- 24 Sep, 2022 40 commits
-
-
Hans de Goede authored
When atomisp_open() fails then it must call v4l2_fh_release() to undo the results of v4l2_fh_open(). Reviewed-by:
Andy Shevchenko <andriy.shevchenko@intel.com> Signed-off-by:
Hans de Goede <hdegoede@redhat.com> Signed-off-by:
Mauro Carvalho Chehab <mchehab@kernel.org>
-
Hans de Goede authored
Now that the registering of the /dev/* video / subdev nodes has been moved to the end of atomisp_pci_probe() the workaround with the loading mutex to delay opens until init is done is no longer necessary. Reviewed-by:
-
Hans de Goede authored
Register /dev/* nodes at the end of atomisp_pci_probe(), this is a prerequisite for dropping the loading mutex + ready flag kludge for delaying open() calls on the /dev/* nodes . Reviewed-by:
-
Hans de Goede authored
Split subdev and video-node registration into 2 steps, this is a preparation step for moving video-node registration to the end of probe() so that the loading() mutex can be removed. Reviewed-by:
-
Hans de Goede authored
atomisp_css_set_cont_prev_start_time() is a no-op, remove it. Reviewed-by:
-
Hans de Goede authored
Remove some more code which is no longer referenced after the removal of the ATOMISP_ACC_* custom ioctls. Reviewed-by:
-
Hans de Goede authored
The ACC /dev/video# device node uses a struct video_device embedded in an atomisp_acc_pipe struct instead of in an atomisp_video_pipe struct. Yet it uses the same file-ops and ioctl-ops even though it does not have a videobuf queue, which makes e.g. the mmap fop nonsense. Worse the only file-ops / ioctls which differentiate between the 2 types and correctly do container_of on the right type are the open/release fops and the vidioc_default handler. The mmap and poll fops and *all* other ioctl handlers unconditionally do container_of on the passed in struct video_device blindly assuming they are dealing with the one embedded in the atomisp_video_pipe struct. This makes it trivial for userspace to cause all sort of undefined behavior by calling mmap, poll or the other ioctls on the ACC device node! Presumably the use of the ACC device node was to allow making the special ioctls to load custom programs while the other /dev/video# nodes were already open, since the /dev/video# nodes can currently all be opened only once (which needs to be fixed). commit 4bbca788 ("media: atomisp: remove private acceleration ioctls") has removed the custom ATOMISP_ACC_* ioctls, so there no longer is any reason to keep the ACC device node. As explained above its presence can easily cause the kernel to crash, so remove the ACC device node and the code for handling it. Reviewed-by:
-
Hans de Goede authored
After the file-injection support removal the file_input flag is always false. Remove the flag and replace any code checking it with the code-path for when it is false. Reviewed-by:
-
Hans de Goede authored
After the file-injection support removal the outq videobuf queue is no longer used, remove it. Reviewed-by:
-
Hans de Goede authored
After the file-injection support removal, atomisp_video_pipe->type never is V4L2_BUF_TYPE_VIDEO_OUTPUT anymore, so the V4L2_BUF_TYPE_VIDEO_OUTPUT support path in atomisp_video_init() is never hit and this path is the only user of atomisp_file_fops and atomisp_file_ioctl_ops. Remove atomisp_file_fops and atomisp_file_ioctl_ops and all of the functions which are only referenced by these ops structs. Reviewed-by:
-
Hans de Goede authored
The file-injection support of the atomisp driver has not been tested and is not necessary for camera support, remove it. Note the main reason for removing this is because it depends on the videobuf (version 1) outq and we want to remove or replace all videobuf usage in the driver. Reviewed-by:
-
Hans de Goede authored
atomisp_subdev_register_entities() had V4L2_CAP_VIDEO_CAPTURE / V4L2_CAP_VIDEO_OUT swapped. Or-ing in V4L2_CAP_VIDEO_OUT for the nodes which allow capturing from the camera and or-ing in V4L2_CAP_VIDEO_CAPTURE for the file-injection node (mem2mem use of the ISP). Things happen to still work for the capture device-nodes because the "shared" caps also included V4L2_CAP_VIDEO_CAPTURE, so those shared nodes advertised V4L2_CAP_VIDEO_CAPTURE | V4L2_CAP_VIDEO_OUT. Fix things so that only the correct caps are advertised. Reviewed-by:
-
Hans de Goede authored
The atomisp code needs USERPTR pointers to be page aligned, otherwise bad things (scribbling over other parts of the process' RAM) happen. Add a check to ensure this and exit VIDIOC_QBUF calls with unaligned pointers with -EINVAL. Reviewed-by:
Andy Shevchenko <andy.shevchenko@gmail.com> Signed-off-by:
-
Hans de Goede authored
alloc_user_pages() is only ever called on qbuf for USERPTR buffers which always hits the get_user_pages_fast() path, so the pin_user_pages() path can be removed. Getting the vma then also is no longer necessary since that is only done to determine which path to use. And this also removes the only users of the mem_type struct hmm_bo member, so remove that as well. Reviewed-by:
-
Hans de Goede authored
Rewrite free_private_pages() using pages_array helper funcs. Reviewed-by:
-
Hans de Goede authored
Rewrite alloc_private_pages() using pages_array helper funcs. Note alloc_pages_bulk_array() skips non NULL pages, so switch the allocating of the pages pointer array to kcalloc to ensure the pages are initially all set to NULL. Reviewed-by:
-
Hans de Goede authored
Further simplify alloc_private_pages(). Reviewed-by:
-
Hans de Goede authored
Since lack_mem starts initialized to true, alloc_private_pages() will always set order to HMM_MIN_ORDER aka 0 / will always alloc 1 page at a time. So all the magic to decrease order if allocs fail is not necessary and can be removed. Reviewed-by:
-
Hans de Goede authored
atomisp_try_fmt() gives results with padding included. So when userspace asks for e.g. 1600x1200 then we should pass 1616x1216 to atomisp_try_fmt() this will then get adjusted back to 1600x1200 before returning it to userspace by the atomisp_adjust_fmt() call at the end of atomisp_try_fmt(). This fixes the resolution list in camorama showing resolutions like e.g. 1584x1184 instead of 1600x1200. Reviewed-by:
-
Hans de Goede authored
atomisp_try_fmt() calls the sensor's try_fmt handler but it does not copy the result back to the passed in v4l2_pix_format under some circumstances. Potentially returning an unsupported resolution to userspace, which VIDIOC_TRY_FMT is not supposed to do. atomisp_set_fmt() also uses atomisp_try_fmt() and relies on this wrong behavior. The VIDIOC_TRY_FMT call passes NULL for the res_overflow argument where as the atomisp_set_fmt() call passes non NULL. Use the res_overflow argument to differentiate between the 2 callers and always propagate the sensors result in the VIDIOC_TRY_FMT case. This fixes the resolution list in camorama showing resolutions like e.g. 1584x1184 instead of 1600x1200. Reviewed-by:
-
Hans de Goede authored
Add info about sensors v4l2_get_subdev_hostdata() use, to notes.txt. Reviewed-by:
-
Hans de Goede authored
Exit with an error on any i2c-write errors, rather then only exiting with an error when ov2680_get_intg_factor() fails. Reviewed-by:
-
Hans de Goede authored
On ov2680_set_fmt() calls with format->which == V4L2_SUBDEV_FORMAT_TRY, ov2680_set_fmt() does not talk to the sensor, so there is no need to lock the dev->input_lock mutex in this case. Reviewed-by:
-
Hans de Goede authored
On sets actually store the set (closest) format inside ov2680_device.dev, so that it also properly gets returned by get_fmt. This fixes the following problem: 1. App does an VIDIOC_SET_FMT 640x480, calling ov2680_set_fmt() 2. Internal buffers (atomisp_create_pipes_stream()) get allocated at 640x480 size by atomisp_set_fmt() 3. ov2680_get_fmt() gets called later on and returns 1600x1200 since ov2680_device.dev was not updated. So things get configured to stream at 1600x1200, but the internal buffers created during atomisp_create_pipes_stream() do not get updated in size 4. streaming starts, internal buffers overflow and the entire machine freezes eventually due to memory being corrupted Reviewed-by:
-
Andy Shevchenko authored
The acpi_evaluate_dsm_typed() provides a way to check the type of the object evaluated by _DSM call. Use it instead of open coded variant. Link: https://lore.kernel.org/r/20220730155905.90091-1-andriy.shevchenko@linux.intel.comSigned-off-by:
Andy Shevchenko <andriy.shevchenko@linux.intel.com> Acked-by:
Sakari Ailus <sakari.ailus@linux.intel.com> Tested-by:
-
Krzysztof Kozlowski authored
Convert the Samsung Exynos SoC G-Scaler bindings to DT schema. Changes done during conversion: 1. A typical (already used) properties like clocks, iommus and power-domains. 2. Require clocks, because they are essential for the block to operate. 3. Describe the differences in clocks between the Exynos5250/5420 and the Exynos5433 G-Scalers. This includes the fifth Exynos5433 clock "gsd" (GSCL Smart Deck) which was added to the DTS, but not to the bindings and Linux driver. Similarly to Exynos5433 DECON change [1], the clock should be used. [1] https://lore.kernel.org/all/6270db2d-667d-8d6f-9289-be92da486c25@samsung.com/Signed-off-by:
Krzysztof Kozlowski <krzysztof.kozlowski@linaro.org> Reviewed-by:
Rob Herring <robh@kernel.org> Signed-off-by:
-
Krzysztof Kozlowski authored
Convert Dongwoon Anatech DW9714 camera voice coil lens driver to DT schema and extend the bindings with vcc-supply (already used by driver) and powerdown-gpios (based on datasheet, not used by the driver). Signed-off-by:
-
Tomi Valkeinen authored
Now that the pipeline is stored into pads instead of entities, we can change the relevant functions to take pads instead of entities. Signed-off-by:
Tomi Valkeinen <tomi.valkeinen@ideasonboard.com> Reviewed-by:
Laurent Pinchart <laurent.pinchart+renesas@ideasonboard.com> Signed-off-by:
-
Tomi Valkeinen authored
Add a new media entity operation, has_pad_interdep. The optional op is used to discover the pad interdependencies inside an entity during the pipeline construction. Signed-off-by:
-
Laurent Pinchart authored
[Note: the code is mostly from Laurent but the patch description is from Tomi] The media_pipeline_start() and media_pipeline_stop() functions use the media graph walk API to traverse the graph and validate the pipeline. The graph walk traverses the media graph following links between the entities. Also, while the pipeline can't change between the start and stop calls, the graph is walked again from scratch at stop time, or any time a driver needs to inspect the pipeline. With the upcoming multiplexed streams support we will need a bit more intelligent pipeline construction, as e.g. two independent streams may be passing through a single entity via separate pads in which case those pads should not be part of the same pipeline. This patch essentially rewrites the media_pipeline_start/stop so that a pipeline is defined as a set of pads instead of entities and the media graph traversal considers the pad interdependencies when choosing which links to follow. Currently all the entity's pads are considered as interdependent. This means that the behavior with all the current drivers stays the same, but in the future we can define a more fine-grained pipeline construction. Additionally the media pipeline's pads are cached at media_pipeline_start() time, and re-used at media_pipeline_stop() which avoid the need to re-walk the whole graph as the previous implementation did. Also, caching pads in the pipeline can serve in the future as the foundation to provide a better API than the media graph walk to drivers to iterate over pads and entities in the pipeline. Note that the old media_pipeline_start/stop used the media graph walk API. The new version does not use the media graph walk API, but instead a new implementation. There are two reasons for not changing the graph walk: it proved to be rather difficult to change the graph walk to have the features implemented in this patch, and second, this keeps the backward compatibility of the graph walk as there are users of the graph walk API The long term plan is that all the existing code would be converted to use the new cached pipeline, thus allowing us to remove the graph walk. Signed-off-by:
-
Tomi Valkeinen authored
Use video_device_pipeline_alloc_start() instead of manually allocating/managing the media pipeline storage. Signed-off-by:
-
Tomi Valkeinen authored
Add new variant of media_pipeline_start(), media_pipeline_alloc_start(). media_pipeline_alloc_start() can be used by drivers that do not need to extend the media_pipeline. The function will either use the pipeline already associated with the entity, if such exists, or allocate a new pipeline. When media_pipeline_stop() is called and the pipeline's use count drops to zero, the pipeline is automatically freed. Signed-off-by:
-
Tomi Valkeinen authored
Use video_device_pipeline() in the drivers instead of media_entity_pipeline(). Signed-off-by:
-
Tomi Valkeinen authored
Convert the media drivers to use video device based pipeline start/stop where possible. Signed-off-by:
-
Tomi Valkeinen authored
With the upcoming stream related improvements to the pipelines, the pipelines are moved from media entities to media pads. As the drivers currently use the pipelines with the entity based model, moving the pipelines to pads will cause changes to the drivers. However, most of the uses of media pipelines are related to a video device (a DMA engine) with a single pad, and thus there's never a need to support multiple pads in these use cases. We can avoid pushing the complexities of the pad based model to the drivers by adding video device wrappers for the pipeline related functions. This patch adds a number of wrappers to media_pipeline functions, all of which take a video_device as a parameter (instead of a media_entity), and verify that there's just one pad. Signed-off-by:
-
Laurent Pinchart authored
Replace direct access to the pipe field in drivers with a new helper function. This will allow easier refactoring of media pipeline handling in the MC core behind the scenes without affecting drivers. Signed-off-by:
Laurent Pinchart <laurent.pinchart@ideasonboard.com> Signed-off-by:
-
Laurent Pinchart authored
The media_entity_get_fwnode_pad() function is unrelated to the graph traversal code that it is currently bundled with. Move it with the media_entity_remote_pad() function. Signed-off-by:
-
Laurent Pinchart authored
The media_entity_enum_init() function is a wrapper around __media_entity_enum_init() that turns a media_device pointer argument into the maximum entity ID in the corresponding media graph. __media_entity_enum_init() is never used outside of media_entity_enum_init(), so the two functions can be merged together. Signed-off-by:
-
Jacopo Mondi authored
Add an iterator helper to easily cycle through all pads in an entity and use it in media-entity and media-device code where appropriate. Signed-off-by:
Jacopo Mondi <jacopo+renesas@jmondi.org> Reviewed-by:
-
Tomi Valkeinen authored
'streaming_count' is a bit misleading name, as the count is increased with media_pipeline_start(). Let's rename it to 'start_count' instead. Signed-off-by:
-
