Hi, This series adds Gen3 VIN support to rcar-vin driver for Renesas r8a7795, r8a7796 and r8a77970. It is based on the media-tree and depends on Fabrizio Castro patches as they touches the order of the compatible strings in the documentation to reduce merge conflicts. The dependencies are included in this series. The driver is tested on Renesas H3 (r8a7795, ES2.0), M3-W (r8a7796) together with the rcar-csi2 driver (posted separately and not yet upstream) and the Salvator-X onboard ADV7482. It is also tested on the V3M (r8a77970) on the Eagle board together with its expansion board with a ADV7482 and out of tree patches for GMSL capture using the max9286 and rdacm20 drivers. It is possible to capture both CVBS and HDMI video streams, v4l2-compliance passes with no errors and media-ctl can be used to change the routing and formats for the different entities in the media graph. Gen2 compatibility is verified on Koelsch and no problems where found, video can be captured just like before and v4l2-compliance passes without errors or warnings just like before this series. For convenience the series can be fetched from: git://git.ragnatech.se/linux rcar/vin/mc-v13 I have started on a very basic test suite for the VIN driver at: https://git.ragnatech.se/vin-tests And as before the state of the driver and information about how to test it can be found on the elinux wiki: http://elinux.org/R-Car/Tests:rcar-vin * Changes from v12 - Rebase to latest media-tree/master changed a 'return ret' to a 'goto out' in rvin_start_streaming() to take recent changes to the VIN driver into account. - Moved field != V4L2_FIELD_ANY in 'rcar-vin: set a default field to fallback on' check from a later commit 'rcar-vin: simplify how formats are set and reset' in the series. This is to avoid ignoring the field returned from the sensor if FIELD_ANY was requested by the user. This was only a problem between this change and a few patches later, but better to fix it now. Reported by Hans, thanks for spotting this. - Fix spelling. - Add review tags from Hans. * Changes since v11 - Rewrote commit message for '[PATCH v11 22/32] rcar-vin: force default colorspace for media centric mode'. Also set fixed values for xfer_func, quantization and ycbcr_enc. - Reorderd filed order in struct rvin_group_route. - Renamed chan to channel in struct rvin_group_route. - Rework 'rcar-vin: read subdevice format for crop only when needed' into 'rcar-vin: simplify how formats are set and reset'. - Keep caching the source dimensions and drop all changes to rvin_g_selection() and rvin_s_selection(). - Inline rvin_get_vin_format_from_source() into rvin_reset_format() which now is the only user left. - Add patch to cache the video standard instead of reading it at stream on. - Fix error labels in rvin_mc_open(). - Fixed spelling in commit messages and comment, thanks Laurent! - Added reviewed tags from Laurent, Thanks! * Changes since v10 - Corrected spelling in comments and commit messages. - Reworked 'rcar-vin: read subdevice format for crop only when needed' to only get the source format once per operation. - Moved some patches around to make it easier to review, moved: - rcar-vin: set a default field to fallback on - rcar-vin: fix handling of single field frames (top, bottom and alternate fields) - rcar-vin: update bytesperline and sizeimage calculation - rcar-vin: break out format alignment and checking - rcar-vin: update pixelformat check for M1 Before: - rcar-vin: read subdevice format for crop only when needed - Rename variable 'code' to 'mbus_code' in struct rvin_dev. - Add comment describing no locking is needed in rvin_set_channel_routing(). - Check return value of pm_runtime_get_sync() in rvin_set_channel_routing(). - Rework 'rcar-vin: add check for colorspace' to not try to check the format, instead force a default format. This should be revisited once either v4l2-compliance or v4l2 framework changes are worked out to allow for MC centric drivers to validate user supplied colorspace. - Add error checking for pm_runtime_get_sync() and v4l2_pipeline_pm_use(). - Change mutex_lock() to mutex_lock_interruptible() in rvin_mc_open(). - Rewrote documentation for struct rvin_group_route. - Rename rvin_mc_parse_v4l2() to rvin_mc_parse_of_endpoint(). - Reword error messages in rvin_mc_parse_of_endpoint(). - Removed unneeded loop in rvin_mc_parse_of_endpoint(). - Remove check !is_media_entity_v4l2_subdev() in rvin_group_entity_to_csi_id(). - Add documentation for the algorithm used to figure out if a link can be enabled or not in rvin_group_link_notify(). - Break out format validation to rvin_mc_validate_format(). - Include two DT documentation patches from Fabrizio Castro which previously where mentioned as the only dependency for this series. - Added reviewed tags from Laurent, Thanks! * Changes since v9 - Fixed mistakes in the device tree description pointed out by Laurent. - GenX -> GenX platforms - portX -> port X - Explicitly state the on Gen3 platforms port 0 can only describe one endpoint and that only VIN instances connected to external pins should have a port 0 node. - s/which is/connected to/ in he endpoint description for Gen3 platforms. - Update some poorly written commit messages. - Moved the digital subdevice attach and detach code to two separate functions to increase readability. - Rename the struct rvin_info member chip to model to better describe its purpose. - Change the video name from "rcar_vin e6ef0000.video" to "VINx output" where x is the VIN id. - Dropped patch 'rcar-vin: do not allow changing scaling and composing while streaming' as it removed Gen2 functionality which is valid as pointed out by Laurent. - Rename rvin_get_sd_format() to rvin_get_source_format() and change its parameter from struct v4l2_pix_format* to struct v4l2_mbus_framefmt*. - Clarified commit message and add a few move comments to 'rcar-vin: fix handling of single field frames (top, bottom and alternate fields'. - Update documentation for struct rvin_dev fields mbus_cfg and code. - Fix argument in VNCSI_IFMD_CSI_CHSEL macro. - Renamed rvin_set_chsel() to rvin_set_channel_routing(). - Restore the VNMC register after changing CHSEL setting. - Broke patch 'rcar-vin: break out format alignment and checking' into three parts to ease review. - Add new patch to introduce a default field. - Only include media/v4l2-mc.h in the .c files that needs it and not in rcar-vin.h. - Rename rvin_group_allocate() rvin_group_get() - Rename rvin_group_delete() rvin_group_put() - Updated error message "VIN number %d already occupied\n" to "Duplicate renesas,id property value %u\n". - Removed dev_dbg messages which only where useful during development. - Dropped patch '[PATCH v9 10/28] rcar-vin: do not reset crop and compose when setting format' as it introduces a regression on Gen2. - Inline rvin_group_read_id() and rvin_group_add_vin() into rvin_group_get(). - Define static variables at the top of rcar-core.c. - Fix potential deadlock in rvin_group_get(). - Set media device model name to the VIN module name. - Set media device model to matched complicity string. - Do not use a 2 dimensional sparse array for chsel values in struct rvin_info, instead use a flat array and store VIN and CHSEL value inside each array item. - Reworked DT parsing code to make use of the new v4l2_async_notifier_parse_fwnode_endpoints_by_port() helper removing a lot of iffy custom parsing code. - Rework how CHSEL value is calculated in the link notifier callback. Using a bitmap of possible values instead of looping over an array turned out great reducing both LoC and increasing readability of the code which was a bit difficult before. It also reduced to memory needed to contain the static routing information. - Verify the media bus pixel code when starting a stream. - Take the media device graph lock when figuring out and starting a stream so not to race between simultaneous stream start from multiple rcar-vin instances as they might share common subdevices. - Added review tags from Laurent. - Dropped tags from patches that where changed more then just correcting spelling or other gramma errors. * Changes since v8 - Fixed issue in rvin_group_init() where rvin_group_update_links() was called, but after moving the video device registration to the complete() callback this is now invalid. - Fixed spelling in commit messages. - Added review and acks from Hans, Kieran and Rob. * Changes since v7 - Dropped '[PATCH v7 02/25] rcar-vin: register the video device at probe time' - Add patch which renames four badly name functions. Some of the renaming was in v7 part of the dropped patch 02/25. Make it a own patch and rename all badly named functions in one patch. - Add patch to replace part of the functionality of the dropped patch v7 02/25. The new patch keeps the subdevice (un)registration calls in the async callbacks bind() and unbind() but moves the direct subdevice initialization which only is used on Gen2 from the Gen2 and Gen3 shared rvin_v4l2_register(). - Add patch to enable Renesas V3M (r8a77970) - Patch 'rcar-vin: parse Gen3 OF and setup media graph' have had code additions since v7 since it now registers the video devices in the async complete() callback instead of at probe time as an effect of dropping v7 02/25. - The complete() callback now register all video devices. - The unbind() callback now unregister all video devices. - A new member '*notifier' is added to struct rvin_group which keeps track of which rcar-vin instance have registered its notifier on the groups behalf. For the reason above all Reviewed-by tags have been dropped for this patch. - Replaced all kernel messages which used of_node_full_name() as now only returns the basename and not till full path, thanks Geert. printk("%s", of_node_full_name(ep)); -> printk("%pOF", ep); - Added Reviewed-by tags from Hans, big thanks! * Changes since v6 - Rebase ontop of latest media-tree which brings in the use of the fwnode async helpers for Gen2. - Updated DT binding documentation, thanks Laurent for very helpful input! - Removed help text which where copied in from v4l2_ctrl_handler_init() documentation when moving that code block, this was a residue from the soc_camera conversion and should have been removed at that time. - Removed bad check of tvnorms which disables IOCTLs if it's not set, this was a residue from soc_camera conversion and have use in the current driver. - Moved all subdevice initialization from complete to bound handler while improving the unbind handler. With this move all operations of the ctrl_handler from the subdevice is handled in either bound or unbind removing races pointed out by Laurent. - Renamed rvin_v4l2_probe() -> rvin_v4l2_register() and rvin_v4l2_remove() -> rvin_v4l2_unregister(). - Fold rvin_mbus_supported() into its only caller. - Sort compatible string entries in ascending order. - Improved documentation for struct rvin_group_chsel. - Clarify comment in rvin_group_csi_pad_to_chan(). - Make use of of_device_get_match_data() as suggested by Geert. - Fixed spelling mistakes. - Added review tags from Hans. * Changes since v5 - Extract and make use of common format checking for both Gen2 and Gen3. - Assign pad at declaration time in rvin_get_sd_format() - Always call pm_runtime_{get_sync,put}() and v4l2_pipeline_pm_use() when opening/closing a video device, remove the check of v4l2_fh_is_singular_file(). - Make rvin_set_chsel() return void instead of int since it always return 0. - Simplify the VIN group allocator functions. - Make the group notifier callbacks and setup more robust. - Moved the video device registration back to probe time. - Add H3 ES2.0 support. - Fix handling of single field formats (top, bottom, alternate) as this was obviously wrong before but hidden by the Gen2 scaler support. - Added review tags from Kieran. * Changes since v4 (Not posted to ML) - Updated to the new fwnode functions. - Moved the registration of the video devices to the async notification callback. * Changes since v3 - Only add neighboring subdevices to the async notifier. Instead of parsing the whole OF graph depend on incremental async subnotifier to discover the whole pipeline. This is needed to support arbitrarily long graphs and support the new ADV7482 prototype driver which Kieran is working on. - Fix warning from lockdep, reported by Kieran. - Fix commit messages from feedback from Sergei, thanks. - Fix chip info an OF device ids sorting order, thanks Geert. - Use subdev->of_node instead of subdev->dev->of_node, thanks Kieran. * Changes since v2 - Do not try to control the subdevices in the media graph from the rcar-vin driver. Have user-space configure to format in the pipeline instead. - Add link validation before starting the stream. - Rework on how the subdevices are and the video node behave by defining specific V4L2 operations for the MC mode of operation, this simplified the driver quit a bit, thanks Laurent! - Add a new 'renesas,id' DT property which is needed to to be able to keep the VIN to CSI-2 routing table inside the driver. Previously this information was taken from the CSI-2 DT node which is obviously the wrong way to do things. Thanks Laurent for pointing this out. - Fixed a memory leek in the group allocator function. - Return -EMLINK instead of -EBUSY if a MC link is not possible given the current routing setup. - Add comments to clarify that the 4 channels from the CSI-2 node is not directly related to CSI-2 virtual channels, the CSI-2 node can output any VC on any of its output channels. * Changes since v1 - Remove unneeded casts as pointed out by Geert. - Fix spelling and DT documentation as pointed out by Geert and Sergei, thanks! - Refresh patch 2/32 with an updated version, thanks Sakari for pointing this out. - Add Sakaris Ack to patch 1/32. - Rebase on top of v4.9-rc1 instead of v4.9-rc3 to ease integration testing together with renesas-drivers tree. Fabrizio Castro (2): dt-bindings: media: rcar_vin: Reverse SoC part number list dt-bindings: media: rcar_vin: add device tree support for r8a774[35] Niklas Söderlund (31): rcar-vin: add Gen3 devicetree bindings documentation rcar-vin: rename poorly named initialize and cleanup functions rcar-vin: unregister video device on driver removal rcar-vin: move subdevice handling to async callbacks rcar-vin: move model information to own struct rcar-vin: move max width and height information to chip information rcar-vin: move functions regarding scaling rcar-vin: all Gen2 boards can scale simplify logic rcar-vin: set a default field to fallback on rcar-vin: fix handling of single field frames (top, bottom and alternate fields) rcar-vin: update bytesperline and sizeimage calculation rcar-vin: align pixelformat check rcar-vin: break out format alignment and checking rcar-vin: simplify how formats are set and reset rcar-vin: cache video standard rcar-vin: move media bus configuration to struct rvin_dev rcar-vin: enable Gen3 hardware configuration rcar-vin: add function to manipulate Gen3 chsel value rcar-vin: add flag to switch to media controller mode rcar-vin: use different v4l2 operations in media controller mode rcar-vin: force default colorspace for media centric mode rcar-vin: prepare for media controller mode initialization rcar-vin: add group allocator functions rcar-vin: change name of video device rcar-vin: add chsel information to rvin_info rcar-vin: parse Gen3 OF and setup media graph rcar-vin: add link notify for Gen3 rcar-vin: extend {start,stop}_streaming to work with media controller rcar-vin: enable support for r8a7795 rcar-vin: enable support for r8a7796 rcar-vin: enable support for r8a77970 .../devicetree/bindings/media/rcar_vin.txt | 137 ++- drivers/media/platform/rcar-vin/Kconfig | 2 +- drivers/media/platform/rcar-vin/rcar-core.c | 962 +++++++++++++++++++-- drivers/media/platform/rcar-vin/rcar-dma.c | 785 ++++++++++------- drivers/media/platform/rcar-vin/rcar-v4l2.c | 488 ++++++----- drivers/media/platform/rcar-vin/rcar-vin.h | 146 +++- 6 files changed, 1913 insertions(+), 607 deletions(-) -- 2.16.2
