- 19 Jul, 2020 40 commits
-
-
Dafna Hirschfeld authored
Currently the stats are read in a work queue. Defering the reading of the stats is not needed and it is fine to read them inside the irq handler. This patch fixes and remove the TODO item: 'Use threaded interrupt for rkisp1_stats_isr(), remove work queue.' Signed-off-by:
Dafna Hirschfeld <dafna.hirschfeld@collabora.com> Acked-by:
Helen Koike <helen.koike@collabora.com> Signed-off-by:
Hans Verkuil <hverkuil-cisco@xs4all.nl> Signed-off-by:
Mauro Carvalho Chehab <mchehab+huawei@kernel.org>
-
Dafna Hirschfeld authored
This patch removes two locks in the rkisp1_stats struct: - The mutex 'wq_lock' that is used to protect the buffers list 'stat' - The spin_lock 'irq_lock' that is used to protect the variable 'is_streaming' It replaces them with one spin_lock 'lock' that protects both the buffers list and the 'is_streaming' variable. In later patch the reading of the statistics will move to the isr so there will be no need for the mutex 'wq_lock' Signed-off-by:
-
Dafna Hirschfeld authored
The function 'rkisp1_stats_vb2_stop_streaming' runs in user context therefore it is enough to use spin_lock_irq Signed-off-by:
-
Dafna Hirschfeld authored
The mask of the ready statistics flags is used in several places in the code using bitwise-or. Use a macro for that to make the code more readable. Signed-off-by:
Laurent Pinchart <laurent.pinchart@ideasonboard.com> Signed-off-by:
-
Laurent Pinchart authored
Implementation of media controller centric device configuration will add code to the CAMERARX support section, which is already quite big. Move it to a separate file to make the code more manageable. No functional change is included. The cal_camerarx_init_regmap() function is kept in cal.c and renamed to cal_init_camerarx_regmap() as it's not specific to one CAMERARX instance, but related to the whole CAL device. Signed-off-by:
Benoit Parrot <bparrot@ti.com> Signed-off-by:
-
Laurent Pinchart authored
To prepare for the split of the camerarx code to a separate file, move the CAL I/O accessors to cal.h. This requires renaming the accessors with a cal_prefix, as the current names are too generic and prone to namespace clashes. The reg_read() and read_write() macros, that cover both CAL and CAMERARX register access, are split in two groups of inline functions, one for CAL access and one for CAMERARX access. Signed-off-by:
-
Laurent Pinchart authored
To prepare for implementation of media controller centric device configuration, move all the video node handling support to a separate file. Signed-off-by:
-
Laurent Pinchart authored
Specifying 'int' explicitly is generally preferred in the kernel for unsigned int types. Fix the only wrong occurrence. Signed-off-by:
-
Laurent Pinchart authored
The phy->sensor NULL check in cal_camerarx_get_external_rate() is not needed, as the V4L2 video devices are only registered when the sensor is bound. Remove it. Signed-off-by:
-
Laurent Pinchart authored
The external pixel rate is retrieved when starting the camerarx and only used then. There's no need to store it in the cal_camerarx structure, it can be returned by cal_camerarx_get_external_info() and explicitly passed to cal_camerarx_config(). While at it, rename cal_camerarx_get_external_info() to cal_camerarx_get_external_rate() to better reflect the function's purpose. Signed-off-by:
-
Laurent Pinchart authored
Refactor the camerarx start and stop procedure to group all the corresponding operations in two functions, cal_camerarx_start() and cal_camerarx_stop() instead of splitting them in multiple steps called from the vb2 stream start and stop functions. This reduces the coupling between the camerarx and the contexts, and better models the camerarx hardware. The cal_camerarx_start() is a reworked version of cal_camerarx_init() with the description of the start procedure updated to match the AM65x, DRA7[124567]x and DRA80xM TRMs. The cal_camerarx_wait_ready() function is inlined in cal_camerarx_start() to better describe the start procedure. No functional change is included in the camerarx start and stop procedures themselves, but the interleaving of the start steps with the context configuration has been changed in cal_start_streaming(). Signed-off-by:
-
Laurent Pinchart authored
To prepare for the camerarx refactoring, reorder functions without any functional change to ease review of the refactoring itself. Signed-off-by:
-
Laurent Pinchart authored
The cal_camerarx_max_lanes() function is a one-liner that has a single caller. It doesn't improve readability. Inline it in its caller. Signed-off-by:
-
Laurent Pinchart authored
To avoid making the cal_ctx structure layoug depend on the size of the cal_formats array, allocate the active_fmt array dynamically. This prepares for splitting the driver in multiple files. Signed-off-by:
-
Laurent Pinchart authored
To prepare for splitting the V4L2 API support to a separate file, call cal_ctx_v4l2_init_formats() from cal_ctx_v4l2_register(). Signed-off-by:
-
Laurent Pinchart authored
Enable the media controller API by registering a media device and initializing the media entities corresponding to the video devices. The context initialization is slightly refactored as a result. The media graph will be built in a subsequent change. Signed-off-by:
-
Laurent Pinchart authored
The cal_dev structure stores the platform_device pointer, but most accesses to that field need the device pointer. Store the struct device pointer directly to simplify the code, and use to_platform_device() in the two locations that need the platform device. Signed-off-by:
-
Laurent Pinchart authored
Print the hardware revision in the X.Y.R format, which is more readable that the 32-bit hex value. For the hardware info register, only print its value if it doesn't contain what we expect. Signed-off-by:
-
Laurent Pinchart authored
Read the hardware revision and info right after allocating resources, as there's no need to delay doing so until all initialization is complete. Signed-off-by:
-
Laurent Pinchart authored
Create four functions to handle initialization, cleanup, registration and unregistration of the V4L2 (and soon media controller) objects: - init() is meant to be called early at probe time to initialize the objects, before they get used from within the kernel - cleanup() is the counterpart of init, and is meant to be called at the end of the remove sequence to free all objects - register() is meant to be called at the end of the probe sequence, to register the userspace-facing devices - unregister() is the counterpart of register, and is meant to be called at the beginning for the remove sequence, to disallow access from userspace Signed-off-by:
-
Laurent Pinchart authored
Use the CAL_NUM_CSI2_PORTS macro instead of the hardcoded numerical value 2 to iterate over the CSI-2 ports. Signed-off-by:
-
Laurent Pinchart authored
The isvcirqset() isn't used. The isportirqset() doesn't increase readability. Inline the latter and simply drop the former. Signed-off-by:
-
Laurent Pinchart authored
The CAL_NUM_CSI2_PORTS macro isn't a register definition. Move it to cal.c, and fix indentation of the other macros while at it. Signed-off-by:
-
Laurent Pinchart authored
The driver happens the use the same number of CAMERARX and context, but coupling their cleanup at remove time is wrong. To prepare for the introduction of additional contexts, decouple the two. Signed-off-by:
-
Laurent Pinchart authored
The cal_quickdump_regs() dumps registers for the two CAMERARX instances. Retrieve those instances from the cal_dev directly instead of going through the contexts, and simplify the code by using a loop. Signed-off-by:
-
Laurent Pinchart authored
The async notifier is meant to bind to subdevs connected to the CSI-2 ports. Those ports are modelled by the CAMERARX phy instances. To prepare for additional decoupling of contexts and phys, make the notifier operate on phys. We still initialize and register the context V4L2 support in the async notifier complete operation as that's our signal that the userspace API is ready to be exposed. Signed-off-by:
-
Laurent Pinchart authored
There's no need to create one async notifier per CAL context. Merge them all in a single notifier, stored in cal_dev. Signed-off-by:
-
Laurent Pinchart authored
The subdev notifier handling assumes a 1:1 mapping between CAL contexts and notifiers. To prepare for merging the multiple notifiers into a single one, retrieve the CAL context from the async subdev structure instead of from the notifier. Signed-off-by:
-
Laurent Pinchart authored
The initialization of the context based on the connected sensor is split between the async notifier .bound() and .complete() operations. Move it all to a separate function and call it from .bound() operation to prepare for the move of the notifiers from the contexts to the cal_dev. Only V4L2 registration is kept in the .complete() operation. Signed-off-by:
-
Laurent Pinchart authored
Use ARRAY_SIZE(cal->ctx) to iterate over the cal->ctx array instead of using the numerical value from the CAL_NUM_CONTEXT macro (or, worse, hardcoding the value 2). This will allow reworking contexts with less changes in the code. Signed-off-by:
-
Laurent Pinchart authored
The DT port nodes corresponding to the CSI-2 inputs belong to the CAMERARX instances. Move parsing of the DT properties to a new cal_camerarx_parse_dt() function, called by cal_camerarx_create(). Signed-off-by:
-
Laurent Pinchart authored
The cal_camerarx_create() function allocates resources with devm_*, and thus doesn't need any manual cleanup. Those won't hold true for long, as we will need to store resources that have no devm_* allocation variant in cal_camerarx. Furthermore, devm_kzalloc() is the wrong memory allocation API for structures that can be accessed from userspace, as device nodes can be kept open across device removal. Add a cal_camerarx_destroy() function to destroy a CAMERARX instance explicitly, and switch to kzalloc() for memory allocation. Signed-off-by:
-
Laurent Pinchart authored
In cal_remove(), unregister the video devices as the first operation, before cleaning up the V4L2 objects, to avoid use-after-free. This isn't a complete solution yet, as video nodes can be kept open across unregistration. Signed-off-by:
-
Laurent Pinchart authored
Create cal_ctx_v4l2_unregister() and cal_ctx_v4l2_cleanup() functions to unregister and cleanup the V4L2-related objects from the context, and call them in cal_remove() and in the error path of cal_probe(). Signed-off-by:
-
Laurent Pinchart authored
There's no need to delay most of the video device initialization until the sensor subdevs are bound. Split the initialization and registration, and perform the initialization when creating the context. Signed-off-by:
-
Laurent Pinchart authored
The v4l2_device structure is meant to represent the whole device. In the CAL case, this corresponds to the CAL, the CAMERARX instances and the connected sensors. There should thus be a single v4l2_device instance. Replace the per-context instance with a global instance in the cal_dev structure. Don't set the v4l2_device name manually as v4l2_device_register() sets it to a value that is suitable for the driver. Signed-off-by:
-
Laurent Pinchart authored
To prepare for decoupling the v4l2_device from the cal_ctx, don't set the control handler in the v4l2_device and expect the video node to use it automatically, but set the video node control handler directly. This requires adding the sensor subdev controls to the control handler manually, as that operation was performed on the v4l2_device by v4l2_device_register_subdev(). Signed-off-by:
-
Laurent Pinchart authored
Letting the v4l2_fwnode_endpoint_parse() function determine the bus type automatically is deprecated. Set the bus type to DPHY manually as the TI CAL only supports DPHY. Signed-off-by:
Tomi Valkeinen <tomi.valkeinen@ti.com> Reviewed-by:
-
Laurent Pinchart authored
Replace the manual of_get_next_port() and of_get_next_endpoint() implementations and the corresponding logic in the caller with a call to of_graph_get_endpoint_by_regs(). This greatly simplifies OF parsing. Signed-off-by:
-
Laurent Pinchart authored
Using the endpoint reg value to pass the CSI-2 source virtual channel ID in DT is a hack, isn't documented in the DT bindings, and isn't used in neither upstream DT nor TI official overlays. Hardcode the virtual channel to 0 to simplify reworking the code, proper virtual channel support will be implemented later. Signed-off-by:
-
