Age | Commit message (Collapse) | Author |
|
Many signals used in internal and public APIs carry the emitter pointer
as a signal argument. This was done to allow slots connected to multiple
signal instances to differentiate between emitters. While starting from
a good intention of facilitating the implementation of slots, it turned
out to be a bad API design as the signal isn't meant to know what it
will be connected to, and thus shouldn't carry parameters that are
solely meant to support a use case specific to the connected slot.
These pointers turn out to be unused in all slots but one. In the only
case where it is needed, it can be obtained by wrapping the slot in a
lambda function when connecting the signal. Do so, and drop the emitter
pointer from all signals.
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Reviewed-by: Umang Jain <umang.jain@ideasonboard.com>
|
|
In many cases, the emitter object passed as a pointer from signals to
slots is also available as a class member. Use the class member when
this occurs, to prepare for removal of the emitter object pointer from
signals.
In test/event.cpp, this additionally requires moving the EventNotifier
to a class member.
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Reviewed-by: Umang Jain <umang.jain@ideasonboard.com>
|
|
It can be useful to connect a signal to a functor, and in particular a
lambda function, while still operating in the context of a receiver
object (to support both object-based disconnection and queued
connections to Object instances).
Add a BoundMethodFunctor class to bind a functor, and a corresponding
Signal::connect() function. There is no corresponding disconnect()
function, as a lambda passed to connect() can't be later passed to
disconnect(). Disconnection typically uses disconnect(T *object), which
will cover the vast majority of use cases.
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Reviewed-by: Umang Jain <umang.jain@ideasonboard.com>
|
|
When disconnecting a signal from a receiver, it is usually not necessary
to specify the receiver's slot function explicitly, as the signal is
often connected to a single slot for a given receiver. We can thus use a
simpler version of Signal::disconnect() that takes a pointer to the
receiver object only. This reduces code size, as the disconnect()
function is a template function.
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Reviewed-by: Umang Jain <umang.jain@ideasonboard.com>
|
|
libcamera isn't supposed to log messages after the logger is destroyed,
as the global logger instance is destroyed after the main() function
returns, and the camera manager is supposed to have been stopped and
destroyed before that.
This rule is difficult to enforce in the V4L2 compat implementation, as
there is no location where we can destroy the camera manager manually
before the logger is destroyed. This results in a use-after-free
condition when the camera manager gets stopped during destruction.
Fix it by not trying to print log messages when the global logger
instance has been destroyed.
This is a bit of a hack, but hopefully not too bad. There could be race
conditions when using a CameraManager instance that is destroyed as part
of the destruction of global variables (like the V4L2 compat layer does,
it wraps CameraManager in a singleton V4L2CompatManager class, and
destroys it when V4L2CompatManager is destroyed) as the CameraManager
thread will still be running when the logger gets destroyed, but this
doesn't cause any regression as we destroy the logger without any
safeguard measure today anyway.
There are other options that could be considered. Forcing destruction of
the logger after the camera manager in the V4L2 compat layer is one of
them, but turned out to be difficult. For instance care would need to be
taken *not* to log any message in the mmap() wrapper if the fd doesn't
match a wrapped camera, as mmap() is called very early in the
initialization process, before libcamera and the logger get initialized.
The resulting implementation would likely be fairly complex.
Another option could be to wrap the logger with a shared pointer, and
keep a reference to it in CameraManager. That's more intrusive, and it's
not clear if it would be worth it.
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Reviewed-by: Paul Elder <paul.elder@ideasonboard.com>
|
|
The SimplePipelineHandler activeCamera_ member pointer is set but never
used. Drop it.
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Tested-by: Martin Kepplinger <martin.kepplinger@puri.sm>
|
|
To use multiple cameras at the same time, a per-camera buffer ready
handler is needed. Move the bufferReady() function connected to the
V4L2VideoDevice bufferReady signal from the SimplePipelineHandler class
to the SimpleCameraData class.
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Tested-by: Martin Kepplinger <martin.kepplinger@puri.sm>
|
|
To use multiple cameras at the same time, each camera instance will need
its own converter. Store the converter in SimpleCameraData, and open it
in init().
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Tested-by: Martin Kepplinger <martin.kepplinger@puri.sm>
|
|
The cameras created by the same pipeline handler instance may share
hardware resources, prohibiting usage of multiple cameras concurrently.
Implement a heuristic to reserve resources and handle mutual exclusiong
in a generic way.
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Tested-by: Martin Kepplinger <martin.kepplinger@puri.sm>
|
|
Move opening of video devices at match() time, the same way as subdevs
are opened, to make the handling of V4L2 video devices and subdevices
more consistent.
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Tested-by: Martin Kepplinger <martin.kepplinger@puri.sm>
|
|
Merge the SimplePipelineHandler videos_ and subdevs_ maps, which
respectively store V4L2 video devices and subdevices associated with
entities, into a single entities_ map that contains an EntityData
structure. This gathers all data about entities in a single place,
allowing for easy extension of entity data in the future.
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Tested-by: Martin Kepplinger <martin.kepplinger@puri.sm>
|
|
Store the entity corresponding to the video node at the end of the
pipeline in the SimpleCameraData::entities_ list. This requires special
handling of the video node in the loops that iterate over all entities,
but will be useful to implement mutually exclusive access to entities
for concurrent camera usage.
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Tested-by: Martin Kepplinger <martin.kepplinger@puri.sm>
|
|
The video device is currently opened in the SimpleCameraData
constructor. This requires opening the video devices on-demand the first
time they're accessed, which gets in the way of refactoring of
per-entity data storage in the pipeline handler.
Move opening of the video device to the SimpleCameraData::init()
function. The on-demand open mechanism isn't touched yet, it will be
refactored later.
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Tested-by: Martin Kepplinger <martin.kepplinger@puri.sm>
|
|
Record the sink and source pads through which an entity is traversed in
the list of entities stored in the camera data. This prepares for
implementing mutually exclusive access to entities between cameras.
The debug message that displays detected pipelines now prints pads to
describe the pipeline more precisely:
[0:00:35.901275750] [260] DEBUG SimplePipeline simple.cpp:404 Found pipeline: [imx290 2-001a|0] -> [0|csis-32e40000.csi|1] -> [0|mxc_isi.0|1] -> [0|mxc_isi.0.capture]
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Reviewed-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Tested-by: Martin Kepplinger <martin.kepplinger@puri.sm>
|
|
Add a new field to the MediaEntity class to identify the type of
interface it exposes to userspace. The MediaEntity constructor is
changed to take a media_v2_interface pointer instead of just the device
node major and minor to have access to the interface type.
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Reviewed-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Tested-by: Martin Kepplinger <martin.kepplinger@puri.sm>
|
|
MappedFrameBuffer::maps() returns planes_. This renames the function
name to planes().
Signed-off-by: Hirokazu Honda <hiroh@chromium.org>
Reviewed-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Reviewed-by: Jacopo Mondi <jacopo@jmondi.org>
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
|
|
Do not compare higher precision of the ratios, as it might lead to
absurd selection of sensor size for a relatively low requested
resolution size.
For example:
The imx258 driver supports the following sensor resolutions:
- 4208x3118 = 1.349583066
- 2104x1560 = 1.348717949
- 1048x780 = 1.343589744
It can be inferred that, that the aspect ratio only differs by a small
mantissa with each other. It does not makes sense to select a 4208x3118
for a requested size of say 640x480 or 1280x720, which is what is
happening currently.
($) cam -c1 -swidth=640,height=480,role=raw
- CIO2 configuration: 4208x3118-SGRBG10_IPU3 [*]
In order to address this constraint, only compare the ratio with single
precision to make a better decision on the sensor resolution policy
selection.
($) cam -c1 -srole=raw,width=640,height=480
- CIO2 configuration: 1048x780-SGRBG10_IPU3 [*]
[*] Please revert 0536a9aa7189("ipu3: Disallow raw only camera configuration")
if the configuration is reported as invalid.
Signed-off-by: Umang Jain <umang.jain@ideasonboard.com>
Tested-by: Jacopo Mondi <jacopo@jmondi.org>
Reviewed-by: Jacopo Mondi <jacopo@jmondi.org>
Reviewed-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
|
|
The current implementation of getSensorFormat() prioritizes sensor
sizes that match the output size FOV ratio.
Modify the frame size selection procedure to prioritize resolutions
with the same FOV as the sensor's native one, to avoid cropping in the
sensor pixel array.
For example:
- on a Soraka device equipped with ov13858 as back sensor, with a
native resolution of 4224x3136 (4:3), when requested to select the
sensor output size to produce 1080p (16:9) a frame size of 2112x1188
(16:9) is selected causing the ImgU configuration procedure to fail.
If a resolution with the same FOV as the sensor's native size, such
as 2112x1568 (4:3), is selected the pipeline works correctly.
Suggested-by:: Jacopo Mondi <jacopo@jmondi.org>
Signed-off-by: Umang Jain <umang.jain@ideasonboard.com>
Reviewed-by: Jacopo Mondi <jacopo@jmondi.org>
Tested-by: Jacopo Mondi <jacopo@jmondi.org>
Reviewed-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
|
|
This prepares a base to introduce custom selection of sensor format
based on platform(Soraka or Nautilus) constraints. The changes in
selection policy will be introduced in a subsequent patch.
The function is copied from CameraSensor::getFormat() in the IPU3
pipeline handler code to be later changed on top.
The copy is not verbatim and has a minor change as follows:
CameraSensor::getFormats() has access to a V4L2Subdevice::Formats
internally and use it directly to iterate over supported camera sensor
frame sizes. The copy is adapted to use the CameraSensor::sizes(mbusCode)
instead, to enumerate over the supported frame sizes as per
mbusCodesToPixelFormat map.
No functional changes in this patch.
Signed-off-by: Umang Jain <umang.jain@ideasonboard.com>
Reviewed-by: Jacopo Mondi <jacopo@jmondi.org>
Tested-by: Jacopo Mondi <jacopo@jmondi.org>
Reviewed-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
|
|
In CameraSensor, the mbusCodes() and sizes() accessor functions
retrieves all the supported media bus codes and the supported sizes
respectively. However, this is quite limiting since the caller
probably isn't in a position to match which range of sizes are
supported for a particular mbusCode.
Hence, the caller is most likely interested to know about the sizes
supported for a particular media bus code. This patch transforms the
existing CameraSensor::sizes() to CameraSensor::sizes(mbuscode) to
achieve that goal.
The patch also transforms existing CIO2Device::sizes() in IPU3 pipeline
handler to CIO2Device::sizes(PixelFormat) on a similar principle. The
function is then plumbed to CameraSensor::sizes(mbusCode) to enumerate
the per-format sizes as required in
PipelineHandlerIPU3::generateConfiguration().
Signed-off-by: Umang Jain <umang.jain@ideasonboard.com>
Tested-by: Jacopo Mondi <jacopo@jmondi.org>
Reviewed-by: Jacopo Mondi <jacopo@jmondi.org>
Reviewed-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
|
|
The offset variable is introduced to FrameBuffer::Plane. In order to
detect that the plane is used while the offset is not set, this adds
the assertion to FrameBuffer::planes(). It should be removed in the
future.
Signed-off-by: Hirokazu Honda <hiroh@chromium.org>
Reviewed-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
|
|
V4L2VideDevice::createBuffer() creates the same number of
FrameBuffer::Planes as V4L2 format planes. Therefore, if the v4l2 format
single is single-planar format, the created number of
FrameBuffer::Planes is 1. It should rather create the same number of
FrameBuffer::Planes as the color format planes.
Signed-off-by: Hirokazu Honda <hiroh@chromium.org>
Reviewed-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
|
|
MappedBuffer::maps()
MappedBuffer::maps() returns std::vector<MappedBuffer::Plane>.
Plane has the address, but the address points the beginning of the
buffer containing the plane.
This makes the Plane point the beginning of the plane. So
MappedBuffer::maps()[i].data() returns the address of i-th plane.
Signed-off-by: Hirokazu Honda <hiroh@chromium.org>
Reviewed-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
|
|
This adds offset to FrameBuffer::Plane. It enables representing frame
buffers that store planes in the same dmabuf at different offsets, as
for instance required by the V4L2 NV12 pixel format.
Signed-off-by: Hirokazu Honda <hiroh@chromium.org>
Reviewed-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
|
|
The documentation suggests to use CameraConfiguration::operator[] to
access the StreamConfiguration it contains, but as CameraConfiguration
instances are generated by the Camera class and are returned wrapped in
a unique_ptr<>, the usage of operator[] would require an awkward syntax such
as (*config)[i].
Better to suggest the usage of the CameraConfiguration::at() function
instead to access the StreamConfigurations.
Signed-off-by: Jacopo Mondi <jacopo@jmondi.org>
Reviewed-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
|
|
IPCMessage::payload() converts the IPCMessage into an IPCUnixSocket
payload. However, if IPCMessage is constructed with one of the
following constructors -
IPCMessage::IPCMessage(),
IPCMessage::IPCMessage(uint32_t cmd)
IPCMessage::IPCMessage(const Header &header)
The data_ vector of IPCMessage is empty and uninitialised. In that
case, IPCMessage::payload will try to memcpy() an empty data_ vector
which can lead to invoking memcpy() with a nullptr parameter, which
is then identified by the address sanity checker.. Add a non-empty
data_ vector check to avoid it.
The issue is noticed by running a test manually, testing the vimc
IPA code paths in isolated mode. It is only noticed when the test
is compiled with -Db_sanitize=address,undefined meson built-in option.
ipc_pipe.cpp:110:8: runtime error: null pointer passed as argument 2, which is declared to never be null
Signed-off-by: Umang Jain <umang.jain@ideasonboard.com>
Reviewed-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Reviewed-by: Paul Elder <paul.elder@ideasonboard.com>
Reviewed-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
|
|
In IPCUnixSocket, a payload can be sent/received with empty fd vector,
which leads to passing a nullptr in memcpy() in both sendData()
and recvData(). Add a null check for fd vector's data pointer
to avoid invoking memcpy() with nullptr.
The issue is noticed by running a test manually testing the vimc
IPA code paths in isolated mode. It is only noticed when the test
is compiled with -Db_sanitize=address,undefined meson built-in option.
ipc_unixsocket.cpp:268:8: runtime error: null pointer passed as argument 2, which is declared to never be null
ipc_unixsocket.cpp:312:8: runtime error: null pointer passed as argument 1, which is declared to never be null
Signed-off-by: Umang Jain <umang.jain@ideasonboard.com>
Reviewed-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Reviewed-by: Paul Elder <paul.elder@ideasonboard.com>
|
|
Regarding (de)serialization in isolated IPA calls, we have four layers:
- struct
- byte vector + fd vector
- IPCMessage
- IPC payload
The proxy handles the upper three layers (with help from the
IPADataSerializer), and passes an IPCMessage to the IPC mechanism
(implemented as an IPCPipe), which sends an IPC payload to its worker
counterpart.
When a FileDescriptor is involved, previously it was only a
FileDescriptor in the first layer; in the lower three it was an int. To
reduce the risk of potential fd leaks in the future, keep the
FileDescriptor as-is throughout the upper three layers. Only the IPC
mechanism will deal with ints, if it so wishes, when it does the actual
IPC. IPCPipeUnixSocket does deal with ints for sending fds, so the
conversion between IPCMessage and IPCUnixSocket::Payload converts
between FileDescriptor and int.
Additionally, change the data portion of the serialized form of
FileDescriptor to a 32-bit unsigned integer, for alightnment purposes
and in preparation for conversion to an index into the fd array.
Also update the deserializer of FrameBuffer::Plane accordingly.
Signed-off-by: Paul Elder <paul.elder@ideasonboard.com>
Tested-by: Umang Jain <umang.jain@ideasonboard.com>
Reviewed-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Reviewed-by: Umang Jain <umang.jain@ideasonboard.com>
|
|
There is no point in recording sensor's timestamp when V4L2VideoDevice
has marked the frame buffers with FrameMetadata::FrameCancelled
(happens when the streams are stopped). The metadata is mostly invalid
for cancelled buffers hence, setting timestamp on invalid metadata
does not make sense (however some metadata can be considered valid
for e.g. returning cause of failure through metadata on cancelled
buffers).
Signed-off-by: Umang Jain <umang.jain@ideasonboard.com>
Reviewed-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Reviewed-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
|
|
Replace manual static casts from the PipelineHandler pointer to a
derived class pointer with helper functions in the camera data classes.
This simplifies code accessing the pipeline from the camera data.
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Reviewed-by: Jacopo Mondi <jacopo@jmondi.org>
Reviewed-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
|
|
The PipelineHandler controls() and properties() functions are only used
by the Camera class. Now that the controls and properties are stored in
the Camera::Private class, we can drop those functions and access the
private data directly in Camera::controls() and Camera::properties().
Suggested-by: Jacopo Mondi <jacopo@jmondi.org>
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Reviewed-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Reviewed-by: Jacopo Mondi <jacopo@jmondi.org>
|
|
The CameraData class isn't used anymore. Drop it.
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Reviewed-by: Niklas Söderlund <niklas.soderlund@ragnatech.se>
Reviewed-by: Jacopo Mondi <jacopo@jmondi.org>
|
|
As part of the effort to remove the CameraData class, migrate the
pipeline handler-specific camera data from CameraData to the
Camera::Private class.
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Reviewed-by: Niklas Söderlund <niklas.soderlund@ragnatech.se>
Reviewed-by: Jacopo Mondi <jacopo@jmondi.org>
|
|
As part of the effort to remove the CameraData class, migrate the
pipeline handler-specific camera data from CameraData to the
Camera::Private class.
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Reviewed-by: Niklas Söderlund <niklas.soderlund@ragnatech.se>
Reviewed-by: Jacopo Mondi <jacopo@jmondi.org>
|
|
As part of the effort to remove the CameraData class, migrate the
pipeline handler-specific camera data from CameraData to the
Camera::Private class.
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Reviewed-by: Niklas Söderlund <niklas.soderlund@ragnatech.se>
Reviewed-by: Jacopo Mondi <jacopo@jmondi.org>
|
|
As part of the effort to remove the CameraData class, migrate the
pipeline handler-specific camera data from CameraData to the
Camera::Private class.
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Reviewed-by: Niklas Söderlund <niklas.soderlund@ragnatech.se>
Reviewed-by: Jacopo Mondi <jacopo@jmondi.org>
|
|
As part of the effort to remove the CameraData class, migrate the
pipeline handler-specific camera data from CameraData to the
Camera::Private class.
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Reviewed-by: Niklas Söderlund <niklas.soderlund@ragnatech.se>
Reviewed-by: Jacopo Mondi <jacopo@jmondi.org>
|
|
As part of the effort to remove the CameraData class, migrate the
pipeline handler-specific camera data from CameraData to the
Camera::Private class.
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Reviewed-by: Jacopo Mondi <jacopo@jmondi.org>
Reviewed-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
|
|
Improve the Camera::Private documentation by fixing minor spelling or
style issues.
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Reviewed-by: Jacopo Mondi <jacopo@jmondi.org>
Reviewed-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
|
|
With pipeline handlers now being able to subclass Camera::Private, start
the migration from CameraData to Camera::Private by moving the members
of the base CameraData class. The controlInfo_, properties_ and pipe_
members are duplicated for now, to allow migrating pipeline handlers one
by one.
The Camera::Private class is now properly documented, don't exclude it
from documentation generation.
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Reviewed-by: Niklas Söderlund <niklas.soderlund@ragnatech.se>
Reviewed-by: Jacopo Mondi <jacopo@jmondi.org>
|
|
In order to allow subclassing Camera::Private in pipeline handlers, pass
the pointer to the private data to the Camera constructor, and to the
Camera::createCamera() function.
The Camera::Private id_ and streams_ members now need to be initialized
by the Camera constructor instead of the Camera::Private constructor, to
allow storage of the streams in a pipeline handler-specific subclass of
Camera::Private.
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Reviewed-by: Niklas Söderlund <niklas.soderlund@ragnatech.se>
Reviewed-by: Jacopo Mondi <jacopo@jmondi.org>
|
|
The Extensible constructor takes a pointer to a Private instance, whose
lifetime it then manages. Make this explicit in the API by passing the
pointer as a std::unique_ptr<Private>.
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Reviewed-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Reviewed-by: Jacopo Mondi <jacopo@jmondi.org>
|
|
When the stream is stopped, the V4L2VideoDevice sends back all
the queued buffers with FrameMetadata::FrameCancelled status.
It is the responsibility of the pipeline handler to handle
these buffers with FrameMetadata::FrameCancelled. VIMC is
currently missing this handling path.
As the FrameMetadata::FrameCancelled is set when the stream is
stopped, we can be sure that no more queued and re-use of request
shall happen. Hence, cancel all the requests' buffers force a
complete with completeBuffer().
The issue is caught by the gstreamer_single_stream_test.cpp running
with vimc. During the check with meson built-in option
'-Db_sanitize=address,undefined'
it was observed:
==118003==ERROR: AddressSanitizer: heap-use-after-free on address 0x60e000037108 at pc 0x7f225160c9ac bp 0x7f224a47b620 sp 0x7f224a47b618
READ of size 4 at 0x60e000037108 thread T1
#0 0x7f225160c9ab in libcamera::Request::sequence() const ../include/libcamera/request.h:55
#1 0x7f22518297aa in libcamera::VimcCameraData::bufferReady(libcamera::FrameBuffer*) ../src/libcamera/pipeline/vimc/vimc.cpp:577
#2 0x7f225183b1ef in libcamera::BoundMethodMember<libcamera::VimcCameraData, void, libcamera::FrameBuffer*>::activate(libcamera::FrameBuffer*, bool) ../include/libcamera/base/bound_method.h:194
#3 0x7f22515cc91f in libcamera::Signal<libcamera::FrameBuffer*>::emit(libcamera::FrameBuffer*) ../include/libcamera/base/signal.h:126
#4 0x7f22515c3305 in libcamera::V4L2VideoDevice::streamOff() ../src/libcamera/v4l2_videodevice.cpp:1605
#5 0x7f225181f345 in libcamera::PipelineHandlerVimc::stop(libcamera::Camera*) ../src/libcamera/pipeline/vimc/vimc.cpp:365
The VimcCameraData::bufferReady seems to emit even after the stream
is stopped. It's primarily due to vimc's lack of handling
FrameMetadata::FrameCancelled in its pipeline handler.
Signed-off-by: Umang Jain <umang.jain@ideasonboard.com>
Reviewed-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Tested-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Reviewed-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Tested-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
|
|
Plumb through VIMC mojo interface to enable buffers passing.
VIMC does not have parameters or statistics buffers but we can
mimick the typical case of passing IPA buffers from pipeline
handler to IPA using mock buffers. The mock IPA buffers are
FrameBuffers which are dmabuf backed (in other words, mmap()able
through MappedFramebuffer inside the IPA).
This commits shows:
- Passing the parameter buffer from the pipeline handler to
the IPA through functions defined in mojom interface.
- Passing request controls ControlList to the IPA.
Any tests using VIMC will now loop in the IPA paths. Any tests running
in isolated mode will help us to test IPA IPC code paths especially
around (de)serialization of data passing from pipeline handlers to the
IPA. Future IPA interface tests can simply extend the vimc mojom
interface to achieve/test a specific use case as required.
Signed-off-by: Umang Jain <umang.jain@ideasonboard.com>
Reviewed-by: Paul Elder <paul.elder@ideasonboard.com>
Reviewed-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
|
|
VIMC pipeline handler has dmabuf-backed mock FrameBuffers which are
specifically targetted mimicking IPA buffers (parameter and statistics).
Map these mock buffers to the VIMC IPA that would enable exercising IPA
IPC code paths. This will provide leverage to our test suite to test
IPA IPC code paths, which are common to various platforms.
Signed-off-by: Umang Jain <umang.jain@ideasonboard.com>
Reviewed-by: Paul Elder <paul.elder@ideasonboard.com>
Reviewed-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
|
|
VIMC is a virtual test driver that doesn't have statistics or
parameters buffers that are typically passed from a pipeline
handler to its platform IPA. To increase the test coverage going
forward, we can at least mimick the typical interaction of how
a pipeline handler and IPA interact, and use it to increase the
test coverage.
Hence, create simple (single plane) dmabuf-backed FrameBuffers,
which can act as mock IPA buffers and can be memory mapped (mmap)
to VIMC IPA. To create these buffers, temporarily hijack the output
video node and configure it with a V4L2DeviceFormat. Buffers then
can be exported from the output video node using
V4L2VideoDevice::exportBuffers(). These buffers will be mimicked as
IPA buffers in subsequent commits.
Signed-off-by: Umang Jain <umang.jain@ideasonboard.com>
Reviewed-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Reviewed-by: Paul Elder <paul.elder@ideasonboard.com>
|
|
As part of an effort to make the vimc IPA usable for testing, extend it
with a configure function. The configuration is currently ignored by the
IPA.
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Reviewed-by: Paul Elder <paul.elder@ideasonboard.com>
Reviewed-by: Umang Jain <umang.jain@ideasonboard.com>
Signed-off-by: Umang Jain <umang.jain@ideasonboard.com>
|
|
All the IPU3 Camera controls are currently initialized by the pipeline
handler which initializes them using the camera sensor configuration and
platform specific requirements.
However, some controls are better initialized by the IPA, which might,
in example, cap the exposure times and frame duration to the constraints
of its algorithms implementation.
Also, moving forward, the IPA should register controls to report its
capabilities, in example the ability to enable/disable 3A algorithms on
request.
Move the existing controls initialization to the IPA, by providing
the sensor configuration and its controls to the IPU3IPA::init()
function, which initializes controls and returns them to the pipeline
through an output parameter.
The existing controls initialization has been copied verbatim from the
pipeline handler to the IPA, if not a for few line breaks adjustments
and the resulting Camera controls values are not changed.
Signed-off-by: Jacopo Mondi <jacopo@jmondi.org>
Reviewed-by: Paul Elder <paul.elder@ideasonboard.com>
Reviewed-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Reviewed-by: Umang Jain <umang.jain@ideasonboard.com>
|
|
Introduce a new field in the controls serialization protocol to
allow discerning which ControlIdMap a ControlInfoMap refers to.
The newly introduced IdMapType enumeration describes the possible
info maps:
- Either the globally available controls::controls and
properties::properties maps, which are valid across IPC boundaries
- A ControlIdMap created locally by the V4L2 device, which is not valid
across the IPC boundaries
At de-serialization time the idMapType field is inspected and
- If the idmap is a globally defined one, there's no need to create
new ControlId instances when populating the de-serialized
ControlInfoMap. Use the globally available map to retrieve the
ControlId reference and use it.
- If the idmap is a map only available locally, create a new ControlId
as it used to happen before this patch.
As a direct consequence, this change allows us to perform lookup by
ControlId reference on de-serialized ControlIdMap that refers to the
libcamera defined controls::controls and properties::properties.
Signed-off-by: Jacopo Mondi <jacopo@jmondi.org>
Reviewed-by: Paul Elder <paul.elder@ideasonboard.com>
Reviewed-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
|
|
ControlInfoMap does not have a ControlId map associated, but rather
creates one with the generateIdMap() function at creation time.
As a consequence, when in the need to de-serialize a ControlInfoMap all
the ControlId it contains are created by the deserializer instance, not
being able to discern if the controls the ControlIdMap refers to are the
global libcamera controls (and properties) or instances local to the
V4L2 device that has first initialized the controls.
As a consequence the ControlId stored in a de-serialized map will always
be newly created entities, preventing lookup by ControlId reference on a
de-serialized ControlInfoMap.
In order to make it possible to use globally available ControlId
instances whenever possible, create ControlInfoMap with a reference to
an externally allocated ControlIdMap instead of generating one
internally.
As a consequence the class constructors take and additional argument,
which might be not pleasant to type in, but enforces the concepts that
ControlInfoMap should be created with controls part of the same id map.
As the ControlIdMap the ControlInfoMap refers to needs to be allocated
externally:
- Use the globally available controls::controls (or
properties::properties) id map when referring to libcamera controls
- The V4L2 device that creates ControlInfoMap by parsing the device's
controls has to allocate a ControlIdMap
- The ControlSerializer that de-serializes a ControlInfoMap has to
create and store the ControlIdMap the de-serialized info map refers to
Signed-off-by: Jacopo Mondi <jacopo@jmondi.org>
Reviewed-by: Paul Elder <paul.elder@ideasonboard.com>
Reviewed-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
|