radeon must load only after amdkfd's loading has been completed. If that
is not enforced, then radeon's call into amdkfd's functions will cause a
kernel BUG.

When radeon and amdkfd are built as kernel modules, that rule is
enforced by the kernel's modules loading mechanism. When radeon and
amdkfd are built inside the kernel image, that rule is enforced by
ordering in the drm Makefile (amdkfd before radeon).

Instead of using drm Makefile ordering, we can now use deferred
loading as amdkfd now returns -EPROBE_DEFER in kgd2kfd_init() when it is
not yet loaded.

This patch defers radeon loading by propagating -EPROBE_DEFER to the
kernel's drivers loading infrastructure. That will put radeon into the
pending drivers list (see description in dd.c). Once amdkfd is loaded,
a call to kgd2kfd_init() will return successfully and radeon will be
able to load.
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>

Current dependencies between amdkfd and radeon/amdgpu force the loading
of amdkfd _before_ radeon and/or amdgpu are loaded. When all these kernel
drivers are built as modules, this ordering is enforced by the kernel
built-in mechanism of loading dependent modules.

However, there is no such mechanism in case where all these drivers are
compiled inside the kernel image (not as modules). The current way to
enforce loading of amdkfd before radeon/amdgpu, is to put amdkfd before
radeon/amdgpu in the drm Makefile, but that method is way too fragile.

In addition, there is no kernel mechanism to check whether a kernel
driver that is built inside the kernel image, has already been loaded.

To solve this, this patch adds to kfd_module.c a new static variable,
amdkfd_init_completed, that is set to 1 only when amdkfd's
module initialization function has been completed (successfully).

kgd2kfd_init(), which is the initialization function of the
kgd-->kfd interface, and which is the first function in amdkfd called by
radeon/amdgpu, will return successfully only if amdkfd_init_completed is
equal 1.

If amdkfd_init_completed is not equal to 1, kgd2kfd_init() will
return -EPROBE_DEFER to signal radeon/amdgpu they need to defer
their loading until amdkfd is loaded.
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>

As previously discussed, this is my first pull request for the DCU DRM
driver along with the change in MAINTAINERS.
https://lkml.org/lkml/2016/1/7/26

The pull contains some code cleanup changes (e.g. removing all error
handling for the regmap calls) and several fixes.

* 'for-next' of http://git.agner.ch/git/linux-drm-fsl-dcu:
  drm/fsl-dcu: fix register initialization
  drm/fsl-dcu: use mode flags for hsync/vsync polarity
  drm/fsl-dcu: fix alpha blending
  drm/fsl-dcu: mask all interrupts on initialization
  drm/fsl-dcu: handle initialization errors properly
  drm/fsl-dcu: avoid memory leak on errors
  drm/fsl-dcu: remove regmap return value checks
  drm/fsl-dcu: specify volatile registers
  drm: fsl-dcu: Fix no fb check bug
  MAINTAINERS: update for Freescale DCU DRM driver

The layer enumeration start with 0 (0-15 for LS1021a and 0-63 for
Vybrid) whereas the register enumeration start from 1 (1-10 for
LS1021a and 1-9 for Vybrid). The loop started off from 0 for both
iterations and initialized the number of layers inclusive, which
is one layer too many.

All extensively written registers seem to be unassigned, it seems
that the write to those registers did not do any harm in practice.
Signed-off-by: Stefan Agner <stefan@agner.ch>

The current default configuration is as follows:
- Invert VSYNC signal (active LOW)
- Invert HSYNC signal (active LOW)

The mode flags allow to specify the required polarity per
mode. Furthermore, none of the current driver settings is
actually a standard polarity.

This patch applies the current driver default polarities as
explicit flags to the display which has been introduced with
the driver (NEC WQVGA "nec,nl4827hc19-05b"). The driver now
also parses the flags field and applies the configuration
accordingly, by using the following values as standard
polarities: (e.g. when no flags are specified):
- VSYNC signal not inverted (active HIGH)
- HSYNC signal not inverted (active HIGH)
Acked-by: Thierry Reding <treding@nvidia.com>
Signed-off-by: Stefan Agner <stefan@agner.ch>

Fix alpha blending by enabling alpha blending for the whole frame if
a color mode with alpha channel is selected (DRM_FORMAT_ARGB*). Also
support color modes without alpha channel (DRM_FORMAT_XRGB*) by just
not enabling alpha blending on layer level.
Signed-off-by: Stefan Agner <stefan@agner.ch>

The state of the interrupt mask register on initialization is
unknown, e.g. U-Boot could already used the DCU. So depending on
the boot loader, the outcome of the interrupt mask register could
be different. A defined state is much more preferable. Also, there
is no value in keeping interrupts enabled which we don't need.
Therefor, mask all interrupts on initialization.
Signed-off-by: Stefan Agner <stefan@agner.ch>

If initialization fails (e.g. due to missing panel node or deferred
probe) make sure to roll-back all operations and return the error
code.
Signed-off-by: Stefan Agner <stefan@agner.ch>

Improve error handling during CRTC initialization. Especially avoid
memory leaks in the primary plane initialization error path.
Signed-off-by: Stefan Agner <stefan@agner.ch>

It is not common to do regmap return value checks, especially not
for memory mapped device. We can rule out most error returns since
the conditions are static and we know they are ok (e.g. offset
aligned to register stride). Also without proper error handling
they are not really valuable for the user. Hence remove most of
them.

The check in the interrupt handler is worth keeping since a
volatile register won't be readable in case register caching is
still enabled.
Signed-off-by: Stefan Agner <stefan@agner.ch>

Since we are using cached registers, we need to specify volatile
registers explicitly to avoid reading their value from the cache.
This allows to read the correct interrupt status in fsl_dcu_drm_irq
and clear the asserted bits only.
Signed-off-by: Stefan Agner <stefan@agner.ch>

For state->fb or state->crtc may be NULL in fsl_dcu_drm_plane_atomic_check
function, if so, return 0.
Signed-off-by: Meng Yi <meng.yi@nxp.com>
Signed-off-by: Jianwei Wang <jianwei.wang.chn@gmail.com>
Signed-off-by: Stefan Agner <stefan@agner.ch>

Promote myself as new maintainer of the Freescale DCU DRM driver.
Acked-by: Jianwei Wang <jianwei.wang.chn@gmail.com>
Signed-off-by: Stefan Agner <stefan@agner.ch>

rcar-du updates.

* 'drm/next/du' of git://linuxtv.org/pinchartl/fbdev: (281 commits)
  drm: rcar-du: Add tri-planar memory formats support
  drm: rcar-du: Add probe deferral debug messages
  drm: rcar-du: lvds: Add R-Car Gen3 support
  drm: rcar-du: lvds: Rename PLLEN bit to PLLON
  drm: rcar-du: lvds: Fix PLL frequency-related configuration
  drm: rcar-du: lvds: Avoid duplication of clock clamp code
  drm: rcar-du: Add R8A7795 device support
  drm: rcar-du: Output the DISP signal on the ODDF pin
  drm: rcar-du: Output the DISP signal on the DISP pin
  drm: rcar-du: Support up to 4 CRTCs
  drm: rcar-du: Drop LVDS double dependency on OF
  drm: rcar-du: Enable compilation on ARM64
  drm: rcar-du: Fix compile warning on 64-bit platforms
  drm: rcar-du: Expose the VSP1 compositor through KMS planes
  drm: rcar-du: Move plane allocator to rcar_du_plane.c
  drm: rcar-du: Restart the DU group when a plane source changes
  drm: rcar-du: Add VSP1 compositor support
  drm: rcar-du: Add VSP1 support to the planes allocator
  drm: rcar-du: Refactor plane setup
  drm: rcar-du: Compute plane DDCR4 register value directly
  ...

Those formats are supported on Gen3 only.
Signed-off-by: Laurent Pinchart <laurent.pinchart+renesas@ideasonboard.com>

Print a message when the HDMI I2C slave encoder can't be found to help
debugging probe deferral issues.
Signed-off-by: Laurent Pinchart <laurent.pinchart+renesas@ideasonboard.com>

The LVDS encoder differs slightly in Gen3 SoCs in its PLL configuration.
Add support for the Gen3 LVDS PLL parameters and startup procedure.
Signed-off-by: Koji Matsuoka <koji.matsuoka.xm@renesas.com>
Signed-off-by: Laurent Pinchart <laurent.pinchart+renesas@ideasonboard.com>

The bit is named PLLON in the datasheet, rename it.
Signed-off-by: Laurent Pinchart <laurent.pinchart+renesas@ideasonboard.com>

The frequency checks don't match the datasheet, fix them.
Signed-off-by: Laurent Pinchart <laurent.pinchart+renesas@ideasonboard.com>

Replace the duplicate code by a single central function.
Signed-off-by: Laurent Pinchart <laurent.pinchart+renesas@ideasonboard.com>

Document the R8A7795-specific DT bindings and support them in the
driver. The HDMI and LVDS outputs are currently not supported.
Signed-off-by: Koji Matsuoka <koji.matsuoka.xm@renesas.com>
Signed-off-by: Laurent Pinchart <laurent.pinchart+renesas@ideasonboard.com>

The ODDF signal, output by default on the ODDF pin, isn't used on any
board supported in the kernel. As the Gen3 Salvator-X board uses the
ODDF pin as a DISP signal, hardcode that configuration in the driver.

Use of the ODDF signal will be implemented later through proper DT-based
configuration of the DU pins.
Signed-off-by: Laurent Pinchart <laurent.pinchart+renesas@ideasonboard.com>

The DE signal is currently configured to be identical to the DISP
signal and is used for the same purpose. To make it clearer that the
DISP pin outputs the DISP signal, select it explicitly.
Signed-off-by: Laurent Pinchart <laurent.pinchart+renesas@ideasonboard.com>

The Gen3 R8A7795 DU has 4 CRTCs, support them all.
Signed-off-by: Koji Matsuoka <koji.matsuoka.xm@renesas.com>
Signed-off-by: Laurent Pinchart <laurent.pinchart+renesas@ideasonboard.com>

LVDS support depends on DRM_RCAR_DU which already depends on OF. Drop
the explicit dependency.
Signed-off-by: Laurent Pinchart <laurent.pinchart+renesas@ideasonboard.com>

The R8A7795 SoC is ARM64-based and include a DU. Enable driver
compilation on ARM64.
Signed-off-by: Koji Matsuoka <koji.matsuoka.xm@renesas.com>
Signed-off-by: Laurent Pinchart <laurent.pinchart+renesas@ideasonboard.com>

Use %tu instead of %u to print difference between pointers.
Signed-off-by: Koji Matsuoka <koji.matsuoka.xm@renesas.com>
Signed-off-by: Laurent Pinchart <laurent.pinchart+renesas@ideasonboard.com>

On R-Car Gen3 SoCs the DU lost its ability to access memory directly and
needs to work in conjunction with the VSP to do so. This commit handles
the VSP internally to hide it from the user.
Signed-off-by: Laurent Pinchart <laurent.pinchart+renesas@ideasonboard.com>

The plane allocator is specific to DU planes and won't be used for
VSP-based planes, move it with the rest of the DU planes code where it
belongs.
Signed-off-by: Laurent Pinchart <laurent.pinchart+renesas@ideasonboard.com>

Plane sources are configured by the VSPS bit in the PnDDCR4 register.
Although the datasheet states that the bit is updated during vertical
blanking, it seems that updates only occur when the DU group is held in
reset through the DSYSR.DRES bit. Restart the group if the source
changes.
Signed-off-by: Laurent Pinchart <laurent.pinchart+renesas@ideasonboard.com>

Configure the plane source at plane setup time to source frames from
memory or from the VSP1.
Signed-off-by: Laurent Pinchart <laurent.pinchart+renesas@ideasonboard.com>

The R8A7790 DU can source frames directly from the VSP1 devices VSPD0
and VSPD1. VSPD0 feeds DU0/1 plane 0, and VSPD1 feeds either DU2 plane 0
or DU0/1 plane 1.

Allocate the correct fixed plane when sourcing frames from VSPD0 or
VSPD1, and allocate planes in reverse index order otherwise to ensure
maximum availability of planes 0 and 1.
Signed-off-by: Laurent Pinchart <laurent.pinchart+renesas@ideasonboard.com>

Rename plane setup functions using more explicit names, and use local
variables to store intermediate values to increase readability. This
also prepares for VSP compositor support.
Signed-off-by: Laurent Pinchart <laurent.pinchart+renesas@ideasonboard.com>

There's no need for a read-modify-write pattern, all register bits can
be set explicitly.
Signed-off-by: Laurent Pinchart <laurent.pinchart+renesas@ideasonboard.com>

A disabled CRTC can't display planes, the driver shouldn't try to
configure it when updating planes. The DRM core will store the plane
state for us, and the plane will be configured appropriately the next
time the CRTC is enabled.
Signed-off-by: Laurent Pinchart <laurent.pinchart+renesas@ideasonboard.com>

The atomic framework guarantees that the CRTC enable and disable
functions will only be called when needed, there's no need to duplicate
the CRTC state check. By replacing the local CRTC enabled state check at
resume time with the CRTC active state we can remove the local CRTC
enabled state altogether.
Signed-off-by: Laurent Pinchart <laurent.pinchart+renesas@ideasonboard.com>

The drm driver .load() operation is prone to race conditions as it
initializes the driver after registering the device nodes. Its usage is
deprecated, inline it in the probe function and call drm_dev_alloc() and
drm_dev_register() explicitly.

For consistency inline the .unload() handler in the remove function as
well.
Signed-off-by: Laurent Pinchart <laurent.pinchart+renesas@ideasonboard.com>
Acked-by: Daniel Vetter <daniel.vetter@ffwll.ch>

drivers/media/platform/vsp1/vsp1_drm.c:47:5: warning: no previous prototype for 'vsp1_du_init' [-Wmissing-prototypes]
 int vsp1_du_init(struct device *dev)
     ^
drivers/media/platform/vsp1/vsp1_drm.c:76:5: warning: no previous prototype for 'vsp1_du_setup_lif' [-Wmissing-prototypes]
 int vsp1_du_setup_lif(struct device *dev, unsigned int width,
     ^
drivers/media/platform/vsp1/vsp1_drm.c:221:6: warning: no previous prototype for 'vsp1_du_atomic_begin' [-Wmissing-prototypes]
 void vsp1_du_atomic_begin(struct device *dev)
      ^
drivers/media/platform/vsp1/vsp1_drm.c:273:5: warning: no previous prototype for 'vsp1_du_atomic_update' [-Wmissing-prototypes]
 int vsp1_du_atomic_update(struct device *dev, unsigned int rpf_index,
     ^
drivers/media/platform/vsp1/vsp1_drm.c:451:6: warning: no previous prototype for 'vsp1_du_atomic_flush' [-Wmissing-prototypes]
 void vsp1_du_atomic_flush(struct device *dev)
      ^
Signed-off-by: Mauro Carvalho Chehab <mchehab@osg.samsung.com>

The IP version number carries enough information to identify the exact
device instance features. Drop the related DT properties and use the IP
version instead.
Signed-off-by: Laurent Pinchart <laurent.pinchart+renesas@ideasonboard.com>
Signed-off-by: Mauro Carvalho Chehab <mchehab@osg.samsung.com>