The libcamera_generated_ipa_headers variable, containing the list of
generated IPA headers, is listed in the sources of IPA modules, as well
as IPA tests. This was done to ensure that the modules and tests get
rebuilt when the generate IPA headers change. However, the dependency is
already handled through the libcamera_private dependency object,
specified for all those modules and tests. There's no need to list the
IPA generated headers as sources. Drop them.
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Reviewed-by: Daniel Scally <dan.scally@ideasonboard.com>
Reviewed-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
The *::Private classes are part of the internal API, as their name
implies. They are defined in internal headers, but implemented in the
same source file as their public counterparts. This will cause Doxygen
to complain about missing class definition when splitting the public and
internal API documents, as the internal headers won't be parsed by
Doxygen for the public API documentation.
Marking the classes with \internal isn't enough. The directive prevents
the documentation block from being included in the output, but this
occurs at the generation stage, after the documentation blocks are
parsed. Fix this by completely hidding the implementation of the
*::Private classes from Doxygen using preprocessor conditional
compilation. To do so, introduce a new macro, __DOXYGEN_PUBLIC__, that
will be defined for the public API documentation only.
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Reviewed-by: Daniel Scally <dan.scally@ideasonboard.com>
Reviewed-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
The libcamera public API exposes classes that have parts considered
internal. They inherit the Extensible class, and their internal parts
are split into a Private class. Those classes are defined in public API
headers, and their Private counterparts are defined in internal headers
sharing a common file name (in a different directory). Both headers are
documented in the same source file.
For instance, include/libcamera/camera.h contains the public API of the
Camera class, and include/libcamera/internal/camera.h its internal
counterpart. Both are documented in src/libcamera/camera.cpp.
As the internal headers are not part of the public API, they need to be
hidden from the future public API builds. To prepare for doing so, mark
them with the \internal Doxygen directive. Hardcode the Doxygen
INTERNAL_DOCS option to YES to include the internal API. This will be
changed later for the public API documentation build.
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Reviewed-by: Daniel Scally <dan.scally@ideasonboard.com>
Reviewed-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
libcamera has two formats.h headers, an internal one in
include/libcamera/internal/, and a public one generated at build time.
The convention is to prefix the internal header name with
libcamera/internal/ in the Doxygen file directive, but formats.cpp only
uses internal/ as a prefix. Unify it with the rest of the code base.
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Reviewed-by: Daniel Scally <dan.scally@ideasonboard.com>
Reviewed-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
The Doxygen directive only requires qualifying header file names with a
path to differentiate between multiple header files with the same name.
Most file directives that refer to unambiguous files do not have a
libcamera/ and/or internal/ path prefix, but a few do, most likely due
to copy&paste. Drop the prefix in those few files for consistency.
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Reviewed-by: Daniel Scally <dan.scally@ideasonboard.com>
Reviewed-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
libcamera is implemented in C++, use std::vector<> to manage the
dynamically allocated line buffers instead of malloc() and free(). This
simplifies the code and improves memory safety by ensuring no allocation
will be leaked.
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Reviewed-by: Milan Zamazal <mzamazal@redhat.com>
Reviewed-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Tested-by: Hans de Goede <hdegoede@redhat.com>
When the configuration file for an IPA module is missing, it is reported
as an error in the log, for example:
ERROR IPAProxy ipa_proxy.cpp:149 Configuration file 'imx219.yaml' not found for IPA module 'simple'
This is misleading because several pipelines use uncalibrated.yaml in
such a case and can continue working. And in case of software ISP,
there is currently no other configuration file so the error is always
reported.
On the other hand, in some other cases the presence of the configuration
file is required and it is an error if it is missing.
Let's introduce a new optional argument to IPAProxy::configurationFile
that specifies a fallback file if the requested file is not found. If
the primary requested file is not found and a non-empty fallback file is
specified then a warning is logged and the fallback file is looked up.
If neither the fallback file can be found then only then an error is
logged and the method returns an empty string. This change has also the
benefit of putting the common fallback file ("uncalibrated.yaml")
pattern to a single place.
Signed-off-by: Milan Zamazal <mzamazal@redhat.com>
Reviewed-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Reviewed-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Signed-off-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
SimplePipelineHandler::match may be called several times for different
pipeline configurations. Not all of these calls must succeed. For
example, for TI AM69 board with a single camera attached, the following
error is reported in the log even when libcamera works fine:
ERROR SimplePipeline simple.cpp:1558 No sensor found
This is because a sensor is found for /dev/media0 but not for
/dev/media1. The error is harmless in such a case and only confuses
users who may think no camera is detected at all. Let's change the
error to info and add the device node to the message to indicate the
error is specific to the given media only. It's up to the callers to
report a fatal error condition if libcamera cannot work due to no
matching pipeline configuration.
Signed-off-by: Milan Zamazal <mzamazal@redhat.com>
Reviewed-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Reviewed-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Signed-off-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
uClibc doesn't provide memfd_create(), which led libcamera to open-code
the call using syscall(). Sprinkling the code with #ifdef's isn't the
most readable option, so improve it by providing a local implementation
of memfd_create(), and call the function unconditionally from
MemFd::create(). This makes the main code path more readable.
Suggested-by: Nicolas Dufresne <nicolas@ndufresne.ca>
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Reviewed-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Tested-by: Milan Zamazal <mzamazal@redhat.com>
Tested-by: Hans de Goede <hdegoede@redhat.com>
Reviewed-by: Hans de Goede <hdegoede@redhat.com>
Reviewed-by: Nicolas Dufresne <nicolas.dufresne@collabora.com>
Unlike in C where they have been standardized since C99, variable-length
arrays in C++ are an extension supported by gcc and clang. Clang started
warning about this with -Wall in version 18:
src/libcamera/ipc_unixsocket.cpp:250:11: error: variable length arrays in C++ are a Clang extension [-Werror,-Wvla-cxx-extension]
250 | char buf[CMSG_SPACE(num * sizeof(uint32_t))];
| ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
One simple option is to disable the warning. However, usage of VLAs in
C++ is discouraged by some, usually due to security reasons, based on
the rationale that developers are often unaware of unintentional use of
VLAs and how they may affect the security of the code when the array
size is not properly validated.
This rationale may sound dubious, as the most commonly proposed fix is
to replace VLAs with vectors (or just arrays dynamically allocated with
new() wrapped in unique pointers), without adding any size validation.
This will not produce much better results. However, keeping the VLA
warning and converting the code to dynamic allocation may still be
slightly better, as it can prompt developers to notice VLAs and check if
size validation is required.
For these reasons, convert all VLAs to std::vector. Most of the VLAs
don't need extra size validation, as the size is bound through different
constraints (e.g. image width for line buffers). An arguable exception
may be the buffers in IPCUnixSocket::sendData() and
IPCUnixSocket::recvData() as the number of fds is not bound-checked
locally, but we will run out of file descriptors before we could
overflow the buffer size calculation.
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Reviewed-by: Milan Zamazal <mzamazal@redhat.com>
Acked-by: Jacopo Mondi <jacopo.mondi@ideasonboard.com>
V4L2VideoDevice is using the caps to determine which kind of buffers to
use with the video-device in 2 different cases:
1. V4L2VideoDevice::open()
2. V4L2VideoDevice::[get|try|set]Format()
And the order in which the caps are checked is different between
these 2 cases. This is a problem for /dev/video# nodes which support
both video-capture and metadata buffers. open() sets bufferType_ to
V4L2_BUF_TYPE_VIDEO_CAPTURE[_MPLANE] in this case, where as
[get|try|set]Format() will call [get|set]FormatMeta() which does not
work with V4L2_BUF_TYPE_VIDEO_CAPTURE[_MPLANE] buffers.
Switch [get|try|set]Format() to use the bufferType_ to determine on what
sort of buffers they should be operating, leaving the V4L2VideoDevice
code with only a single place where the decision is made what sort
of buffers it should operate on for a specific /dev/video# node.
This will also allow to modify open() in the future to take a bufferType
argument to allow overriding the default bufferType it selects for
/dev/video# nodes which are capable of supporting more then 1 buffer type.
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Reviewed-by: Jacopo Mondi <jacopo.mondi@ideasonboard.com>
Reviewed-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
The converter interface uses the unsigned int output stream index to map
to the output frame buffers. This is cumbersome to implement new
converters because one has to keep around additional book keeping
to track the streams with their correct indexes.
The v4l2_converter_m2m and simple pipeline handler are adapted to
use the new interface. This work roped in software ISP as well,
which also seems to use indexes (although it doesn't implement converter
interface) because of a common conversionQueue_ queue used for
converter_ and swIsp_.
The logPrefix is no longer able to generate an index from a stream, and
is updated to be more expressive by reporting the stream configuration
instead, for example, reporting "1920x1080-MJPEG" in place of
"stream0".
Signed-off-by: Umang Jain <umang.jain@ideasonboard.com>
Reviewed-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Reviewed-by: Paul Elder <paul.elder@ideasonboard.com>
Tested-by: Andrei Konovalov <andrey.konovalov.ynk@gmail.com> # sm8250 RB5
Rename the private Stream class from V4L2M2MConverter::Stream to
V4L2M2MConverter::V4L2M2MStream. This is done to improve readability
of the code when we drop the handling of stream by indexes in a
subsequent patch.
Signed-off-by: Umang Jain <umang.jain@ideasonboard.com>
Reviewed-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Reviewed-by: Paul Elder <paul.elder@ideasonboard.com>
Currently the soft-isp outputs a single output stream. Hence,
drop the unnecessary check for stream indexes.
Another reason to drop is actually the stream indexes is meant to be
unique in outputs std::map<>, hence checking for unique stream indexes
is redundant.
Signed-off-by: Umang Jain <umang.jain@ideasonboard.com>
Reviewed-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Reviewed-by: Paul Elder <paul.elder@ideasonboard.com>
The streams sanity check tries to determine if all the stream indexes
passed in outputs std::map<> are unique. However, since the data
container is std::map<>, all its keys (stream indexes in this case),
are already unique.
Instead, rectify the sanity check to ensure all the framebuffers passed
in the outputs std::map<> are unique to each index. Hence, no two stream
indexes should have same framebuffer. Update the comment to reflect
the change.
Signed-off-by: Umang Jain <umang.jain@ideasonboard.com>
Reviewed-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Reviewed-by: Paul Elder <paul.elder@ideasonboard.com>
The YamlObject::get<T>() function template has a specialization for
double but not for float. When used in an IPA module, the issue is
caught at module load time only, when dynamic links are resolved,
causing errors such as
Failed to open IPA module shared object: /usr/lib/libcamera/ipa_rkisp1.so: undefined symbol: _ZNK9libcamera10YamlObject6GetterIfE3getERK_
Fix it by adding a float specialization. The alternative would be to use
double only in IPA modules, but the lack of enforcement at compile time
makes this dangerous.
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Tested-by: Jacopo Mondi <jacopo.mondi@ideasonboard.com>
Reviewed-by: Stefan Klug <stefan.klug@ideasonboard.com>
In order to be more compatible with modern hardware and APIs. This
notably allows GL implementations to directly import the buffers more
often and seems to be required for Wayland.
Further more, as we already enforce a 8 byte stride, these formats work
better for clients that don't support padding - such as libwebrtc at the
time of writing.
Tested devices:
- Librem5
- PinePhone
- Thinkpad X13s
Signed-off-by: Robert Mader <robert.mader@collabora.com>
Tested-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Reviewed-by: Milan Zamazal <mzamazal@redhat.com>
Reviewed-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Signed-off-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
The YamlObject::get() function is a function template that gets fully
specialized for various types. This works fine for non-template types,
but specializing it for template types (e.g. a std::vector<U>) would
require partial template specialization, which C++ allows for classes
and variables but not functions.
To work around this problem, delegate the implementation to a new
YamlObject::Getter structure template, which will support partial
specialization.
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Reviewed-by: Paul Elder <paul.elder@ideasonboard.com>
Reviewed-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
generateConfiguration() called validate() as a final step, causing the
stride and frameSize fields in StreamConfiguration to be filled in based
on the pixel format and width/height.
If a user application did not clear the stride field when setting up a
custom pixel format and width/height, the pipeline handler would respect
this stride and possibly overallocate buffers with a larger stride than
needed.
Fix this by removing the call to validate() completely, leaving the
stride and frameSize fields defaulting to 0. Removal of this call is
inconsequential as we hard-code a valid configuration for Raspberry Pi
platforms in generateConfiguration().
Bug: https://github.com/raspberrypi/libcamera/issues/138
Bug: https://github.com/raspberrypi/libcamera/issues/141
Signed-off-by: Naushir Patuck <naush@raspberrypi.com>
Reviewed-by: David Plowman <david.plowman@raspberrypi.com>
Reviewed-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Signed-off-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
This way the construction of the default value of type `T`
can be delayed until it is really needed, which is useful,
for example when `T == std::string` and the default value comes
from a string literal, as the default value string would always
be constructed otherwise, even if not needed.
Signed-off-by: Barnabás Pőcze <pobrn@protonmail.com>
Reviewed-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Reviewed-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
uClibc doesn't provide the macros defining parameters for the file
sealing API. Define them manually as a work around.
Fixes: ea4baaacc3 ("libcamera: DmaBufAllocator: Support allocating from /dev/udmabuf")
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Reviewed-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Reviewed-by: Hans de Goede <hdegoede@redhat.com>
uClibc doesn't provide a memfd_create() implementation. Fix it by using
a direct syscall when the function isn't available.
Fixes: ea4baaacc3 ("libcamera: DmaBufAllocator: Support allocating from /dev/udmabuf")
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Reviewed-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Reviewed-by: Hans de Goede <hdegoede@redhat.com>
Allow the DmaBufAllocator used by the software ISP to use memfd() +
/dev/udmabuf for the software ISP destination buffers.
This is useful on Linux distributions where normal users are not allowed
to access /dev/dma_heap/* while they are allowed to access /dev/udmabuf.
Reviewed-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Tested-by: Bryan O'Donoghue <bryan.odonoghue@linaro.org> # Lenovo-x13s
Reviewed-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Signed-off-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Users of the DmaHeap class really just want some way to allocate
dma-buffers from userspace. This can also be done by using /dev/udmabuf
instead of using /dev/dma_heap/*.
Rename DmaHeap class to DmaBufAllocator in preparation of adding
/dev/udmabuf support.
And update the DmaHeap class docs to match including replacing references
to "dma-heap type" with "dma-buf provider".
This is a pure automated rename on the code ('s/DmaHeap/DmaBufAllocator/')
+ file renames + doc updates. There are no functional changes.
The DmaBufAllocator objects in vc4.cpp and software_isp.cpp are left named
dmaHeap_ to keep the changes to those 2 files to a minimum.
Reviewed-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Tested-by: Bryan O'Donoghue <bryan.odonoghue@linaro.org> # Lenovo-x13s
Reviewed-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Signed-off-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
The subdev embedded data support series includes a change to the
VIDIOC_SUBDEV_G_ROUTING and VIDIOC_SUBDEV_S_ROUTING ioctls that impacts
the userspace API.
Update to the new API, while preserving backward compatibility to ease
the transition. Document the backward compatibility to only be supported
for two kernel releases. As the routing API isn't enabled in any
upstream kernel yet, users of the API need kernel patches, and are
expected to be able to upgrade quickly.
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Reviewed-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Starting in kernel v5.16, the vimc driver stopped hardcoding the scaler
factor. Use this to lift constraints on the camera configuration, and in
particular on the exotic output size alignment to a multiple of 6. As a
result, vimc-based cameras can more easily match common display
resolutions.
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Reviewed-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Reviewed-by: Daniel Scally <dan.scally@ideasonboard.com>
TODO #4 was recorded at a time where the IPA module computed gain values
and the ISP computed the look up tables. The gains were higher-level
parameters. Now that the look up tables are computed in the IPA module,
the IPA and ISP are more tightly coupled and the TODO item is less
relevant.
Let's drop the TODO item. We may or may not need to switch to a
different representation in future but there is currently no good need
for this and the conversion of the values would be just waste of CPU
cycles.
Signed-off-by: Milan Zamazal <mzamazal@redhat.com>
Reviewed-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Constructing the color mapping tables is related to stats rather than
debayering, where they are applied. Let's move the corresponding code
to stats processing.
The same applies to the auxiliary gamma table. As the gamma value is
currently fixed and used in a single place, with the temporary exception
mentioned below, there is no need to share it anywhere anymore.
It's necessary to initialize SoftwareIsp::debayerParams_ to default
values. These initial values are used for the first two frames, before
they are changed based on determined stats. To avoid sharing the gamma
value constant in artificial ways, we use 0.5 directly in the
initialization. This all is not a particularly elegant thing to do,
such a code belongs conceptually to the similar code in stats
processing, but doing better is left for larger refactoring.
This is a preliminary step towards building this functionality on top of
libipa/algorithm.h, which should follow.
Signed-off-by: Milan Zamazal <mzamazal@redhat.com>
Reviewed-by: Andrei Konovalov <andrey.konovalov.ynk@gmail.com>
Reviewed-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
To match the enumerated media devices, each registered pipeline handler
is used in no specific order. It is a limitation when several pipelines
can match the devices, and user has to select a specific pipeline.
For this purpose, environment variable LIBCAMERA_PIPELINES_MATCH_LIST is
created to give the option to define an ordered list of pipelines to
match on.
LIBCAMERA_PIPELINES_MATCH_LIST="<name1>[,<name2>[,<name3>...]]]"
Example:
LIBCAMERA_PIPELINES_MATCH_LIST="rkisp1,simple"
Signed-off-by: Julien Vuillaumier <julien.vuillaumier@nxp.com>
Reviewed-by: Jacopo Mondi <jacopo.mondi@ideasonboard.com>
Reviewed-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Signed-off-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
