The following ioctl sequence causes fimc_dma_run() to start processing without
complete scaler and DMA initialization which causes missing interrupt and
blocking on DQBUF:
S_FMT, STREAMON, QBUF, DQBUF, STREAMOFF, STREAMON, QBUF, DQBUF.

Fix this regression caused by moving pm_runtime* calls to start/stop_streaming
callback by making sure the fimc_m2m_resume() is always invoked when expected.
Reported-by: NTomasz Stanislawski <t.stanislaws@samsung.com>
Signed-off-by: NSylwester Nawrocki <s.nawrocki@samsung.com>
Signed-off-by: NKyungmin Park <kyungmin.park@samsung.com>
Signed-off-by: NMauro Carvalho Chehab <mchehab@redhat.com>

On Exynos SoCs the FIMC IP allows to configure globally the alpha
component of all pixels for V4L2_PIX_FMT_RGB32, V4L2_PIX_FMT_RGB555
and V4L2_PIX_FMT_RGB444 image formats. This patch adds a v4l2 control
in order to let the applications control the alpha component value.

The alpha value range depends on the pixel format, for RGB32 it's
0..255 (8-bits), for RGB555 - 0..1 (1-bit) and for RGB444 - 0..15
(4-bits). The v4l2 control range is always 0..255 and the alpha
component data width is determined by currently set format on the
V4L2_BUF_TYPE_VIDEO_CAPTURE_MPLANE buffer queue. The applications
need to match the alpha channel value range and the pixel format
since the driver will clamp the alpha component. Depending on
fourcc the valid alpha bits are:

 - V4L2_PIX_FMT_RGB555  [0]
 - V4L2_PIX_FMT_RGB444  [3:0]
 - V4L2_PIX_FMT_RGB32   [7:0]

When switching to a pixel format with smaller alpha component width
the currently set alpha value will be clamped to maximum value valid
for current format. When switching to a format with wider alpha the
alpha value remains unchanged.

The variant description data structure is extended with a new entry
so an additional control is created only where really supported by
the hardware.

V4L2_PIX_FMT_RGB555 and V4L2_PIX_FMT_RGB444 formats are only valid
for V4L2_BUF_TYPE_VIDEO_CAPTURE buffer queue.
Signed-off-by: NSylwester Nawrocki <s.nawrocki@samsung.com>
Signed-off-by: NKyungmin Park <kyungmin.park@samsung.com>
Signed-off-by: NMauro Carvalho Chehab <mchehab@redhat.com>

With 16-bit RGB565 colour format pixels are stored by the device in memory
in the following order:

    | b3  | b2  | b1  | b0  |
   ~+-----+-----+-----+-----+
    | R5 G6 B5  | R5 G6 B5  |

This corresponds to V4L2_PIX_FMT_RGB565 fourcc, not V4L2_PIX_FMT_RGB565X.
This change is required to avoid trouble when setting up video pipeline
with the s5p-tv devices, so the colour formats at both devices can be
properly matched.

Cc: <stable@kernel.org>
Signed-off-by: NSylwester Nawrocki <s.nawrocki@samsung.com>
Signed-off-by: NKyungmin Park <kyungmin.park@samsung.com>
Signed-off-by: NMauro Carvalho Chehab <mchehab@redhat.com>

Minimum vertical pixel size alignment for input and output DMA and
the scaler depend on color format, rotation, the IP instance and revision.

Make vertical pixel size of format and crop better fit for each SoC
revision and the IP instance by adding min_vsize_align attribute to
the FIMC variant data structure. It's now common for the DMA engines
and the scaler.
Signed-off-by: NSylwester Nawrocki <s.nawrocki@samsung.com>
Signed-off-by: NKyungmin Park <kyungmin.park@samsung.com>
Signed-off-by: NMauro Carvalho Chehab <mchehab@redhat.com>

When requested more than 2 buffers the buffer dequeue order was wrong
due to erroneous updating FIMC registers in every interrupt handler
call. This also fixes regression of resetting the output DMA buffer
pointer at wrong time, when some buffers are already queued in hardware.
The hardware is reset in the start_streaming callback in order to align
the H/W state with the software output buffer pointer (buf_index).

Additionally a simple write to S5P_CISCCTRL register is replaced with
a read/modification/write to make sure the scaler is not being disabled
in fimc_hw_set_scaler().
Signed-off-by: NSylwester Nawrocki <s.nawrocki@samsung.com>
Signed-off-by: NKyungmin Park <kyungmin.park@samsung.com>
Signed-off-by: NMauro Carvalho Chehab <mchehab@redhat.com>

ST_LPM bit must not be initially set, so the first resume helper
call properly quiesce the device's operation.
Also fimc_runtime_suspend() at device remove is unneeded and
leads to unbalanced clock disable so remove it.
Signed-off-by: NSylwester Nawrocki <s.nawrocki@samsung.com>
Signed-off-by: NKyungmin Park <kyungmin.park@samsung.com>
Signed-off-by: NMauro Carvalho Chehab <mchehab@redhat.com>

In preparation for the forthcoming VIDIOC_CREATE_BUFS ioctl add a
"const struct v4l2_format *" argument to the .queue_setup() vb2
operation.
Signed-off-by: NGuennadi Liakhovetski <g.liakhovetski@gmx.de>
Signed-off-by: NMauro Carvalho Chehab <mchehab@redhat.com>

The driver supports only multi-planar API and conversion to single-planar
API should be done in libv4l2. Remove misleading single planar capability
flags to avoid problems in applications and libv4l2.
Signed-off-by: NSylwester Nawrocki <s.nawrocki@samsung.com>
Signed-off-by: NKyungmin Park <kyungmin.park@samsung.com>
Signed-off-by: NMauro Carvalho Chehab <mchehab@redhat.com>

Horizontal crop offset must be multiple of 2 otherwise
color distortion occurs.
Signed-off-by: NSylwester Nawrocki <s.nawrocki@samsung.com>
Signed-off-by: NKyungmin Park <kyungmin.park@samsung.com>
Signed-off-by: NMauro Carvalho Chehab <mchehab@redhat.com>

Add support for whole pipeline suspend/resume. Sensors must support
suspend/resume through s_power subdev operation.
Signed-off-by: NSylwester Nawrocki <s.nawrocki@samsung.com>
Signed-off-by: NKyungmin Park <kyungmin.park@samsung.com>
Signed-off-by: NMauro Carvalho Chehab <mchehab@redhat.com>

Also correct and improve *_queue_add/pop functions description.
Signed-off-by: NSylwester Nawrocki <s.nawrocki@samsung.com>
Signed-off-by: NKyungmin Park <kyungmin.park@samsung.com>
Signed-off-by: NMauro Carvalho Chehab <mchehab@redhat.com>

Add support for transparent DMA transfer of JPEG data with MIPI-CSI2
USER1 format. In JPEG mode the color effect, scaling and cropping
is not supported as well as image rotation and flipping thus these
controls are marked as inactive if V4L2_PIX_FMT_JPEG pixel format
was selected.
Signed-off-by: NSylwester Nawrocki <s.nawrocki@samsung.com>
Signed-off-by: NKyungmin Park <kyungmin.park@samsung.com>
Signed-off-by: NMauro Carvalho Chehab <mchehab@redhat.com>

Add a subdev to expose the host's scaling and composition functions.
The camera frame composition onto an output buffer may be configured
through set/get_crop at FIMC.{n} source pad.
Additionally allow crop, composition and controls to be modified
during streaming. Make sure the default format is set when opening
the video capture node.
Rename struct fimc_vid_cap::fmt to more relevant 'mf' to avoid
confusion.
Signed-off-by: NSylwester Nawrocki <s.nawrocki@samsung.com>
Signed-off-by: NKyungmin Park <kyungmin.park@samsung.com>
Signed-off-by: NMauro Carvalho Chehab <mchehab@redhat.com>

In the camera capture driver use fimc_pipeline_* calls provided by
the media device driver part, in place where v4l2_subdev_call() were
used. This way the capture driver don't need to differentiate between
various H/W pipeline setups, i.e. if the MIPI-CSI receiver subdev is
used or not.

Remove the sync_capture_fmt() function instead of which
fimc_pipeline_try_format() is introduced in the following patch adding
the FIMC capture subdev.

The TRY_FMT ioctl function is completed by a subsequent patch adding
the capture subdev, so the try_fmt routines can be reused in the subdev
and the video node ioctl handlers.
Signed-off-by: NSylwester Nawrocki <s.nawrocki@samsung.com>
Signed-off-by: NKyungmin Park <kyungmin.park@samsung.com>
Signed-off-by: NMauro Carvalho Chehab <mchehab@redhat.com>

Replace fimc_find_format() and find_mbus_format() with single function
that can return a pointer to the private format description based
on fourcc, media bus code or index in the table.

Create separate VIDIOC_ENUM_FMT ioctl handlers for video capture
and mem-to-mem video node. This is needed because some formats are
valid only for the video capture and some only for the mem-to-mem
video node.
Signed-off-by: NSylwester Nawrocki <s.nawrocki@samsung.com>
Signed-off-by: NKyungmin Park <kyungmin.park@samsung.com>
Signed-off-by: NMauro Carvalho Chehab <mchehab@redhat.com>

Move the camera capture H/W initialization sequence to a separate
function. This is needed for reuse in the following runtime PM code.
Signed-off-by: NSylwester Nawrocki <s.nawrocki@samsung.com>
Signed-off-by: NKyungmin Park <kyungmin.park@samsung.com>
Signed-off-by: NMauro Carvalho Chehab <mchehab@redhat.com>

Convert the v4l controls code to use the new control framework.

fimc_ctrls_activate/deactivate functions are introduced for the
transparent DMA transfer mode (JPEG), where the rotation and flipping
controls are not supported.

The capture video node does not inherit sensors' controls when the
subdevs are configured by the user space (user_subdev_api == true).
However by default after the driver's initialization
the 'user-subdev_api' flag is false and any sensor controls will
also be available at the video node.

When the pipeline links are disconnected through the media device
the FIMC and any sensor inherited controls are destroyed and then
again created when the pipeline connection completes.
Signed-off-by: NSylwester Nawrocki <s.nawrocki@samsung.com>
Signed-off-by: NKyungmin Park <kyungmin.park@samsung.com>
Signed-off-by: NMauro Carvalho Chehab <mchehab@redhat.com>

This is a prerequisite for the patch converting the driver to use
the control framework. As the capture driver does not use per file
handle contexts, two separate ioctl handlers are created for it
(vidioc_try_fmt_mplane, and vidioc_g_fmt_mplane) so there is no
handlers shared between the memory-to-memory and capture video node.
Signed-off-by: NSylwester Nawrocki <s.nawrocki@samsung.com>
Signed-off-by: NKyungmin Park <kyungmin.park@samsung.com>
Signed-off-by: NMauro Carvalho Chehab <mchehab@redhat.com>

Add a top level media device driver aggregating FIMC video devnodes,
MIPI-CSIS and sensor subdevs. This driver gathers all media entities
and creates the possible links between them during initialization. By
default some links will be activated to enable access to all available
sensors in the system. For example if there are sensors S0, S1 listed
in the media device platform data definition they will be by default
assigned to FIMC0, FIMC1 respectively, which in turn will corresponds
to separate /dev/video?.
There is enough FIMC H/W entities to cover all available physical camera
interfaces in the system.

The fimc media device driver is bound to the "s5p-fimc-md" platform device.
Such platform device should be created by board initialization code
and camera sensors description array need to be specified as its
platform data.

The media device driver also implements various video pipeline operations,
for enabling subdevs power, streaming, etc., which will be used by the
capture video node driver.
Signed-off-by: NSylwester Nawrocki <s.nawrocki@samsung.com>
Signed-off-by: NKyungmin Park <kyungmin.park@samsung.com>
Signed-off-by: NMauro Carvalho Chehab <mchehab@redhat.com>

Currently there is a v4l2_device instance being registered per each
(capture and memory-to-memory) video node created per FIMC H/W instance.
This patch is a prerequisite for using the top level v4l2_device
instantiated by the media device driver.
To retain current debug trace semantic (so it's possible to distinguish
between the capture and m2m FIMC) the video_device is used in place
of v4l2_device where appropriate.
Signed-off-by: NSylwester Nawrocki <s.nawrocki@samsung.com>
Signed-off-by: NKyungmin Park <kyungmin.park@samsung.com>
Signed-off-by: NMauro Carvalho Chehab <mchehab@redhat.com>

There are 2 separate clock outputs available in the SoC for external sensors.
These two clocks can be shared among all FIMC entities and there is
currently no any arbitration of the clocks in the driver.

So make the capture driver not touching these clocks and let them be
be properly handled at the media device driver level, enabling proper
arbitration between FIMC entities.
Signed-off-by: NSylwester Nawrocki <s.nawrocki@samsung.com>
Signed-off-by: NKyungmin Park <kyungmin.park@samsung.com>
Signed-off-by: NMauro Carvalho Chehab <mchehab@redhat.com>

Do not register video nodes during FIMC device probe. Also make
fimc_register_m2m_device() public for use by the media device driver.
The video nodes are to be registered during the media device driver
initialization, altogether with the subdev devnodes. The video
capture nodes need to be registered as last ones when the remaining
pipeline elements are already initialized.
Signed-off-by: NSylwester Nawrocki <s.nawrocki@samsung.com>
Signed-off-by: NKyungmin Park <kyungmin.park@samsung.com>
Signed-off-by: NMauro Carvalho Chehab <mchehab@redhat.com>

Add intialization of the media entities for video capture
and mem-to-mem video nodes. The mem-to-mem entity has no pads
whereas the capture entity has single sink pad.
Also clean up the video node naming for consistency.
Signed-off-by: NSylwester Nawrocki <s.nawrocki@samsung.com>
Signed-off-by: NKyungmin Park <kyungmin.park@samsung.com>
Signed-off-by: NMauro Carvalho Chehab <mchehab@redhat.com>

Add runtime PM and system sleep support in the memory-to-memory
driver. It's required to enable the FIMC operation on Exynos4
SoCs. This patch prevents system boot failure when the driver
is compiled in, as it now tries to access its I/O memory without
first enabling the corresponding power domain.

The camera capture device suspend/resume is not fully covered,
the capture device is just powered on/off during the video
node open/close. However this enables it's normal operation
on Exynos4 SoCs.

[mchehab@redhat.com: fix a small checkpatch error]
Signed-off-by: NSylwester Nawrocki <s.nawrocki@samsung.com>
Signed-off-by: NKyungmin Park <kyungmin.park@samsung.com>
Signed-off-by: NMauro Carvalho Chehab <mchehab@redhat.com>

Use the correct 'dma_addr' name for the buffer address. 'paddr' suggested
that this is the physical address in system memory. For most ARM platforms
these two are the same, but this is not a generic rule. 'dma_addr' will
also point better to dma-mapping api.
Signed-off-by: NMarek Szyprowski <m.szyprowski@samsung.com>
Signed-off-by: NKyungmin Park <kyungmin.park@samsung.com>
CC: Pawel Osciak <pawel@osciak.com>
Acked-by: NGuennadi Liakhovetski <g.liakhovetski@gmx.de>
Signed-off-by: NMauro Carvalho Chehab <mchehab@redhat.com>

Plane sizes array was declared as unsigned long[], while unsigned int is
more than enough for storing size of the video buffer. This patch reduces
the size of the array by definiting it as unsigned int[].
Reported-by: NLaurent Pinchart <laurent.pinchart@ideasonboard.com>
Signed-off-by: NMarek Szyprowski <m.szyprowski@samsung.com>
Signed-off-by: NKyungmin Park <kyungmin.park@samsung.com>
CC: Pawel Osciak <pawel@osciak.com>
Signed-off-by: NMauro Carvalho Chehab <mchehab@redhat.com>

Standardize the remaining video drivers to return the API version
for the VIDIOC_QUERYCAP version, instead of a per-driver version.

Those drivers had the version updated more recently or are SoC
drivers. Even so, it doesn't sound a good idea to keep a per-driver
version control, so, let's use the per-subsystem version control
instead.
Acked-by: NHans Verkuil <hans.verkuil@cisco.com>
Signed-off-by: NMauro Carvalho Chehab <mchehab@redhat.com>

Signed-off-by: NSylwester Nawrocki <s.nawrocki@samsung.com>
Signed-off-by: NKyungmin Park <kyungmin.park@samsung.com>
Signed-off-by: NMauro Carvalho Chehab <mchehab@redhat.com>

With multi-planar formats fmt.pix_mp member of struct v4l2_format
should be used rather than fmt.pix. Fix find_fmt() function to do
the right thing.
Signed-off-by: NSylwester Nawrocki <s.nawrocki@samsung.com>
Signed-off-by: NKyungmin Park <kyungmin.park@samsung.com>
Signed-off-by: NMauro Carvalho Chehab <mchehab@redhat.com>

Avoid dereferencing of NULL f->fmt. Correct size of the allocated
buffer in case the crop rectangle is smaller than the bounds
rectangle (configured with S_FMT). Also remove redundant check
for *num_buffer == 0 as this case is handled in videobuf2.
Signed-off-by: NSylwester Nawrocki <s.nawrocki@samsung.com>
Signed-off-by: NKyungmin Park <kyungmin.park@samsung.com>
Signed-off-by: NMauro Carvalho Chehab <mchehab@redhat.com>

Correct inconsistencies in data structures' documentation.
Remove meaningless debug traces.
Signed-off-by: NSylwester Nawrocki <s.nawrocki@samsung.com>
Signed-off-by: NKyungmin Park <kyungmin.park@samsung.com>
Signed-off-by: NMauro Carvalho Chehab <mchehab@redhat.com>

Remove V4L2_MBUS_FMT_RGB565_2X8_BE media code entry as
camera interface supports only packed YUYV formats.
Signed-off-by: NSylwester Nawrocki <s.nawrocki@samsung.com>
Signed-off-by: NKyungmin Park <kyungmin.park@samsung.com>
Signed-off-by: NMauro Carvalho Chehab <mchehab@redhat.com>

Add support for buffer timestamps and the sequence number in
the video capture driver.
Signed-off-by: NSylwester Nawrocki <s.nawrocki@samsung.com>
Signed-off-by: NKyungmin Park <kyungmin.park@samsung.com>
Signed-off-by: NMauro Carvalho Chehab <mchehab@redhat.com>

Make sure the sizeimage for 3-planar color formats is
width * height * 3/2 and the bytesperline is same for each
plane in case of a multi-planar format.
Signed-off-by: NSylwester Nawrocki <s.nawrocki@samsung.com>
Signed-off-by: NKyungmin Park <kyungmin.park@samsung.com>
Signed-off-by: NMauro Carvalho Chehab <mchehab@redhat.com>

Protecting the color format with vb2_is_streaming() is not sufficient
as this prevents changing the format only after VIDIOC_STREAMON.
To prevent the color format reconfiguration as soon as buffers
are allocated use vb2_is_busy() instead.
Also make the videobuf queue ops structure static.
Signed-off-by: NSylwester Nawrocki <s.nawrocki@samsung.com>
Signed-off-by: NKyungmin Park <kyungmin.park@samsung.com>
Signed-off-by: NMauro Carvalho Chehab <mchehab@redhat.com>

Correct pixel limits for the fourth FIMC entity on Exynos4 SoCs.
FIMC3 only supports the writeback input from the LCD mixer.
Also rename s5pv310 variant to exynos4 which is needed after
renaming s5pv310 series to Exynos4.
Signed-off-by: NSylwester Nawrocki <s.nawrocki@samsung.com>
Signed-off-by: NKyungmin Park <kyungmin.park@samsung.com>
Signed-off-by: NMauro Carvalho Chehab <mchehab@redhat.com>

Use pix_mp member of struct v4l2_format to return a format
description rather than pix. Also fill in the plane_fmt array.
This is a missing bit of conversion to the multiplanar API.
Signed-off-by: NSylwester Nawrocki <s.nawrocki@samsung.com>
Signed-off-by: NKyungmin Park <kyungmin.park@samsung.com>
Signed-off-by: NMauro Carvalho Chehab <mchehab@redhat.com>

Add num_clients field to struct s5p_fimc_isp_info to define exactly
size of clients array which simplifies a bit the sensors management.
Signed-off-by: NSylwester Nawrocki <s.nawrocki@samsung.com>
Signed-off-by: NKyungmin Park <kyungmin.park@samsung.com>
Signed-off-by: NMauro Carvalho Chehab <mchehab@redhat.com>

Rework the locking in m2m driver to assure proper operation on SMP systems.

When job_abort or stop_streaming was called to immediately shutdown
a memory-to-memory transaction video buffers scheduled for processing
were never returned to vb2 and v4l2_m2m_job_finish was not called
which led to hanging.

Correct this and also return the unprocessed buffers to vb2 marking
them as erroneous, in case the end of frame interrupt do not occur.
Reported-by: NSewoon Park <seuni.park@samsung.com>
Signed-off-by: NSylwester Nawrocki <s.nawrocki@samsung.com>
Signed-off-by: NKyungmin Park <kyungmin.park@samsung.com>
Signed-off-by: NMauro Carvalho Chehab <mchehab@redhat.com>

In some cases fimc H/W did not stop although there were no output
buffers available. So the capture deactivation interrupt routine
is modified and the state of ST_CAPT_RUN is cleared only
in the LAST-IRQ call.

After LAST-IRQ is generated, H/W pointer will be skipped by 1 frame.
(reference by user manual) So, S/W pointer should be increased too.

Reviewed-by Jonghun Han <jonghun.han@samsung.com>
Signed-off-by: NSungchun Kang <sungchun.kang@samsung.com>
Signed-off-by: NSylwester Nawrocki <s.nawrocki@samsung.com>
Signed-off-by: NMauro Carvalho Chehab <mchehab@redhat.com>