- 04 Jan, 2021 40 commits
-
-
Michael Tretter authored
The dst_handle field in the PUT_STREAM_BUFFER and ENCODE_FRAME_RESPONSE is used to retrieve the V4L2 CAPTURE buffer. Calling it stream_id is confusing. Call it dst_handle inspired by src_handle for the OUTPUT buffer. Signed-off-by:
Michael Tretter <m.tretter@pengutronix.de> Signed-off-by:
Hans Verkuil <hverkuil-cisco@xs4all.nl> Signed-off-by:
Mauro Carvalho Chehab <mchehab+huawei@kernel.org>
-
Michael Tretter authored
The v4l2-m2m has various helpers for correctly handle the draining. Drop the driver specific state machine and use the m2m helper functions. Signed-off-by:
-
Michael Tretter authored
Convert the Allegro DVT video IP codec text binding to Yaml. Add the converted binding to the MAINTAINERS file. Signed-off-by:
Rob Herring <robh@kernel.org> Signed-off-by:
-
Michael Tretter authored
The stateful encoder API was finalized. Nothing is blocking the driver from being moved out of staging. Signed-off-by:
-
Dan Carpenter authored
The "channel" is added to the "dev->channels" but then if v4l2_m2m_ctx_init() fails then we free "channel" but it's still on the list so it could lead to a use after free. Let's not add it to the list until after v4l2_m2m_ctx_init() succeeds. Fixes: cc62c747 ("media: allegro: add missed checks in allegro_open()") Signed-off-by:
Dan Carpenter <dan.carpenter@oracle.com> Reviewed-by:
-
Hans Verkuil authored
If the CEC device was unregistered, then add EPOLLPRI to the poll() mask. Otherwise a select() that only waits for exceptions will not wake up. A select() that waits for read and/or write events *will* wake up on an EPOLLERR, but not (for some reason) if it just waits for exceptions. Strangly the epoll functionality will wakeup on EPOLLERR if you just wait for an exception, so in this respect select() and epoll differ. In the end it doesn't really matter, what matters is that polling file handles are woken up on device unregistration. Signed-off-by:
-
Hans Verkuil authored
If the V4L2 device was unregistered, then add EPOLLPRI to the poll mask. Otherwise a select() that only waits for exceptions will not wake up. A select() that waits for read and/or write events *will* wake up on an EPOLLERR, but not (for some reason) if it just waits for exceptions. Strangly the epoll functionality will wakeup on EPOLLERR if you just wait for an exception, so in this respect select() and epoll differ. In the end it doesn't really matter, what matters is that polling file handles are woken up on device unregistration. It also improves the code a bit if vdev->fops->poll is NULL: this didn't check for device unregistration. Signed-off-by:
-
Hans Verkuil authored
When the disconnect error injection control is set, then besides faking unregistering the device nodes, also call v4l2_event_wake_all() to ensure any userspace applications will wake up as per a 'normal' unregister. Signed-off-by:
-
Hans Verkuil authored
When unregistering a V4L2 device node, make sure any filehandles that are waiting for an event are woken up. Add v4l2_event_wake_all() to v4l2-event.c and call it from video_unregister_device(). Otherwise userspace might never know that a device node was removed. [hverkuil: checkpatch: replaced 'if (vdev == NULL)' by 'if (!vdev)'] Signed-off-by:
-
Alexandre Courbot authored
do_poll()/do_select() seem to set the _qproc member of poll_table to NULL the first time they are called on a given table, making subsequent calls of poll_wait() on that table no-ops. This is a problem for mem2mem which calls poll_wait() on the V4L2 queues' waitqueues only when a queue-related event is requested, which may not necessarily be the case during the first poll. For instance, a stateful decoder is typically only interested in EPOLLPRI events when it starts, and will switch to listening to both EPOLLPRI and EPOLLIN after receiving the initial resolution change event and configuring the CAPTURE queue. However by the time that switch happens and v4l2_m2m_poll_for_data() is called for the first time, poll_wait() has become a no-op and the V4L2 queues waitqueues thus cannot be registered. Fix this by moving the registration to v4l2_m2m_poll() and do it whether or not one of the queue-related events are requested. Signed-off-by:
Alexandre Courbot <gnurou@gmail.com> Signed-off-by:
-
Alexandre Courbot authored
do_poll()/do_select() seem to set the _qproc member of poll_table to NULL the first time they are called on a given table, making subsequent calls of poll_wait() on that table no-ops. This is a problem for vb2 which calls poll_wait() on the V4L2 queues' waitqueues only when a queue-related event is requested, which may not necessarily be the case during the first poll. Fix this by making the call to poll_wait() happen first thing and unconditionally in vb2_core_poll(). Signed-off-by:
-
Sowjanya Komatineni authored
This patch adds custom V4L2 control for syncpt timeout retry to continue capture on error for specified retries count through this control. This is useful for HDMI-to-CSI bridge debug purposes like for hotplug scenarios or for ignoring captures till HDMI input is stabilized. Signed-off-by:
Sowjanya Komatineni <skomatineni@nvidia.com> Signed-off-by:
-
Sowjanya Komatineni authored
Tegra VI/CSI hardware don't have native 8 lane capture support. Each CSI port has max 4 lanes only. So for x8 captures, consecutive ports are ganged up for left half and right half captures on to each x4 ports with buffer offsets based on source image split width to align side-by-side. All ports in gang are configured together during the corresponding video device node streaming for x8 captures. x8 capture with gang ports are supported with HDMI-to-CSI bridges where they split 4K image into left half onto one x4 port and right half onto second x4 port. Signed-off-by:
-
Sowjanya Komatineni authored
Tegra VI/CSI hardware don't have native 8 lane CSI RX port. But x8 capture can be supported by using consecutive x4 ports simultaneously with HDMI-to-CSI bridges where source image is split on to two x4 ports. This patch updates dt-bindings for csi endpoint data-lane property with maximum of 8 lanes. Acked-by:
Sakari Ailus <sakari.ailus@linux.intel.com> Signed-off-by:
-
Sowjanya Komatineni authored
Some CSI2 receivers support 8 data lanes. So, this patch updates CSI2 maximum data lanes to be 8. Acked-by:
-
Sowjanya Komatineni authored
Implement V4L2 device notify callback to handle V4L2_EVENT_SOURCE_CHANGE event from subdevices. HDMI-to-CSI bridge drivers trigger V4L2_EVENT_SOURCE_CHANGE when source DV timing changes are detected or when HDMI hotplug happens. Runtime source parameter changes during active streaming is not allowed and Tegra video driver calls vb2_queue_error to signal a fatal error if a notification of this event happens during an active streaming. Signed-off-by:
-
Sowjanya Komatineni authored
Current implementation uses v4l2_ctrl_subscribe_event() and this does not handle V4L2_EVENT_SOURCE_CHANGE. So, update driver to handle V4L2_EVENT_SOURCE_CHANGE. Signed-off-by:
-
Sowjanya Komatineni authored
This patch adds support for log_status ioctl. Signed-off-by:
-
Sowjanya Komatineni authored
This patch adds support for EDID get and set v4l2 ioctl ops to use with HDMI to CSI bridges. Signed-off-by:
-
Sowjanya Komatineni authored
This patch adds below v4l2 DV timing ioctls to support HDMI-to-CSI bridges. Signed-off-by:
-
Sowjanya Komatineni authored
NV16 are two-plane versions of YUV422 format. VI/CSI surface0 registers corresponds to first Y plane and surface1 registers corresponds to seconds UV plane. This patch updates image size for NV16 format to include both planes and programs VI/CSI surface1 registers for UV plane capture. Signed-off-by:
-
Sowjanya Komatineni authored
V4L2 pixel format is incorrect for RGB and YUV formats. This patch fixes it. Signed-off-by:
-
Sowjanya Komatineni authored
VI Pixel transforms converts source pixel data to selected destination pixel formats in memory and aligns properly. YUV and RGB formats need this pixel transform to be enabled. RAW formats use T_R16_I destination pixel format in memory and does not need pixel transform as they support direct write to memory. So, this patch enables pixel transform for YUV and RGB and keeps it bypass for RAW formats. Signed-off-by:
-
Sowjanya Komatineni authored
Currently try format implementation doesn't check if subdevice has get_selection ops implemented and returns -EINVAL on error from direct v4l2_subdev_call of get_selection. Selection API's are not mandatory for all V4L2 subdevices. This patch fixes it by adding v4l2_subdev_has_ops check prior to calling get_selection ops and continues with try or set format with zero crop settings for subdevices that don't have get_selection and returns -EINVAL only for subdevices that has get_selection ops. Signed-off-by:
-
Laurent Pinchart authored
Rework the context API exposed to cal-video.c to simplify it. The configuration and enable steps are all grouped in a single cal_ctx_start() function, and the DMA stop and IRQ disable are similarly groupd in cal_ctx_stop(). The cal_ctx_wr_dma_addr() function is renamed to cal_ctx_set_dma_addr() for consistency with the cal_ctx_ prefix of the start and stop functions. Signed-off-by:
Laurent Pinchart <laurent.pinchart@ideasonboard.com> Reviewed-by:
Benoit Parrot <bparrot@ti.com> Signed-off-by:
-
Laurent Pinchart authored
dma_addr_t is the correct type to store DMA addresses. Replace incorrect usage of unsigned long and unsigned int. Signed-off-by:
-
Laurent Pinchart authored
When a pending buffer becomes active, the cal_dmaqueue.active field is updated, but the pending field keeps the same value until a new buffer becomes pending. This requires handling the special case of pending == active in different places. Simplify the code by setting the pending field to NULL when the pending buffer becomes active. Buffers are now simply moved from queue to pending and from pending to active. Signed-off-by:
-
Laurent Pinchart authored
The cal_dmaqueue structure only contains the list of queued buffers. Move the other fields that are logically related to the DMA queue (current and next buffer points, state, wait queue and lock) from cal_ctx to cal_dmaqueue. Take this as an opportunity to document the fields usage and to give them more appropriate names. The 'active' field stored the list of all queued buffers, not the active buffers, so rename it to 'queue'. The 'cur_frm' and 'next_frm' are respectively renamed to 'active' and 'pending' to better explain their purpose. The 'dma_state' and 'dma_wait' fields are stripped of their 'dma_' prefix as they're now part of cal_dmaqueue. Finally, 'slock' is renamed to 'lock'. Signed-off-by:
-
Laurent Pinchart authored
Use the list_first_entry() macro where appropriate to replace manual usage of list_entry(head.next). Signed-off-by:
-
Laurent Pinchart authored
The vb2 queue ensures that the start streaming operation will only be called with a minimal number of buffers queued to the driver. There's thus no need to manually check if the DMA queue is empty. Remove the check. Signed-off-by:
-
Laurent Pinchart authored
The cal_calc_format_size() function checks that the passed fmtinfo argument is not NULL. All callers ensure that the pointer is valid. Drop the check. Signed-off-by:
-
Laurent Pinchart authored
The V4L2 read/write API is inefficient and makes little sense for devices that handle frame-based formats. Applications shouldn't use it, drop its support from the driver. Signed-off-by:
-
Laurent Pinchart authored
The cal_start_streaming() and cal_stop_streaming() functions duplicate the same buffer release logic. split it to a separate function to share the code. Signed-off-by:
-
Laurent Pinchart authored
The cal_start_streaming() and cal_stop_streaming() functions are called with interrupts enabled. spin_lock_irq() can thus be used instead of spin_lock_irqsave(). Signed-off-by:
-
Laurent Pinchart authored
When stopping the stream, the driver needs to ensure that ongoing DMA completes and that no new DMA is started. It does so using a feature of the PPI that can be stopped on a frame boundary. The downside of this mechanism is that the DMA can't be stopped independently of the source, which prevents usage of multiple contexts for the same source (to handle CSI-2 virtual channels or data types). Rework the stream stop mechanism to stop the write DMA without disabling the PPI first. The new mechanism relies on the combination of a state machine in the driver and shadowing of the CAL_WR_DMA_CTRL_j.MODE field in the hardware. When a stop is requested, the DMA start interrupt handler will request the hardware to stop at the end of the current frame by disabling the write DMA context in the shadowed register, and flag that a stop is in progress. The next DMA end interrupt will then report that the stop is complete. This makes it possible to stop the PPI after stopping the DMA, and fold the cal_camerarx_ppi_disable() call into cal_camerarx_stop(). Signed-off-by:
-
Laurent Pinchart authored
To further decouple the context and CAMERARX components, move the call to cal_camerarx_ppi_enable() from cal_start_streaming() to the CAMERARX .s_stream() operation. The DMA destination address has to be set before starting the CAMERARX, which is desirable anyway. cal_camerarx_ppi_disable() will be addressed separately. Signed-off-by:
-
Laurent Pinchart authored
Interrupts are enabled and disabled by the cal_camerarx_enable_irqs() and cal_camerarx_disable_irqs(). Despite their name, they deal with all interrupts, not just the CAMERARX interrupts, and they hardcode the assumption that the context index is identical to the CAMERARX index. Split the context-related interrupt management to two new functions, cal_ctx_enable_irqs() and cal_ctx_disable_irqs(), called from the cal_start_streaming() and cal_stop_streaming() functions. The explicit calls to cal_camerarx_enable_irqs() and cal_camerarx_disable_irqs() are folded with the CAMERARX .s_stream() operation to simplify the CAMERARX API. Enabling the OCPO error interrupt is moved to the PM runtime resume operation, as it's global to the device, not related to a CAMERARX or context. The VC IRQ enable and disable are removed as they're not used, the parent interrupt bit (CAL_HL_IRQ_VC_MASK) never being set. Signed-off-by:
-
Laurent Pinchart authored
The cal_fmt structure stores information about a format. Its name is ambiguous, as it could be understood as storing a format. Furthermore, instances of the structure are called 'fmt' through the code, which leads to confusion with the 'format' variables. Rename the structure to cal_format_info, and the corresponding 'fmt' variables to 'fmtinfo'. Signed-off-by:
-
Laurent Pinchart authored
The cal_ctx_wr_dma_config() function has access to the context, there's no need to give it format-related values retrieved from the context. Access the values internally, and reword internal comments. The comment regarding the CAL_WR_DMA_OFST register not being well understood is dropped, as the datasheet explicitly documents it as "Offset between two consecutive line starts". Signed-off-by:
-
Laurent Pinchart authored
Replace calls to cal_camerarx_start() and cal_camerarx_stop() with usage of the .s_stream() subdev operation. This requires calling the .set_fmt() operation as the CAMERARX now relies on the format information set through the subdev API instead of receiving it in the cal_camerarx_start() function. This change prepare for exposing the CAMERARX subdev operations to userspace by using the same API within the kernel. Signed-off-by:
-
