From 45267fed3e55845c5b4b279162b273040ae4f587 Mon Sep 17 00:00:00 2001 From: Steve Longerbeam Date: Thu, 14 Dec 2017 20:04:47 -0500 Subject: [PATCH] media: staging/imx: update TODO Update TODO file: - Remove TODO info about the OV564x driver, while this still needs to be done (add a OV5642 driver or merge with OV5640 driver), it is not relevant here. - Update TODO about methods for retrieving CSI bus config. - Add some TODO's about OF graph parsing restrictions. Signed-off-by: Steve Longerbeam Signed-off-by: Hans Verkuil Signed-off-by: Mauro Carvalho Chehab --- drivers/staging/media/imx/TODO | 63 ++++++++++++++++++++++++++++++++++-------- 1 file changed, 51 insertions(+), 12 deletions(-) diff --git a/drivers/staging/media/imx/TODO b/drivers/staging/media/imx/TODO index 0bee313..9eb7326 100644 --- a/drivers/staging/media/imx/TODO +++ b/drivers/staging/media/imx/TODO @@ -1,19 +1,14 @@ -- Clean up and move the ov5642 subdev driver to drivers/media/i2c, or - merge support for OV5642 into drivers/media/i2c/ov5640.c, and create - the binding docs for it. - - The Frame Interval Monitor could be exported to v4l2-core for general use. -- At driver load time, the device-tree node that is the original source - (the "sensor"), is parsed to record its media bus configuration, and - this info is required in imx-media-csi.c to setup the CSI. - Laurent Pinchart argues that instead the CSI subdev should call its - neighbor's g_mbus_config op (which should be propagated if necessary) - to get this info. However Hans Verkuil is planning to remove the - g_mbus_config op. For now this driver uses the parsed DT mbus config - method until this issue is resolved. +- The CSI subdevice parses its nearest upstream neighbor's device-tree + bus config in order to setup the CSI. Laurent Pinchart argues that + instead the CSI subdev should call its neighbor's g_mbus_config op + (which should be propagated if necessary) to get this info. However + Hans Verkuil is planning to remove the g_mbus_config op. For now this + driver uses the parsed DT bus config method until this issue is + resolved. - This media driver supports inheriting V4L2 controls to the video capture devices, from the subdevices in the capture device's @@ -21,3 +16,47 @@ link_notify callback when the pipeline is modified. It should be decided whether this feature is useful enough to make it generally available by exporting to v4l2-core. + +- The OF graph is walked at probe time to form the list of fwnodes to + be passed to v4l2_async_notifier_register(), starting from the IPU + CSI ports. And after all async subdevices have been bound, + v4l2_fwnode_parse_link() is used to form the media links between + the entities discovered by walking the OF graph. + + While this approach allows support for arbitrary OF graphs, there + are some assumptions for this to work: + + 1. All port parent nodes reachable in the graph from the IPU CSI + ports bind to V4L2 async subdevice drivers. + + If a device has mixed-use ports such as video plus audio, the + endpoints from the audio ports are followed to devices that must + bind to V4L2 subdevice drivers, and not for example, to an ALSA + driver or a non-V4L2 media driver. If the device were bound to + such a driver, imx-media would never get an async completion + notification because the device fwnode was added to the async + list, but the driver does not interface with the V4L2 async + framework. + + 2. Every port reachable in the graph is treated as a media pad, + owned by the V4L2 subdevice that is bound to the port's parent. + + This presents problems for devices that don't make this port = pad + assumption. Examples are SMIAPP compatible cameras which define only + a single output port node, but which define multiple pads owned + by multiple subdevices (pixel-array, binner, scaler). Or video + decoders (entity function MEDIA_ENT_F_ATV_DECODER), which also define + only a single output port node, but define multiple pads for video, + VBI, and audio out. + + A workaround at present is to set the port reg properties to + correspond to the media pad index that the port represents. A + possible long-term solution is to implement a subdev API that + maps a port id to a media pad index. + + 3. Every endpoint of a port reachable in the graph is treated as + a media link, between V4L2 subdevices that are bound to the + port parents of the local and remote endpoints. + + Which means a port must not contain mixed-use endpoints, they + must all refer to media links between V4L2 subdevices. -- 2.7.4