This introduces the Cedrus VPU driver that supports the VPU found in
Allwinner SoCs, also known as Video Engine. It is implemented through
a V4L2 M2M decoder device and a media device (used for media requests).
So far, it only supports MPEG-2 decoding.

Since this VPU is stateless, synchronization with media requests is
required in order to ensure consistency between frame headers that
contain metadata about the frame to process and the raw slice data that
is used to generate the frame.

This driver was made possible thanks to the long-standing effort
carried out by the linux-sunxi community in the interest of reverse
engineering, documenting and implementing support for the Allwinner VPU.

Signed-off-by: Paul Kocialkowski <paul.kocialkowski@bootlin.com>
Acked-by: Maxime Ripard <maxime.ripard@bootlin.com>
Signed-off-by: Hans Verkuil <hans.verkuil@cisco.com>
[hans.verkuil@cisco.com: dropped obsolete MEDIA_REQUEST_API from Kconfig]
Signed-off-by: Hans Verkuil <hans.verkuil@cisco.com>
Signed-off-by: Mauro Carvalho Chehab <mchehab+samsung@kernel.org>

This adds a device-tree binding document that specifies the properties
used by the Cedrus VPU driver, as well as examples.

Signed-off-by: Paul Kocialkowski <paul.kocialkowski@bootlin.com>
Reviewed-by: Rob Herring <robh@kernel.org>
Signed-off-by: Maxime Ripard <maxime.ripard@bootlin.com>
Signed-off-by: Hans Verkuil <hans.verkuil@cisco.com>
Signed-off-by: Mauro Carvalho Chehab <mchehab+samsung@kernel.org>

This introduces support for the Sunxi tiled NV12 format, where each
component of the YUV frame is divided into macroblocks. Hence, the size
of each plane requires specific alignment. The pixels inside each
macroblock are coded in linear order (line after line from top to
bottom).

This tiled NV12 format is used by the video engine on Allwinner
platforms: it is the default format for decoded frames (and the only
one available in the oldest supported platforms).

Signed-off-by: Paul Kocialkowski <paul.kocialkowski@bootlin.com>
Signed-off-by: Maxime Ripard <maxime.ripard@bootlin.com>
Signed-off-by: Hans Verkuil <hans.verkuil@cisco.com>
Signed-off-by: Mauro Carvalho Chehab <mchehab+samsung@kernel.org>

Stateless video decoding engines require both the MPEG-2 slices and
associated metadata from the video stream in order to decode frames.

This introduces definitions for a new pixel format, describing buffers
with MPEG-2 slice data, as well as control structure sfor passing the
frame metadata to drivers.

This is based on work from both Florent Revest and Hugues Fruchet.

Signed-off-by: Paul Kocialkowski <paul.kocialkowski@bootlin.com>
Signed-off-by: Maxime Ripard <maxime.ripard@bootlin.com>
Signed-off-by: Hans Verkuil <hans.verkuil@cisco.com>
Signed-off-by: Mauro Carvalho Chehab <mchehab+samsung@kernel.org>

The helper indicating whether buffers are associated with the request is
reworked and renamed to return the number of associated buffer objects.

This is useful for drivers that need to check how many buffers are in
the request to validate it.

Existing users of the helper don't need particular adaptation since the
meaning of zero/non-zero remains consistent.

Signed-off-by: Paul Kocialkowski <contact@paulk.fr>
Signed-off-by: Maxime Ripard <maxime.ripard@bootlin.com>
Signed-off-by: Hans Verkuil <hans.verkuil@cisco.com>
Signed-off-by: Mauro Carvalho Chehab <mchehab+samsung@kernel.org>

Fix this smatch error:

drivers/media/v4l2-core/v4l2-ctrls.c:2971 v4l2_ctrl_request_clone() error: uninitialized symbol 'err'.

Signed-off-by: Hans Verkuil <hans.verkuil@cisco.com>
Acked-by: Sakari Ailus <sakari.ailus@linux.intel.com>
Signed-off-by: Mauro Carvalho Chehab <mchehab+samsung@kernel.org>

Fix this sparse warning:

drivers/media/v4l2-core/v4l2-compat-ioctl32.c:256: warning:
Function parameter or member 'capabilities' not described in 'v4l2_create_buffers32'

Signed-off-by: Hans Verkuil <hans.verkuil@cisco.com>
Acked-by: Sakari Ailus <sakari.ailus@linux.intel.com>
Signed-off-by: Mauro Carvalho Chehab <mchehab+samsung@kernel.org>

Various clarifications and readability improvements based on
Laurent Pinchart's review of the documentation.

Signed-off-by: Hans Verkuil <hans.verkuil@cisco.com>
Reviewed-by: Tomasz Figa <tfiga@chromium.org>
Acked-by: Sakari Ailus <sakari.ailus@linux.intel.com>
Signed-off-by: Mauro Carvalho Chehab <mchehab+samsung@kernel.org>

If requests are not supported by the driver, then return EACCES, not
EPERM.

If you attempt to mix queueing buffers directly and using requests,
then EBUSY is returned instead of EPERM: once a specific queueing mode
has been chosen the queue is 'busy' if you attempt the other mode
(i.e. direct queueing vs via a request).

Signed-off-by: Hans Verkuil <hans.verkuil@cisco.com>
Reviewed-by: Tomasz Figa <tfiga@chromium.org>
Acked-by: Sakari Ailus <sakari.ailus@linux.intel.com>
Signed-off-by: Mauro Carvalho Chehab <mchehab+samsung@kernel.org>

The request reference count was decreased again once a reference to the
request object was taken. Postpone this until we finished using the object.

In theory I think it is possible that the request_fd can be closed by
the application from another thread. In that case when request_put is
called the whole request would be freed.

It's highly unlikely, but let's just be safe and fix this potential
race condition.

Signed-off-by: Hans Verkuil <hans.verkuil@cisco.com>
Reviewed-by: Tomasz Figa <tfiga@chromium.org>
Acked-by: Sakari Ailus <sakari.ailus@linux.intel.com>
Signed-off-by: Mauro Carvalho Chehab <mchehab+samsung@kernel.org>

When getting control values from a completed request, we have
to protect the request against being re-inited when it is
being accessed by calling media_request_(un)lock_for_access.

Signed-off-by: Hans Verkuil <hans.verkuil@cisco.com>
Reviewed-by: Tomasz Figa <tfiga@chromium.org>
Acked-by: Sakari Ailus <sakari.ailus@linux.intel.com>
Signed-off-by: Mauro Carvalho Chehab <mchehab+samsung@kernel.org>

Add helper functions to prevent a completed request from being
re-inited while it is being accessed.

Signed-off-by: Hans Verkuil <hans.verkuil@cisco.com>
Reviewed-by: Tomasz Figa <tfiga@chromium.org>
Acked-by: Sakari Ailus <sakari.ailus@linux.intel.com>
Signed-off-by: Mauro Carvalho Chehab <mchehab+samsung@kernel.org>

Set the capabilities field of v4l2_requestbuffers and v4l2_create_buffers.

The various mapping modes were easy, but for signaling the request capability
a new 'supports_requests' bitfield was added to videobuf2-core.h (and set in
vim2m and vivid). Drivers have to set this bitfield for any queue where
requests are supported.

Signed-off-by: Hans Verkuil <hansverk@cisco.com>
Reviewed-by: Tomasz Figa <tfiga@chromium.org>
Acked-by: Sakari Ailus <sakari.ailus@linux.intel.com>
Signed-off-by: Mauro Carvalho Chehab <mchehab+samsung@kernel.org>

VIDIOC_REQBUFS and VIDIOC_CREATE_BUFFERS will return capabilities
telling userspace what the given buffer type is capable of.

Signed-off-by: Hans Verkuil <hansverk@cisco.com>
Reviewed-by: Tomasz Figa <tfiga@chromium.org>
Acked-by: Sakari Ailus <sakari.ailus@linux.intel.com>
Signed-off-by: Mauro Carvalho Chehab <mchehab+samsung@kernel.org>

Document that V4L2_BUF_FLAG_REQUEST_FD should only be used with
VIDIOC_QBUF and cleared otherwise.

Signed-off-by: Hans Verkuil <hansverk@cisco.com>
Reviewed-by: Tomasz Figa <tfiga@chromium.org>
Acked-by: Sakari Ailus <sakari.ailus@linux.intel.com>
Signed-off-by: Mauro Carvalho Chehab <mchehab+samsung@kernel.org>

For now (this might be relaxed in the future) we do not allow getting
controls from a request that isn't completed. In that case we return
-EACCES. Update the documentation accordingly.

Signed-off-by: Hans Verkuil <hansverk@cisco.com>
Reviewed-by: Tomasz Figa <tfiga@chromium.org>
Acked-by: Sakari Ailus <sakari.ailus@linux.intel.com>
Signed-off-by: Mauro Carvalho Chehab <mchehab+samsung@kernel.org>

Instead of returning -ENOENT when a request_fd was not found (VIDIOC_QBUF
and VIDIOC_G/S/TRY_EXT_CTRLS), we now return -EINVAL. This is in line
with what we do when invalid dmabuf fds are passed to e.g. VIDIOC_QBUF.

Also document that EINVAL is returned for invalid m.fd values, we never
documented that.

Signed-off-by: Hans Verkuil <hansverk@cisco.com>
Reviewed-by: Tomasz Figa <tfiga@chromium.org>
Acked-by: Sakari Ailus <sakari.ailus@linux.intel.com>
Signed-off-by: Mauro Carvalho Chehab <mchehab+samsung@kernel.org>

Add support for requests to vivid.

Signed-off-by: Hans Verkuil <hans.verkuil@cisco.com>
Reviewed-by: Mauro Carvalho Chehab <mchehab+samsung@kernel.org>
Signed-off-by: Mauro Carvalho Chehab <mchehab+samsung@kernel.org>

Add support for the media_device to vivid. This is a prerequisite
for request support.

Signed-off-by: Hans Verkuil <hans.verkuil@cisco.com>
Reviewed-by: Mauro Carvalho Chehab <mchehab+samsung@kernel.org>
Signed-off-by: Mauro Carvalho Chehab <mchehab+samsung@kernel.org>

Add support for requests to vim2m.

Signed-off-by: Hans Verkuil <hans.verkuil@cisco.com>
Reviewed-by: Mauro Carvalho Chehab <mchehab+samsung@kernel.org>
Signed-off-by: Mauro Carvalho Chehab <mchehab+samsung@kernel.org>

v4l2_ctrl uses mutexes, so we can't setup a ctrl_handler in
interrupt context. Switch to a workqueue instead and drop the timer.

Signed-off-by: Hans Verkuil <hans.verkuil@cisco.com>
Reviewed-by: Mauro Carvalho Chehab <mchehab+samsung@kernel.org>
Signed-off-by: Mauro Carvalho Chehab <mchehab+samsung@kernel.org>

For mem2mem devices we have to make sure that v4l2_m2m_try_schedule()
is called whenever a request is queued.

We do that by creating a vb2_m2m_request_queue() helper that should
be used instead of the 'normal' vb2_request_queue() helper. The m2m
helper function will call v4l2_m2m_try_schedule() as needed.

In addition we also avoid calling v4l2_m2m_try_schedule() when preparing
or queueing a buffer for a request since that is no longer needed.
Instead this helper function will do that when the request is actually
queued.

Signed-off-by: Hans Verkuil <hans.verkuil@cisco.com>
Reviewed-by: Mauro Carvalho Chehab <mchehab+samsung@kernel.org>
Signed-off-by: Mauro Carvalho Chehab <mchehab+samsung@kernel.org>

Check if the vb2 queue uses requests, and if so refuse to
add buffers that are not part of a request. Also check for
the reverse: a vb2 queue did not use requests, and an attempt
was made to queue a buffer to a request.

We might relax this in the future, but for now just return
-EPERM in that case.

Signed-off-by: Hans Verkuil <hans.verkuil@cisco.com>
Reviewed-by: Mauro Carvalho Chehab <mchehab+samsung@kernel.org>
Signed-off-by: Mauro Carvalho Chehab <mchehab+samsung@kernel.org>

Set the first time a buffer from a request is queued to vb2
(uses_requests) or directly queued (uses_qbuf).
Cleared when the queue is canceled.

Signed-off-by: Hans Verkuil <hans.verkuil@cisco.com>
Reviewed-by: Mauro Carvalho Chehab <mchehab+samsung@kernel.org>
Signed-off-by: Mauro Carvalho Chehab <mchehab+samsung@kernel.org>

The generic vb2_request_validate helper function checks if
there are buffers in the request and if so, prepares (validates)
all objects in the request.

The generic vb2_request_queue helper function queues all buffer
objects in the validated request.

Signed-off-by: Hans Verkuil <hans.verkuil@cisco.com>
Reviewed-by: Mauro Carvalho Chehab <mchehab+samsung@kernel.org>
Signed-off-by: Mauro Carvalho Chehab <mchehab+samsung@kernel.org>

Add a new helper function to tell if a request object is a buffer.

Add a new helper function that returns true if a media_request
contains at least one buffer.

Signed-off-by: Hans Verkuil <hans.verkuil@cisco.com>
Reviewed-by: Mauro Carvalho Chehab <mchehab+samsung@kernel.org>
Signed-off-by: Mauro Carvalho Chehab <mchehab+samsung@kernel.org>

This implements the V4L2 part of the request support. The main
change is that vb2_qbuf and vb2_prepare_buf now have a new
media_device pointer. This required changes to several drivers
that did not use the vb2_ioctl_qbuf/prepare_buf helper functions.

Signed-off-by: Hans Verkuil <hans.verkuil@cisco.com>
Reviewed-by: Mauro Carvalho Chehab <mchehab+samsung@kernel.org>
Signed-off-by: Mauro Carvalho Chehab <mchehab+samsung@kernel.org>

Buffers can now be prepared or queued for a request.

A buffer is unbound from the request at vb2_buffer_done time or
when the queue is cancelled.

Signed-off-by: Hans Verkuil <hans.verkuil@cisco.com>
Reviewed-by: Mauro Carvalho Chehab <mchehab+samsung@kernel.org>
Signed-off-by: Mauro Carvalho Chehab <mchehab+samsung@kernel.org>

Make vb2_buffer a request object.

Signed-off-by: Hans Verkuil <hans.verkuil@cisco.com>
Reviewed-by: Mauro Carvalho Chehab <mchehab+samsung@kernel.org>
Signed-off-by: Mauro Carvalho Chehab <mchehab+samsung@kernel.org>

We need to initialize the request_fd field in struct vb2_v4l2_buffer
to -1 instead of the default of 0. So we need to add a new op that
is called when struct vb2_v4l2_buffer is allocated.

Signed-off-by: Hans Verkuil <hans.verkuil@cisco.com>
Reviewed-by: Mauro Carvalho Chehab <mchehab+samsung@kernel.org>
Signed-off-by: Mauro Carvalho Chehab <mchehab+samsung@kernel.org>

When queuing buffers allow for passing the request that should
be associated with this buffer.

If V4L2_BUF_FLAG_REQUEST_FD is set, then request_fd is used as
the file descriptor.

If a buffer is stored in a request, but not yet queued to the
driver, then V4L2_BUF_FLAG_IN_REQUEST is set.

Signed-off-by: Hans Verkuil <hans.verkuil@cisco.com>
Reviewed-by: Mauro Carvalho Chehab <mchehab+samsung@kernel.org>
Signed-off-by: Mauro Carvalho Chehab <mchehab+samsung@kernel.org>

The PREPARED state becomes a problem with the request API: a buffer
could be PREPARED but dequeued, or PREPARED and in state IN_REQUEST.

PREPARED is really not a state as such, but more a property of the
buffer. So make new 'prepared' and 'synced' bools instead to remember
whether the buffer is prepared and/or synced or not.

V4L2_BUF_FLAG_PREPARED is only set if the buffer is both synced and
prepared and in the DEQUEUED state.

Signed-off-by: Hans Verkuil <hans.verkuil@cisco.com>
Reviewed-by: Mauro Carvalho Chehab <mchehab+samsung@kernel.org>
Signed-off-by: Mauro Carvalho Chehab <mchehab+samsung@kernel.org>

There is no need to check the vb2 state in the buf_prepare
callback: it can never be wrong.

Since VB2_BUF_STATE_PREPARED will be removed in the next patch
we'll remove this unnecessary check (and use of that state) first.

Signed-off-by: Hans Verkuil <hans.verkuil@cisco.com>
Reviewed-by: Mauro Carvalho Chehab <mchehab+samsung@kernel.org>
Signed-off-by: Mauro Carvalho Chehab <mchehab+samsung@kernel.org>

The userspace-provided plane data needs to be stored in
vb2_v4l2_buffer. Currently this information is applied by
__fill_vb2_buffer() which is called by the core prepare_buf
and qbuf functions, but when using requests these functions
aren't called yet since the buffer won't be prepared until
the media request is actually queued.

In the meantime this information has to be stored somewhere
and vb2_v4l2_buffer is a good place for it.

The __fill_vb2_buffer callback now just copies the relevant
information from vb2_v4l2_buffer into the planes array.

Signed-off-by: Hans Verkuil <hans.verkuil@cisco.com>
Signed-off-by: Mauro Carvalho Chehab <mchehab+samsung@kernel.org>

Replace 'if' statements by a switch in __fill_vb2_buffer()
in preparation of the next patch.

No other changes.

Signed-off-by: Hans Verkuil <hans.verkuil@cisco.com>
Reviewed-by: Mauro Carvalho Chehab <mchehab+samsung@kernel.org>
Signed-off-by: Mauro Carvalho Chehab <mchehab+samsung@kernel.org>

Move the __fill_v4l2_buffer() to before the vb2_queue_or_prepare_buf()
function to prepare for the next two patches.

No other changes.

Signed-off-by: Hans Verkuil <hans.verkuil@cisco.com>
Reviewed-by: Mauro Carvalho Chehab <mchehab+samsung@kernel.org>
Signed-off-by: Mauro Carvalho Chehab <mchehab+samsung@kernel.org>

If a driver needs to find/inspect the controls set in a request then
it can use these functions.

E.g. to check if a required control is set in a request use this in the
req_validate() implementation:

	int res = -EINVAL;

	hdl = v4l2_ctrl_request_hdl_find(req, parent_hdl);
	if (hdl) {
		if (v4l2_ctrl_request_hdl_ctrl_find(hdl, ctrl_id))
			res = 0;
		v4l2_ctrl_request_hdl_put(hdl);
	}
	return res;

Signed-off-by: Hans Verkuil <hans.verkuil@cisco.com>
Reviewed-by: Mauro Carvalho Chehab <mchehab+samsung@kernel.org>
Signed-off-by: Mauro Carvalho Chehab <mchehab+samsung@kernel.org>

The v4l2_g/s_ext_ctrls functions now support control handlers that
represent requests.

The v4l2_ctrls_find_req_obj() function is responsible for finding the
request from the fd.

Signed-off-by: Hans Verkuil <hans.verkuil@cisco.com>
Reviewed-by: Mauro Carvalho Chehab <mchehab+samsung@kernel.org>
Signed-off-by: Mauro Carvalho Chehab <mchehab+samsung@kernel.org>

Integrate the request support. This adds the v4l2_ctrl_request_complete
and v4l2_ctrl_request_setup functions to complete a request and (as a
helper function) to apply a request to the hardware.

It takes care of queuing requests and correctly chaining control values
in the request queue.

Note that when a request is marked completed it will copy control values
to the internal request state. This can be optimized in the future since
this is sub-optimal when dealing with large compound and/or array controls.

For the initial 'stateless codec' use-case the current implementation is
sufficient.

Signed-off-by: Hans Verkuil <hans.verkuil@cisco.com>
Reviewed-by: Mauro Carvalho Chehab <mchehab+samsung@kernel.org>
Signed-off-by: Mauro Carvalho Chehab <mchehab+samsung@kernel.org>

The next patch needs the reference to a control instead of the
control itself, so change struct v4l2_ctrl_helper accordingly.

Signed-off-by: Hans Verkuil <hans.verkuil@cisco.com>
Reviewed-by: Mauro Carvalho Chehab <mchehab+samsung@kernel.org>
Signed-off-by: Mauro Carvalho Chehab <mchehab+samsung@kernel.org>