Hi all, This is a RFC for the Analog Devices ADV7482 driver. It is based on top of the media_tree and depends on the series '[PATCH 0/2] move s_stream from v4l2_subdev_video_ops to move s_stream from v4l2_subdev_pad_ops'. It's tested on a Renesas Salvator-X board. The driver is not ready for upstream and contains a few hard-coded restrictions and hacks. The restrictions are in large part deepening on from what I see as limitations to the v4l2 framework. The ADV7482 supports two video pipelines which can run (almost) independent of each other, each pipeline terminates in a CSI-2 output TXA and TXB. TXA is a 4-lane CSI-2 output which can be connected to ether a HDMI, CVBS or TTL input. TXB is a 1-lane CSI-2 output which can be connected to the CVBS input. To make things even more complex the datasheet states that the TTL input can also be an output to which HDMI and CVBS inputs can be routed. This RFC hard codes TXA to the HDMI input and TXB to the CVBS input since that is the only configuration I can test. It also do not consider the TTL port at all other then reserving a sink pad id for it. I have modelled the driver with two source pads, one for TXA and one for TXB since I can't see any other way. And in this design I find limitations to the v4l2 framework, there are operations in struct v4l2_subdev_video_ops which now needs to know which source pad it should be acting on. I have tried to solve my immediate problem of the need for a pad aware s_stream in a separate series mention above but there are more operations that needs pad information. I'm looking forward to trying to find solutions in v4l2 to thees problems but in this RFC there is a compile time define which decides which pad is the one used in such operations. Also some operations in struct v4l2_subdev_video_ops are context aware, for example g_std is only valid for TXB since it is connected to the CVBS input. But ofc one can call g_std from a driver hooked up to TXA (which is the HDMI) and would then get the standard for the CVBS. This results in lots of errors in v4l-compliance test. Things that I see needs to be figured out how to be pad aware are or extended in some other way are. - g_std - s_std - querystd - g_tvnorms - g_input_status - cropcap (needed for pixelaspect ratio) - s_dv_timings - g_dv_timings - query_dv_timings - Controlls, TXA and TXB both have a separate set of controls Further more there are some hard-coded logic regarding which sink is connected to which source, this should be moved to s_routing. And default-input DT parameters should be added and acted upon. I did not want to spend time on this until I figured out how to best deal with the more pad aware operations. Other limitations of the RFC that should be solved are: - ADV7482 allows for most of its I2C slave addresses to be user selectable. RFC uses hard coded values, thees should be moved to DT. - EDID, get_edid and set_edid operations need to be added. - Input format change interrupts like adv7180 have. Now one must call querystd or query_dv_timings for the driver to notice that the format have changed. Given all of the above issues the RFC do work and can deliver both HDMI and CVBS video simultaneously. It properly detects what source is connected to it and the two pipes can be independently started and stopped. If anyone is interested to test on Salvator-X the following branch contains all the patches to grab video (rcar-vin for Gen3, rcar-cis2 and adv7482). https://git.ragnatech.se/linux rcar-vin-gen3 Laurent Pinchart (1): media: entity: Add has_route entity operation Niklas Söderlund (1): [media] i2c: adv7482: add adv7482 driver .../devicetree/bindings/media/i2c/adv7482.txt | 62 + drivers/media/i2c/Kconfig | 10 + drivers/media/i2c/Makefile | 1 + drivers/media/i2c/adv7482.c | 1388 ++++++++++++++++++++ include/media/media-entity.h | 5 + 5 files changed, 1466 insertions(+) create mode 100644 Documentation/devicetree/bindings/media/i2c/adv7482.txt create mode 100644 drivers/media/i2c/adv7482.c -- 2.8.3
