| Age | Commit message (Collapse) | Author |
|---|
|
MJPEG is an extremely useful format given its compression allows
high framerates even over limited bandwith USB connections. Add
an MJPEG encoder class that converts YUV420 data into MJPEG data.
Where a libcamera-source does not support MJPEG natively, convert
YUV420 into MJPEG if the user tries to set MJPEG format.
Reviewed-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Signed-off-by: Daniel Scally <dan.scally@ideasonboard.com>
|
|
Add functions to handle a new entry into video_source_type, representing
data with an encoding step between the source and sink. When running
through this route, buffers are allocated and mmaped on both the source
and the sink, then imported to the source to be filled before being queued
to the sink.
Reviewed-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Signed-off-by: Daniel Scally <dan.scally@ideasonboard.com>
|
|
We need to be able to place encoded data into the sink device's bufs.
To facilitate that add a video_source operation that allows the source
to import buffers from the sink.
Add a .import_buffers() callback so that the libcamera-source can
access a sink's buffers.
Reviewed-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Signed-off-by: Daniel Scally <dan.scally@ideasonboard.com>
|
|
Add an enum to struct video_source that details the type of source
we're dealing with. This will be used to make decisions about buffer
allocation and handling in a more explicit way.
Use the video_source_type associated with a video_source to decide on
the appropriate allocation function and buffer handler at stream on time.
This is a more explicit method than relying on the presence of the
.alloc_buffers op and allows more flexibility.
Reviewed-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Signed-off-by: Daniel Scally <dan.scally@ideasonboard.com>
|
|
Provide the integration of a libcamera source within uvc-gadget.
This adds C++ support to the project and adds libcamera as an
optional external dependency.
Reviewed-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Signed-off-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Signed-off-by: Daniel Scally <dan.scally@ideasonboard.com>
|
|
The slideshow_source class is an implementation of the video_source
class that streams a set of images as video.
Reviewed-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Signed-off-by: Daniel Scally <dan.scally@ideasonboard.com>
Signed-off-by: Paul Elder <paul.elder@ideasonboard.com>
|
|
As written the fps output by the jpg-source is unconstrained; it will
simply fill buffers as quickly as the USB subsysem can handle them.
This is not particularly realistic and also means that the frame rates
configured (and thus expected) by the host are not being adhered to.
Use the new timer infrastructure to slow down the buffer fill operation
such that the frame rates seen match those configured by the host.
Reviewed-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Signed-off-by: Daniel Scally <dan.scally@ideasonboard.com>
|
|
Some sources simply fill the buffers passed by the USB subsystem and
return them as quickly as possible. Particularly with compressed
formats operating at superspeed this rapidly results in unrealistic
frame rates.
Add infrastructure that allows us to define a specific
framerate and introduce blocking calls that constrain those sources
to the framerates expected by the host.
Reviewed-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Signed-off-by: Daniel Scally <dan.scally@ideasonboard.com>
|
|
Without a response, the host will assume there has been an error and
throw a bunch of errors. The noise of those errors makes it difficult
to see what's going on properly, so return a dummy response that
silences the errors until the functionality is properly implemented.
Reviewed-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Signed-off-by: Daniel Scally <dan.scally@ideasonboard.com>
Signed-off-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
|
|
It's easier to see what's going on at a glance when the output is
human-readable, so replace the control codes with their names.
Reviewed-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Signed-off-by: Daniel Scally <dan.scally@ideasonboard.com>
Signed-off-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
|
|
The maxsize field of struct uvc_device is probably unecessary even if
it weren't currently fixed at 0. It's used to set dwMaxVideoFrameSize
in addition to the sizeimage member of a struct v4l2_pix_fmt (multiplied
in the latter instance by 1.5x). The documentary definitions for those
two values are such that they should be identical though, so simply
set dwMaxVideoFrameSize to width*height*2 (to maintain commonality
with other UVC Video devices) and use that to set the sizeimage.
Reviewed-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Signed-off-by: Daniel Scally <dan.scally@ideasonboard.com>
Signed-off-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
|
|
For the UVC_EVENT_STREAMOFF event, the callback frees _other_ events
because it's closing down the pipe. The use of the _safe() version
of the list iterator results in a use-after-free of those events
because they're fetched before the callback that frees them. The
_safe() version is unecessary here; no other callback's behaviour
requires it.
Reviewed-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Signed-off-by: Daniel Scally <dan.scally@ideasonboard.com>
Signed-off-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
|
|
The jpg_source class is an implementation of the video_source class that
streams a still jpg image as video.
Reviewed-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Signed-off-by: Paul Elder <paul.elder@ideasonboard.com>
Signed-off-by: Daniel Scally <dan.scally@ideasonboard.com>
Signed-off-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
|
|
The test_source class is an implementation of the video_source class
that generates a test pattern in userspace to provide video. We have
some operations which are implemented as no-ops. To quiet the
compiler warnings this causes, flag their arguments as unused.
Reviewed-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Signed-off-by: Paul Elder <paul.elder@ideasonboard.com>
Signed-off-by: Daniel Scally <dan.scally@ideasonboard.com>
Signed-off-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
|
|
When we dequeue a buffer from a V4L2 device, it would be good to know
the size of the buffer. Save it to our struct video_buffer when
dequeueing.
Reviewed-by: Daniel Scally <dan.scally@ideasonboard.com>
Reviewed-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Signed-off-by: Paul Elder <paul.elder@ideasonboard.com>
Signed-off-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
|
|
Our v4l2_queue_buffer and v4l2_dequeue_buffer functions give us a
wrapper around the raw V4L2 ioctls with a wrapper struct video_buffer
around the raw struct v4l2_buffer. The problem is that the
video_buffer's mem field is never populated, so there's no way to access
the mem that the buffer refers to given only the buffer.
Populate the video_buffer's mem field at v4l2_dequeue_buffer based on
the dequeued v4l2 buf's index.
Reviewed-by: Daniel Scally <dan.scally@ideasonboard.com>
Reviewed-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Signed-off-by: Paul Elder <paul.elder@ideasonboard.com>
Signed-off-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
|
|
We want to allow video sources whose data is generated in userspace.
Currently buffers are allocated on the V4L2-backed video source and
exported to the uvc-gadget's V4L2 device. Since a video source with
data generated in userspace will not be backed by a V4L2 device, we
allow a way for buffers to be allocated on uvc-gadget's V4L2 device.
As a corollary, the way the buffers are passed has to be changed as
well. Previously they were queued and dequeued between the V4L2 video
source and uvc and vice versa. To allow a video source not backed by
a V4L2 device, we allow an alternative flow where they are dequeued and
queued from and to uvc, and giving a chance to the video source to fill
the buffer in between.
Reviewed-by: Daniel Scally <dan.scally@ideasonboard.com>
Reviewed-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Signed-off-by: Paul Elder <paul.elder@ideasonboard.com>
Signed-off-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
|
|
We are preparing to allow video sources whose data is generated in
userspace. To this end, add a fill_buffer function to video_source.
Reviewed-by: Daniel Scally <dan.scally@ideasonboard.com>
Reviewed-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Signed-off-by: Paul Elder <paul.elder@ideasonboard.com>
Signed-off-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
|
|
Provide infrastructure to be able to build the uvc-gadget library as a
shared object and an executable which links against this as a dependency
to implment the reference 'uvc-gadget' application.
All existing cmake infrastructure is removed. This removes custom glob
support that was previously used by an early development for Android,
which is expected to be re-added later when required.
Reviewed-by: Daniel Scally <dan.scally@ideasonboard.com>
Signed-off-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
|
|
Bringing in the meson infrastructure will enable extra warnings,
including -Wunused...
Directly reference the two locations that have known unused arguments.
Reviewed-by: Daniel Scally <dan.scally@ideasonboard.com>
Signed-off-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
|
|
At some point the kernel has been updated to include a .id descriptor
in the gadget directory name - add that to uvc_find_video_device() so
that the video node is located properly.
Rather than hardcode a specific gadget (for instance with device name
gadget.0) use a wildcard. This has the advantage of backwards
compatability with older kernels that did not use the ID as part of the
device name, though still limits us to supporting scenarios where only a
single gadget is bound to the UDC.
Reviewed-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Signed-off-by: Daniel Scally <dan.scally@ideasonboard.com>
|
|
videobuf2 relies on the sizeimage field to calculate buffer size
when the format is MJPEG, because the bytes-per-line calculation
is missing. Fill in that field as part of v4l2_set_format() so the
API works correctly.
Reviewed-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Signed-off-by: Daniel Scally <dan.scally@ideasonboard.com>
|
|
The MJPEG descriptor does not have a guidFormat, and so configfs doesn't
have it either, but we currently error out if there is no guidFormat.
To fix this, extract the stream type part of the real path, and if it is
MJPEG then hardcode it, otherwise read the guid if it is uncompressed.
This does mean that in configfs the MJPEG format descriptor path must be
"mjpeg" for this to work, though this is already guranteed by configfs.
Signed-off-by: Paul Elder <paul.elder@ideasonboard.com>
Reviewed-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
|
|
Previously, the interval rate that would be selected and replied to the
host in response to UVC_GET_MAX, UVC_GET_MIN, and UVC_GET_DEF were all
the same and were the first interval rate. Assuming the interval rates
are sorted in ascending order, this means that UVC_GET_MAX would always
yield the minimum interval rate.
Modify the call to uvc_fill_streaming_control such that UVC_GET_MAX will
yield the maximum interval rate.
Signed-off-by: Paul Elder <paul.elder@ideasonboard.com>
Reviewed-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Reviewed-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
|
|
We would like to be able to set the frame rate of the video source of a
stream. This patch adds such a function to set the source's frame rate
from the stream. The frame rate is set right after the format is set.
Signed-off-by: Paul Elder <paul.elder@ideasonboard.com>
Reviewed-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Reviewed-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
|
|
Implement the video source op set_frame_rate for v4l2-source video
source.
Signed-off-by: Paul Elder <paul.elder@ideasonboard.com>
Reviewed-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Reviewed-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
|
|
We would like to be able to set the frame rate for video sources. This
patch adds a set_frame_rate op for video sources, along with a wrapper
to call video sources' set_frame_rate functions.
Signed-off-by: Paul Elder <paul.elder@ideasonboard.com>
Reviewed-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Reviewed-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
|
|
We would like to be able to set the frame rate of video sources. This
patch adds such a function for v4l2-based video sources to use.
Signed-off-by: Paul Elder <paul.elder@ideasonboard.com>
Reviewed-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Reviewed-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
|
|
Parse /proc/mounts to identify the correct mount location of the
ConfigFS filesystem.
Signed-off-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Reviewed-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
|
|
The Android C library doesn't provide the glob() function which we use
in the configfs code. Add conditional compilation to cmake such that a
local copy of glob is automatically compiled if glob isn't provided.
Signed-off-by: Paul Elder <paul.elder@ideasonboard.com>
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
|
|
The glibc glob() implementation doesn't compile on Android due to
missing support for getlogin_r (before Android ABI 28) and to missing
__THROW, __THROWNL and __attribute_noinline__ macros.
getlogin_r is only used to implement support for the '~' path component
which we don't use in the uvc-gadget library, so we can compile it out.
The three macros can safely be defined as no-op.
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
|
|
The glibc glob() implementation seems to be meant to be compilable
outside of glibc, but that clearly hasn't been tested for some time. Fix
compilation.
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
|
|
The Android libc implementation (bionic) doesn't provide the glob()
function. To enable compilation of the UVC gadget library for Android,
import the glibc glob() implementation.
The glob.c and glob.h files are imported unmodified to ease copyright
management, even if they don't compile as-is. Compilation will be fixed
separately.
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
|
|
Android now uses clang as the default C compiler, and gcc support is
deprecated. clang doesn't support nested functions, which were used to
implement callback functions in the ConfigFS code. We thus move the
callbacks to the global scope.
Signed-off-by: Paul Elder <paul.elder@ideasonboard.com>
Reviewed-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
|
|
The asprintf function returns -1, and leaves the first parameter
undefined in case of an allocation error. This conflicts with the
code's assumption that the parameter is set NULL on error.
Check all return values of asprintf, and ensure that the string
parameter is not used as a return value on error paths.
Signed-off-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
|
|
There is code that is meant to check for the error of an ioctl, but the
return variable that it checks isn't actually assigned from the ioctl.
Assign the return value before checking it.
Signed-off-by: Paul Elder <paul.elder@ideasonboard.com>
[Fixed error message]
Reviewed-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
|
|
In order to avoid depending on an external kernel source tree recent
enough to provide the g_uvc.h header, import in the the project.
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
|
|
To allow usage of the UVC gadget library in closed-source or otherwise
GPL-incompatible applications, relicense the library under the terms of
the LGPL v2.1 or later, as published by the Free Software Foundation.
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
|
|
Ideas on Board SPRL, the original copyright owner of the UVC gadget
application code, has been dissolved. All company copyrights have been
transferred to Laurent Pinchart. Update the copyright notices to
accordingly.
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
|
|
Replace the boilerplate copyright headers with an SPDX header to ease
license compliance. This doesn't change the license of any of the files.
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
|
|
Split the project into a UVC gadget library and a test application. To
avoid rolling out a custom build system, switch to CMake.
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
|
