by: Inki Dae <inki.dae@samsung.com>

drm/exynos/hdmi: Simplify HPD gpio handling

GPIO is tested only in hdmi_detect, so there is no reason to set it in
other places and to preserve its value in context.

Signed-off-by: Andrzej Hajda <a.hajda@samsung.com>
Reviewed-by: Joonyoung Shim <jy0922.shim@samsung.com>
Signed-off-by: Inki Dae <inki.dae@samsung.com>

drm/exynos/hdmi: remove old platform data code

s5p_hdmi_platform_data were used before device tree introduction.
As HDMI driver is DT only we can drop this struct completely.

Signed-off-by: Andrzej Hajda <a.hajda@samsung.com>
Reviewed-by: Joonyoung Shim <jy0922.shim@samsung.com>
Signed-off-by: Inki Dae <inki.dae@samsung.com>

drm/exynos/fimc: fix runtime pm support

Once pm_runtime_set_active() gets called, the kernel assumes that given
device has already enabled runtime pm and will call pm_runtime_suspend()
without matching pm_runtime_resume(). In case of DRM FIMC IPP driver,
this will result in calling clk_disable() without respective call to
clk_enable(). This patch removes call to pm_runtime_set_active() to
ensure that pm_runtime_suspend/resume calls will match.

Signed-off-by: Marek Szyprowski <m.szyprowski@samsung.com>
Signed-off-by: Inki Dae <inki.dae@samsung.com>

drm/exynos/mixer: always update INT_EN cache

INT_EN cache field was updated only by mixer_enable_vblank.
The patch adds update also by mixer_disable_vblank function.

Signed-off-by: Andrzej Hajda <a.hajda@samsung.com>
Reviewed-by: Joonyoung Shim <jy0922.shim@samsung.com>
Signed-off-by: Inki Dae <inki.dae@samsung.com>

drm/exynos/mixer: correct vsync configuration sequence

Specification advises to clear vsync indicator before configuring vsync.

Signed-off-by: Andrzej Hajda <a.hajda@samsung.com>
Reviewed-by: Joonyoung Shim <jy0922.shim@samsung.com>
Signed-off-by: Inki Dae <inki.dae@samsung.com>

drm/exynos/mixer: fix interrupt clearing

The driver used incorrect flags to clear interrupt status.
The patch fixes it.

Signed-off-by: Andrzej Hajda <a.hajda@samsung.com>
Reviewed-by: Joonyoung Shim <jy0922.shim@samsung.com>
Signed-off-by: Inki Dae <inki.dae@samsung.com>

drm/exynos/hdmi: fix edid memory leak

edid returned by drm_get_edid should be freed.
The patch fixes it.

Signed-off-by: Andrzej Hajda <a.hajda@samsung.com>
Reviewed-by: Joonyoung Shim <jy0922.shim@samsung.com>
Signed-off-by: Inki Dae <inki.dae@samsung.com>

drm/exynos: gsc: fix wrong bitwise operation for swap detection

The bits for rotation are not used as exclusively. So GSC_IN_ROT_270 can
not be used for swap detection. The definition of it is same with
GSC_IN_ROT_MASK. It is enough to check GSC_IN_ROT_90 bit is set or not to
check whether width / height size swapping is needed.

Signed-off-by: Hyungwon Hwang <human.hwang@samsung.com>
Signed-off-by: Inki Dae <inki.dae@samsung.com>

Merge tag 'drm-amdkfd-next-fixes-2015-08-05' of git://people.freedesktop.org/~gabbayo/linux into drm-next

Two small bug fixes for the code you pulled for 4.3:

- Used a SHIFT define instead of a MASK define to check if a bit is turned on
  when destroying hqd. Luckily, this is in gfx7 interface file with amdgpu,
  which was used only for bring-up purposes of amdgpu, so no real effect on
  a running system

- Used a logical AND instead of a bitwise AND operator, when initializing
  sdma virtual memory when using SDMA queues

* tag 'drm-amdkfd-next-fixes-2015-08-05' of git://people.freedesktop.org/~gabbayo/linux:
  drm/amdkfd: fix bug when initializing sdma vm
  drm/amdgpu: fix bug when amdkfd destroys hqd

Merge branch 'drm-sti-next-atomic-2015-08-11' of git.linaro.org/people/benjamin.gaignard/kernel into drm-next

This serie of patches fix minor bugs around how driver sub-components are
bind and planes z-ordering.
The main part is about atomic support: using more atomic helpers allow us
to simplify the code (~300 lines removed) and to ahve a better match between
drm concepts (planes and crtc) and hardware split.

[airlied: fixed up conflict in atomic code]

* 'drm-sti-next-atomic-2015-08-11' of http://git.linaro.org/people/benjamin.gaignard/kernel:
  drm/sti: atomic crtc/plane update
  drm/sti: rename files and functions
  drm/sti: code clean up
  drm/sti: fix dynamic z-ordering
  drm: sti: fix sub-components bind

Merge tag 'topic/drm-misc-2015-08-13' of git://anongit.freedesktop.org/drm-intel into drm-next

Final drm-misc pull for 4.3:
- fbdev emulation Kconfig option for everyone thanks to Archit. It's not
  everything yet bit this is fairly tricky since it spawns all drivers.
- vgaarb & vgaswitcheroo polish from Thierry
- some drm_irq.c cleanups (Thierry)
- struct_mutex crusade from me
- more fbdev panic handling removal
- various things all over in drm core&helpers

* tag 'topic/drm-misc-2015-08-13' of git://anongit.freedesktop.org/drm-intel: (65 commits)
  drm/atomic: Use KMS VBLANK API
  drm/irq: Document return values more consistently
  drm/irq: Make pipe unsigned and name consistent
  drm/irq: Check for valid VBLANK before dereference
  drm/irq: Remove negative CRTC index special-case
  drm/plane: Remove redundant extern
  drm/plane: Use consistent data types for format count
  vga_switcheroo: Remove unnecessary checks
  vga_switcheroo: Wrap overly long lines
  vga_switcheroo: Use pr_fmt()
  vga_switcheroo: Cleanup header comment
  vga_switcheroo: Use pr_*() instead of printk()
  vgaarb: Fix a few checkpatch errors and warnings
  vgaarb: Use vgaarb: prefix consistently in messages
  vgaarb: Stop complaining about absent devices
  drm/atomic: fix null pointer access to mode_fixup callback
  drm/i915: Use CONFIG_DRM_FBDEV_EMULATION
  drm/core: Set mode to NULL when connectors in a set drops to 0.
  drm/atomic: Call ww_acquire_done after check phase is complete
  drm/atomic: Paper over locking WARN in default_state_clear
  ...

drm/atomic: Use KMS VBLANK API

Instead of using the legacy VBLANK API, use the new KMS API. This is
part of an effort to convert all existing users so that the KMS API can
be changed to properly use per-CRTC data.

Signed-off-by: Thierry Reding <treding@nvidia.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

drm/irq: Document return values more consistently

Some of the functions are documented inconsistently. Add Returns:
sections where missing and use consistent style to describe the return
value.

Signed-off-by: Thierry Reding <treding@nvidia.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

drm/irq: Make pipe unsigned and name consistent

Name all references to the pipe number (CRTC index) consistently to make
it easier to distinguish which is a pipe number and which is a pointer
to struct drm_crtc.

While at it also make all references to the pipe number unsigned because
there is no longer any reason why it should ever be negative.

Signed-off-by: Thierry Reding <treding@nvidia.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

drm/irq: Check for valid VBLANK before dereference

When accessing the array of per-CRTC VBLANK structures we must always
check that the index into the array is valid before dereferencing to
avoid crashing.

Signed-off-by: Thierry Reding <treding@nvidia.com>
[danvet: Squash in my own whitespace ocd fixup in drm_vblank_count.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

drm/irq: Remove negative CRTC index special-case

The drm_send_vblank_event() function treats negative CRTC indices as
meaning that a driver doesn't have proper VBLANK handling. This is the
only place where DRM needs negative CRTC indices, so in order to enable
subsequent cleanup, remove this special case and replace it by the more
obvious check for whether or not VBLANK support was initialized.

Signed-off-by: Thierry Reding <treding@nvidia.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

drm/plane: Remove redundant extern

Use of the extern keyword for function prototypes is unnecessary, so it
can be removed.

Signed-off-by: Thierry Reding <treding@nvidia.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

drm/plane: Use consistent data types for format count

Rather than a mix of the the sized uint32_t and signed integer, use an
unsized unsigned int to specify the format count.

Signed-off-by: Thierry Reding <treding@nvidia.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

vga_switcheroo: Remove unnecessary checks

debugfs_remove() gracefully ignores NULL parameters, so the explicit
checks can be removed.

Signed-off-by: Thierry Reding <treding@nvidia.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

vga_switcheroo: Wrap overly long lines

Wrap overly long lines to make checkpatch happy. While at it, also add
blank lines after declarations to eliminate additional problems flagged
by checkpatch.

Signed-off-by: Thierry Reding <treding@nvidia.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

vga_switcheroo: Use pr_fmt()

Use pr_fmt() to define the "vga_switcheroo: " prefix that is prepended
to all output messages emitted by pr_*() functions. This allows making
existing strings much shorter and eliminates a bunch of warnings from
checkpatch about lines being overly long.

Signed-off-by: Thierry Reding <treding@nvidia.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

vga_switcheroo: Cleanup header comment

The header comment uses a weird combination of formatting styles. Make
it consistent.

Signed-off-by: Thierry Reding <treding@nvidia.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

vga_switcheroo: Use pr_*() instead of printk()

This silences a bunch of checkpatch warnings and makes the code shorter.

Signed-off-by: Thierry Reding <treding@nvidia.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

vgaarb: Fix a few checkpatch errors and warnings

Wrap overly long lines (offending lines were mostly comments, so trivial
to fix up) and a number of other coding style issues pointed out by the
checkpatch tool.

Signed-off-by: Thierry Reding <treding@nvidia.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

vgaarb: Use vgaarb: prefix consistently in messages

Define the pr_fmt() macro to causes all messages emitted by pr_*()
functions to be prefixed with "vgaarb: ".

Signed-off-by: Thierry Reding <treding@nvidia.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

vgaarb: Stop complaining about absent devices

Some setups do not register a default VGA device, in which case the VGA
arbiter will still complain about the (non-existent) PCI device being a
non-VGA device.

Fix this by making the error message conditional on a default VGA device
having been set up. Note that the easy route of erroring out early isn't
going to work because otherwise priv->target won't be properly updated.

Signed-off-by: Thierry Reding <treding@nvidia.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

drm/atomic: fix null pointer access to mode_fixup callback

This patch fixes null pointer access incurred when
encoder driver didn't set its own mode_fixup callback.

mode_fixup callback shoudn't be called if the callback
of drm_encoder_helper_funcs is NULL.

Changelog v2:
- change it to else if

Signed-off-by: Inki Dae <inki.dae@samsung.com>
Reviewed-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

drm/i915: Use CONFIG_DRM_FBDEV_EMULATION

Instead of our own duplicated one. This fixes a bug in the driver
unload code if DRM_FBDEV_EMULATION=n but DRM_I915_FBDEV=y because we
try to unregister the nonexistent fbdev drm_framebuffer.

Cc: Archit Taneja <architt@codeaurora.org>
Cc: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Reported-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@intel.com>
Reviewed-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

drm/core: Set mode to NULL when connectors in a set drops to 0.

Without this when a MST connector is removed drm_atomic_helper_set_config
can complain about set->mode && !set->num_connectors.

------------[ cut here ]------------
WARNING: CPU: 2 PID: 2403 at drivers/gpu/drm/drm_atomic_helper.c:1673 drm_atomic_helper_set_config+0x22e/0x420()
CPU: 2 PID: 2403 Comm: kms_flip Not tainted 4.2.0-rc5 #4233
Hardware name: NUC5i7RYB, BIOS RYBDWi35.86A.0246.2015.0309.1355 03/09/2015
 ffffffff81ac75e8 ffff88004e4ffbf8 ffffffff81714c34 0000000080000000
 0000000000000000 ffff88004e4ffc38 ffffffff8107bf81 ffff88004e4ffc48
 ffff8800d8ca0690 ffff8800d8d7a080 ffff8800d8cc2290 ffff8800d07bc9f0
Call Trace:
 [<ffffffff81714c34>] dump_stack+0x4f/0x7b
 [<ffffffff8107bf81>] warn_slowpath_common+0x81/0xc0
 [<ffffffff8107c065>] warn_slowpath_null+0x15/0x20
 [<ffffffff813d9e3e>] drm_atomic_helper_set_config+0x22e/0x420
 [<ffffffff813da174>] ? drm_atomic_helper_plane_set_property+0x84/0xc0
 [<ffffffff813ee101>] drm_mode_set_config_internal+0x61/0x100
 [<ffffffff813dc4ed>] restore_fbdev_mode+0xbd/0xe0
 [<ffffffff813de1e4>] drm_fb_helper_restore_fbdev_mode_unlocked+0x24/0x70
 [<ffffffffc0123d11>] intel_fbdev_restore_mode+0x21/0x80 [i915]
 [<ffffffffc014bf69>] i915_driver_lastclose+0x9/0x10 [i915]
 [<ffffffff813e2429>] drm_lastclose+0x29/0x130
 [<ffffffff813e2844>] drm_release+0x314/0x500
 [<ffffffff81194795>] __fput+0xe5/0x1f0
 [<ffffffff811948d9>] ____fput+0x9/0x10
 [<ffffffff810968d8>] task_work_run+0x88/0xb0
 [<ffffffff8107d53f>] do_exit+0x37f/0xa90
 [<ffffffff8127e258>] ? selinux_file_ioctl+0x48/0xc0
 [<ffffffff81277dfe>] ? security_file_ioctl+0x3e/0x60
 [<ffffffff8107ec80>] do_group_exit+0x40/0xa0
 [<ffffffff8107ecef>] SyS_exit_group+0xf/0x10
 [<ffffffff8171bdd7>] entry_SYSCALL_64_fastpath+0x12/0x6a
---[ end trace 0daf358c49351567 ]---

Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

drm/atomic: Call ww_acquire_done after check phase is complete

We want to make sure that no one tries to acquire more locks and
states, and ww mutexes provide debug facilities for that. So use them.

v2: Only call acquire_done when ->atomic_check was successful to avoid
falling over an -EDEADLK (spotted by Maarten).

Cc: Rob Clark <robdclark@gmail.com>
Cc: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Reviewed-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@intel.com>

drm/atomic: Paper over locking WARN in default_state_clear

In

commit 6f75cea66c8dd043ced282016b21a639af176642
Author: Daniel Vetter <daniel.vetter@ffwll.ch>
Date:   Wed Nov 19 18:38:07 2014 +0100

    drm/atomic: Only destroy connector states with connection mutex held

I tried to fix races of atomic commits against connector
hot-unplugging. The idea is to ensure lifetimes by holding the
connection_mutex long enough. That works for synchronous commits, but
not for async ones.

For async atomic commit we really need to fix up connector lifetimes
for real. But that's a much bigger task, so just add more duct-tape:
For cleaning up connector states we currently don't need the connector
itself. So NULL it out and remove the locking check. Of course that
check was to protect the entire sequence, but the modeset itself
should be save since currently DP MST hot-removal does a dpms-off. And
that should synchronize with any outstanding async atomic commit.

Or at least that's my hope, this is all a giant mess.

Reported-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Cc: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Reviewed-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@intel.com>

drm/edid: Use ARRAY_SIZE in drm_add_modes_noedid

Spotted while reading code for random reasons.

Reviewed-by: Thierry Reding <treding@nvidia.com>
Signed-off-by: Daniel Vetter <daniel.vetter@intel.com>

drm/qxl: Don't take dev->struct_mutex in bo_force_delete

It really doesn't protect anything which doesn't have other locks
already. It also doesn't seem to be wired up into the driver unload
code fwiw, but that's a different issue.

Reviewed-by: Thierry Reding <treding@nvidia.com>
Signed-off-by: Daniel Vetter <daniel.vetter@intel.com>

drm/nouveau: Don't take dev->struct_mutex in ttm_fini

This is only called in driver load/unload paths, no need to grab any
locks at all. Also, ttm takes care of itself anyway.

Cc: Ben Skeggs <bskeggs@redhat.com>
Reviewed-by: Thierry Reding <treding@nvidia.com>
Signed-off-by: Daniel Vetter <daniel.vetter@intel.com>

drm/rockchip: Don't grab dev->struct_mutex for in mmap offset ioctl

Since David Herrmann's mmap vma manager rework we don't need to grab
dev->struct_mutex any more to prevent races when looking up the mmap
offset. Drop it and instead don't forget to use the unref_unlocked
variant (since the drm core still cares).

Aside: I stumbled over the mmap handler which directly does a
dma_mmap_attrs. But totally fails to grab a reference on the
underlying object and hence looks like it happily just leaks the ptes
since there's no guarantee the mmap isn't still around when
gem_free_object is called. Which the kerneldoc of dma_mmap_attrs
explicitly forbids.

v2: Fixup compile fail 0-day spotted.

Cc: Mark Yao <mark.yao@rock-chips.com>
Reviewed-by: Thierry Reding <treding@nvidia.com>
Signed-off-by: Daniel Vetter <daniel.vetter@intel.com>

drm/cma-helper: Don't grab dev->struct_mutex for in mmap offset ioctl

Since David Herrmann's mmap vma manager rework we don't need to grab
dev->struct_mutex any more to prevent races when looking up the mmap
offset. Drop it and instead don't forget to use the unref_unlocked
variant (since the drm core still cares).

Cc: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Cc: Rob Clark <robdclark@gmail.com>
Reviewed-by: Thierry Reding <treding@nvidia.com>
Signed-off-by: Daniel Vetter <daniel.vetter@intel.com>

drm/cirrus: Don't grab dev->struct_mutex for in mmap offset ioctl

Since David Herrmann's mmap vma manager rework we don't need to grab
dev->struct_mutex any more to prevent races when looking up the mmap
offset. Drop it and instead don't forget to use the unref_unlocked
variant (since the drm core still cares).

Reviewed-by: Thierry Reding <treding@nvidia.com>
Signed-off-by: Daniel Vetter <daniel.vetter@intel.com>

drm/mga200g: Hold a proper reference for cursor_set

Looking up an obj, immediate dropping the acquired reference and then
continuing to use it isn't how this is supposed to work. Fix this by
holding a reference for the entire function.

While at it stop grabbing dev->struct_mutex, it doesn't protect
anything here.

Reviewed-by: Thierry Reding <treding@nvidia.com>
Signed-off-by: Daniel Vetter <daniel.vetter@intel.com>

drm/mga200g: Don't grab dev->struct_mutex for in mmap offset ioctl

Since David Herrmann's mmap vma manager rework we don't need to grab
dev->struct_mutex any more to prevent races when looking up the mmap
offset. Drop it and instead don't forget to use the unref_unlocked
variant (since the drm core still cares).

Reviewed-by: Thierry Reding <treding@nvidia.com>
Signed-off-by: Daniel Vetter <daniel.vetter@intel.com>

drm/bochs: Don't grab dev->struct_mutex for in mmap offset ioctl

Since David Herrmann's mmap vma manager rework we don't need to grab
dev->struct_mutex any more to prevent races when looking up the mmap
offset. Drop it and instead don't forget to use the unref_unlocked
variant (since the drm core still cares).

Reviewed-by: Thierry Reding <treding@nvidia.com>
Signed-off-by: Daniel Vetter <daniel.vetter@intel.com>

drm/ast: Don't grab dev->struct_mutex for in mmap offset ioctl

Since David Herrmann's mmap vma manager rework we don't need to grab
dev->struct_mutex any more to prevent races when looking up the mmap
offset. Drop it and instead don't forget to use the unref_unlocked
variant (since the drm core still cares).

Reviewed-by: Thierry Reding <treding@nvidia.com>
Signed-off-by: Daniel Vetter <daniel.vetter@intel.com>

drm/gem: Be more friendly with locking checks

BUG_ON kills the driver, WARN_ON is much friendlier. And usually nothing
bad happens when the locking is slightly busted.

v2: Fix typos in commit message Thierry spotted.

Reviewed-by: Thierry Reding <treding@nvidia.com>
Signed-off-by: Daniel Vetter <daniel.vetter@intel.com>

drm: Remove two-level menu in Kconfig

The Direct Rendering Manager Kconfig option is already a separate menu,
so remove the extra level to make it easier to navigate.

Signed-off-by: Thierry Reding <treding@nvidia.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

drm: Add top level Kconfig option for DRM fbdev emulation

Legacy fbdev emulation support via DRM is achieved through KMS FB helpers.
Most modesetting drivers enable provide fbdev emulation by default by
selecting KMS FB helpers. A few provide a separate Kconfig option for the
user to enable or disbale fbdev emulation.

Enabling fbdev emulation is finally a distro-level decision. Having a top
level Kconfig option for fbdev emulation helps by providing a uniform way
to enable/disable fbdev emulation for any modesetting driver. It also lets
us remove unnecessary driver specific Kconfig options that causes bloat.

With a top level Kconfig in place, we can stub out the fb helper functions
when not needed without breaking functionality. Having stub functions also
prevents drivers to require wrapping fb helper function calls with #ifdefs.

DRM_FBDEV_EMULATION defaults to y since many drivers enable fbdev
emulation by default and majority of distributions expect the fbdev
interface in the kernel.

Signed-off-by: Archit Taneja <architt@codeaurora.org>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

drm/fb-helper: Move drm_fb_helper_force_kernel_mode() inside #ifdef

If CONFIG_MAGIC_SYSRQ is not set:

    drivers/gpu/drm/drm_fb_helper.c:390:13: warning: 'drm_fb_helper_force_kernel_mode' defined but not used [-Wunused-function]
     static bool drm_fb_helper_force_kernel_mode(void)
 ^

Move drm_fb_helper_force_kernel_mode() inside the existing #ifdef to fix
this.

Signed-off-by: Geert Uytterhoeven <geert+renesas@glider.be>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

drm/fb-helper: Clarify drm_fb_helper_restore_fbdev_mode*()

As of commit 5ea1f752ae04be40 ("drm: add
drm_fb_helper_restore_fbdev_mode_unlocked()"),
drm_fb_helper_restore_fbdev_mode() is no longer public, and drivers
should call drm_fb_helper_restore_fbdev_mode_unlocked() from their
->lastclose callbacks instead.

Update the documentation to reflect this, and absorb the one liner
drm_fb_helper_restore_fbdev_mode() into its single caller.

Signed-off-by: Geert Uytterhoeven <geert+renesas@glider.be>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

drivers: gpu: Drop unlikely before IS_ERR(_OR_NULL)

IS_ERR(_OR_NULL) already contain an 'unlikely' compiler flag and there
is no need to do that again from its callers. Drop it.

Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Reviewed-by: Sinclair Yeh <syeh@vmware.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

drm/virtio: Use new drm_fb_helper functions

Use the newly created wrapper drm_fb_helper functions instead of calling
core fbdev functions directly. They also simplify the fb_info creation.

v3:
- Don't touch remove_conflicting_framebuffers

v2:
- add missing header for virtgpu_fb.c

Cc: David Airlie <airlied@linux.ie>
Cc: Gerd Hoffmann <kraxel@redhat.com>
Signed-off-by: Archit Taneja <architt@codeaurora.org>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

drm/amdgpu: Use new drm_fb_helper functions

Use the newly created wrapper drm_fb_helper functions instead of calling
core fbdev functions directly. They also simplify the fb_info creation.

v3:
- Don't touch remove_conflicting_framebuffers

v2:
- Fixed PTR_ERR issue mentioned by kbuild bot

Cc: Fengguang Wu <fengguang.wu@intel.com>
Cc: Alex Deucher <alexander.deucher@amd.com>
Cc: Oded Gabbay <oded.gabbay@gmail.com>
Cc: "Christian König" <christian.koenig@amd.com>
Signed-off-by: Archit Taneja <architt@codeaurora.org>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

drm/bochs: Use new drm_fb_helper functions

Use the newly created wrapper drm_fb_helper functions instead of calling
core fbdev functions directly. They also simplify the fb_info creation.

v3:
- Don't touch remove_conflicting_framebuffers

v2:
- remove unused variable device in bochsfb_create

Cc: David Airlie <airlied@linux.ie>
Cc: Gerd Hoffmann <kraxel@redhat.com>
Cc: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Archit Taneja <architt@codeaurora.org>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

drm/nouveau: Use new drm_fb_helper functions

Use the newly created wrapper drm_fb_helper functions instead of calling
core fbdev functions directly. They also simplify the fb_info creation.

v3:
- Don't touch remove_conflicting_framebuffers

v2:
- remove unused variable pdev in nouveau_fbcon_create

Cc: David Airlie <airlied@linux.ie>
Cc: Ben Skeggs <bskeggs@redhat.com>
Cc: Alexandre Courbot <acourbot@nvidia.com>
Signed-off-by: Archit Taneja <architt@codeaurora.org>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

drm/i915: Use new drm_fb_helper functions

Use the newly created wrapper drm_fb_helper functions instead of calling
core fbdev functions directly. They also simplify the fb_info creation.

v3:
- Don't touch remove_conflicting_framebuffers

v2:
- No changes

Cc: Daniel Vetter <daniel.vetter@intel.com>
Cc: Jani Nikula <jani.nikula@linux.intel.com>
Signed-off-by: Archit Taneja <architt@codeaurora.org>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

drm/radeon: Use new drm_fb_helper functions

Use the newly created wrapper drm_fb_helper functions instead of calling
core fbdev functions directly. They also simplify the fb_info creation.

v3:
- Don't touch remove_conflicting_framebuffers

v2:
- Fix build break because of missing include of drm_fb_helper in
  radeon_drv.c

Cc: Alex Deucher <alexander.deucher@amd.com>
Cc: "Christian König" <christian.koenig@amd.com>
Signed-off-by: Archit Taneja <architt@codeaurora.org>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

drm/mgag200: Use new drm_fb_helper functions

Use the newly created wrapper drm_fb_helper functions instead of calling
core fbdev functions directly. They also simplify the fb_info creation.

v3:
- Don't touch remove_conflicting_framebuffers

v2:
- remove unused goto label 'out'

Cc: Daniel Vetter <daniel.vetter@ffwll.ch>
Cc: David Airlie <airlied@linux.ie>
Cc: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Archit Taneja <architt@codeaurora.org>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

drm/omap: Use new drm_fb_helper functions

Use the newly created wrapper drm_fb_helper functions instead of calling
core fbdev functions directly. They also simplify the fb_info creation.

v3:
- Update error handling for new drm_fb_helper funcs. Check using IS_ERR()
  instead of checking for NULL.

Reported-by: Dan Carpenter <dan.carpenter@oracle.com>
v2:
- No changes

Cc: Tomi Valkeinen <tomi.valkeinen@ti.com>
Cc: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Signed-off-by: Archit Taneja <architt@codeaurora.org>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

drm/cirrus: Use new drm_fb_helper functions

Use the newly created wrapper drm_fb_helper functions instead of calling
core fbdev functions directly.

v3:
- Don't touch remove_conflicting_framebuffers

v2:
- Remove stray goto label out_iounmap

Cc: Thierry Reding <treding@nvidia.com>
Cc: Zach Reizner <zachr@google.com>
Cc: Russell King <rmk+kernel@arm.linux.org.uk>
Cc: Fabian Frederick <fabf@skynet.be>
Signed-off-by: Archit Taneja <architt@codeaurora.org>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

drm: Fixup locking WARNINGs in drm_mode_config_reset

With

commit 7a3f3d6667f5f9ffd1517f6b21d64bbf5312042c
Author: Daniel Vetter <daniel.vetter@ffwll.ch>
Date:   Thu Jul 9 23:44:28 2015 +0200

    drm: Check locking in drm_for_each_connector

we started checking the locking in drm_for_each_connector but somehow
I totally missed drm_mode_config_reset. There's no problem there since
this function should only be called in single-threaded contexts
(driver load or resume), so just wrap the loop with the right lock.

v2: Drink coffee and all that ...

Cc: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Reported-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Tested-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Signed-off-by: Daniel Vetter <daniel.vetter@intel.com>

drm: Remove __drm_modeset_lock_all

The last user is gone, no need for trylocking any more in this legacy
helper.

Reviewed-by: Rob Clark <robdclark@gmail.com>
Signed-off-by: Daniel Vetter <daniel.vetter@intel.com>

drm/fb-helper: Stop using trylocks in force_restore

Since the panic handling is gone this is only used for force-restoring
the fbdev/fbcon from sysrq, and that's done with a work item. No need
any more to do trylocks, we can just do normal locking.

Reviewed-by: Rob Clark <robdclark@gmail.com>
Signed-off-by: Daniel Vetter <daniel.vetter@intel.com>

drm/fbdev: Return -EBUSY when oopsing

Trying to do anything with kms drivers when oopsing has become a
failing proposition. But since we can end up in the fbdev code simply
due to the console unblanking that's done unconditionally just
removing our panic handler isn't enough. We need to block all fbdev
callbacks when oopsing.

There was already one in the blank handler, but it failed silently.
That makes it impossible for drivers (like i915) who subclass these
functions to figure this out.

Instead consistently return -EBUSY so that everyone knows that we
really don't want to be bothered right now. This also allows us to
remove a pile of FIXMEs from the i915 fbdev code (since due to the
failure code they now won't attempt to grab dangerous locks any more).

Cc: Dave Airlie <airlied@gmail.com>
Cc: Rodrigo Vivi <rodrigo.vivi@gmail.com>
Reviewed-by: Rob Clark <robdclark@gmail.com>
Signed-off-by: Daniel Vetter <daniel.vetter@intel.com>