Re: [PATCH v5 06/16] media: Documentation: Add Mali-C55 ISP Documentation

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



Hi Laurent

On 29/05/2024 21:22, Laurent Pinchart wrote:
Hi Dan,

Thank you for the patch.

On Wed, May 29, 2024 at 04:28:48PM +0100, Daniel Scally wrote:
Add a documentation page for the mali-c55 driver, which gives a brief
overview of the hardware and explains how to use the driver's capture
devices and the crop/scaler functions.

Acked-by: Nayden Kanchev <nayden.kanchev@xxxxxxx>
Co-developed-by: Jacopo Mondi <jacopo.mondi@xxxxxxxxxxxxxxxx>
Signed-off-by: Jacopo Mondi <jacopo.mondi@xxxxxxxxxxxxxxxx>
Signed-off-by: Daniel Scally <dan.scally@xxxxxxxxxxxxxxxx>
---
Changes in v5:

	- None

Changes in v4:
	- None

Changes in v3:
	- Documented the synchronised buffer sequence numbers (Sakari)
	- Clarified that the downscale pipe cannot output raw data, the ISP'S
	  resolution limits and choice of media bus format code (Kieran)

Changes in v2:

	- none

  .../admin-guide/media/mali-c55-graph.dot      |  19 +
  Documentation/admin-guide/media/mali-c55.rst  | 333 ++++++++++++++++++
  .../admin-guide/media/v4l-drivers.rst         |   1 +
  3 files changed, 353 insertions(+)
  create mode 100644 Documentation/admin-guide/media/mali-c55-graph.dot
  create mode 100644 Documentation/admin-guide/media/mali-c55.rst

diff --git a/Documentation/admin-guide/media/mali-c55-graph.dot b/Documentation/admin-guide/media/mali-c55-graph.dot
new file mode 100644
index 000000000000..0775ba42bf4c
--- /dev/null
+++ b/Documentation/admin-guide/media/mali-c55-graph.dot
@@ -0,0 +1,19 @@
+digraph board {
+        rankdir=TB
+        n00000001 [label="{{} | mali-c55 tpg\n/dev/v4l-subdev0 | {<port0> 0}}", shape=Mrecord, style=filled, fillcolor=green]
+        n00000001:port0 -> n00000003:port0 [style=dashed]
+        n00000003 [label="{{<port0> 0} | mali-c55 isp\n/dev/v4l-subdev1 | {<port1> 1 | <port2> 2}}", shape=Mrecord, style=filled, fillcolor=green]
+        n00000003:port1 -> n00000007:port0 [style=bold]
+        n00000003:port2 -> n00000007:port2 [style=bold]
+        n00000003:port1 -> n0000000b:port0 [style=bold]
+        n00000007 [label="{{<port0> 0 | <port2> 2} | mali-c55 resizer fr\n/dev/v4l-subdev2 | {<port1> 1}}", shape=Mrecord, style=filled, fillcolor=green]
+        n00000007:port1 -> n0000000e [style=bold]
+        n0000000b [label="{{<port0> 0} | mali-c55 resizer ds\n/dev/v4l-subdev3 | {<port1> 1}}", shape=Mrecord, style=filled, fillcolor=green]
+        n0000000b:port1 -> n00000012 [style=bold]
+        n0000000e [label="mali-c55 fr\n/dev/video0", shape=box, style=filled, fillcolor=yellow]
+        n00000012 [label="mali-c55 ds\n/dev/video1", shape=box, style=filled, fillcolor=yellow]
+        n00000022 [label="{{<port0> 0} | csi2-rx\n/dev/v4l-subdev4 | {<port1> 1}}", shape=Mrecord, style=filled, fillcolor=green]
+        n00000022:port1 -> n00000003:port0
+        n00000027 [label="{{} | imx415 1-001a\n/dev/v4l-subdev5 | {<port0> 0}}", shape=Mrecord, style=filled, fillcolor=green]
+        n00000027:port0 -> n00000022:port0 [style=bold]
+}
\ No newline at end of file
diff --git a/Documentation/admin-guide/media/mali-c55.rst b/Documentation/admin-guide/media/mali-c55.rst
new file mode 100644
index 000000000000..cf4176cb1e44
--- /dev/null
+++ b/Documentation/admin-guide/media/mali-c55.rst
@@ -0,0 +1,333 @@
+.. SPDX-License-Identifier: GPL-2.0
+
+==========================================
+ARM Mali-C55 Image Signal Processor driver
+==========================================
+
+Introduction
+============
+
+This file documents the driver for ARM's Mali-C55 Image Signal Processor. The
+driver is located under drivers/media/platform/arm/mali-c55.
+
+The Mali-C55 ISP receives data in either raw Bayer format or RGB/YUV format from
+sensors through either a parallel interface or a memory bus before processing it
+and outputting it through an internal DMA engine. Two output pipelines are
+possible (though one may not be fitted, depending on the implementation). These
+are referred to as "Full resolution" and "Downscale", but the naming is historic
+and both pipes are capable of cropping/scaling operations. The full resolution
+pipe is also capable of outputting RAW data, bypassing much of the ISP's
+processing. The downscale pipe cannot output RAW data. An integrated test
+pattern generator can be used to drive the ISP and produce image data in the
+absence of a connected camera sensor. The driver module is named mali_c55, and
+is enabled through the CONFIG_VIDEO_MALI_C55 config option.
+
+The driver implements V4L2, Media Controller and V4L2 Subdevice interfaces and
+expects camera sensors connected to the ISP to have V4L2 subdevice interfaces.
+
+Mali-C55 ISP hardware
+=====================
+
+A high level functional view of the Mali-C55 ISP is presented below. The ISP
+takes input from either a live source or through a DMA engine for memory input,
+depending on the SoC integration.::
+
+  +---------+    +----------+                                     +--------+
+  | Sensor  |--->| CSI-2 Rx |                "Full Resolution"    |  DMA   |
+  +---------+    +----------+   |\                 Output    +--->| Writer |
+                       |        | \                          |    +--------+
+                       |        |  \    +----------+  +------+---> Streaming I/O
+  +------------+       +------->|   |   |          |  |
+  |            |                |   |-->| Mali-C55 |--+
+  | DMA Reader |--------------->|   |   |    ISP   |  |
+  |            |                |  /    |          |  |      +---> Streaming I/O
+  +------------+                | /     +----------+  |      |
+                                |/                    +------+
+				                             |    +--------+
+                                                             +--->|  DMA   |
+                                               "Downscaled"       | Writer |
+					          Output          +--------+
You have a mix of tabs and spaces here.

+
+Media Controller Topology
+=========================
+
+An example of the ISP's topology (as implemented in a system with an IMX415
+camera sensor and generic CSI-2 receiver) is below:
+
+
+.. kernel-figure:: mali-c55-graph.dot
+    :alt:   mali-c55-graph.dot
+    :align: center
+
+The driver has 4 V4L2 subdevices:
+
+- `mali_c55 isp`: Responsible for configuring input crop and color space
+                  conversion
+- `mali_c55 tpg`: The test pattern generator, emulating a camera sensor.
+- `mali_c55 resizer fr`: The Full-Resolution pipe resizer
+- `mali_c55 resizer ds`: The Downscale pipe resizer
+
+The driver has 2 V4L2 video devices:
+
+- `mali-c55 fr`: The full-resolution pipe's capture device
+- `mali-c55 ds`: The downscale pipe's capture device
+
+Frame sequences are synchronised across to two capture devices, meaning if one
+pipe is started later than the other the sequence numbers returned in its
+buffers will match those of the other pipe rather than starting from zero.
+
+Frame sequences are synchronised across to two capture devices, meaning if one
+pipe is started later than the other the sequence numbers returned in its
+buffers will match those of the other pipe rather than starting from zero.
I think you can explain this once only.


Oops...


+
+Idiosyncrasies
+--------------
+
+**mali-c55 isp**
+The `mali-c55 isp` subdevice has a single sink pad to which all sources of data
+should be connected. The active source is selected by enabling the appropriate
+media link and disabling all others.
Modelling this with links prevents switching between sources at runtime.
It also makes it possible to misconfigure the pipeline by disconnecting
the two sources. This would be caught at pipeline validation time, but
it still adds complexity.

I was considering using the subdev routing API instead, which would
allow runtime reconfiguration, and prevent invalid configuration in the
first place. The downside is that we would need a mux subdev in front of
the ISP. In terms of additional complexity, that's clearly not great.

Given that switching between the sensor and TPG at runtime is likely not
an important use case, and that the harware may not even support it at
all, we can probably keep the existing graph and driver implementation.


I suppose in the long term we need to think about how this should be modeled in a multi-context system...when we have a media graph with 8 cameras connected (somehow) to the ISP's single sink pad how should we select the right input device for a context? Whatever the answer is there, probably we should do it here...if we can't do it at runtime with links then probably it has to be a mux...or some novel thing.


The ISP has two source pads, reflecting the
+different paths through which it can internally route data. Tap points within
+the ISP allow users to divert data to avoid processing by some or all of the
+hardware's processing steps. The diagram below is intended only to highlight how
+the bypassing works and is not a true reflection of those processing steps; for
+a high-level functional block diagram see ARM's developer page for the
+ISP [3]_::
+
+  +--------------------------------------------------------------+
+  |                Possible Internal ISP Data Routes             |
+  |          +------------+  +----------+  +------------+        |
+  +---+      |            |  |          |  |  Colour    |    +---+
+  | 0 |--+-->| Processing |->| Demosaic |->|   Space    |--->| 1 |
+  +---+  |   |            |  |          |  | Conversion |    +---+
+  |      |   +------------+  +----------+  +------------+        |
+  |      |                                                   +---+
+  |      +---------------------------------------------------| 2 |
+  |                                                          +---+
+  |                                                              |
+  +--------------------------------------------------------------+
+
+
+.. flat-table::
+    :header-rows: 1
+
+    * - Pad
+      - Direction
+      - Purpose
+
+    * - 0
+      - sink
+      - Data input, connected to the TPG and camera sensors
+
+    * - 1
+      - source
+      - RGB/YUV data, connected to the FR and DS V4L2 subdevices
+
+    * - 2
+      - source
+      - RAW bayer data, connected to the FR V4L2 subdevices
+
+The ISP is limited to both input and output resolutions between 640x480 and
+8192x8192, and this is reflected in the ISP and resizer subdevice's .set_fmt()
+operations.
+
+**mali-c55 resizer fr**
+The `mali-c55 resizer fr` subdevice has two _sink_ pads to reflect the different
+insertion points in the hardware (either RAW or demosaiced data):
+
+.. flat-table::
+    :header-rows: 1
+
+    * - Pad
+      - Direction
+      - Purpose
+
+    * - 0
+      - sink
+      - Data input connected to the ISP's demosaiced stream.
+
+    * - 1
+      - source
+      - Data output connected to the capture video device
+
+    * - 2
+      - sink
+      - Data input connected to the ISP's raw data stream
+
+The data source in use is selected through the routing API; two routes each of a
+single stream are available:
+
+.. flat-table::
+    :header-rows: 1
+
+    * - Sink Pad
+      - Source Pad
+      - Purpose
+
+    * - 0
+      - 1
+      - Demosaiced data route
+
+    * - 2
+      - 1
+      - Raw data route
+
+
+If the demosaiced route is active then the FR pipe is only capable of output
+in RGB/YUV formats. If the raw route is active then the output reflects the
+input (which may be either Bayer or RGB/YUV data).
+
+Using the driver to capture video
+=================================
+
+Using the media controller APIs we can configure the input source and ISP to
+capture images in a variety of formats. In the examples below, configuring the
+media graph is done with the v4l-utils [1]_ package's media-ctl utility.
+Capturing the images is done with yavta [2]_.
+
+Configuring the input source
+----------------------------
+
+The first step is to set the input source that we wish by enabling the correct
+media link. Using the example topology above, we can select the TPG as follows:
+
+.. code-block:: none
+
+    media-ctl -l "'lte-csi2-rx':1->'mali-c55 isp':0[0]"
+    media-ctl -l "'mali-c55 tpg':0->'mali-c55 isp':0[1]"
+
+Capturing bayer data from the source and processing to RGB/YUV
+--------------------------------------------------------------
Missing blank line.

+To capture 1920x1080 bayer data from the source and push it through the ISP's
+full processing pipeline, we configure the data formats appropriately on the
+source, ISP and resizer subdevices and set the FR resizer's routing to select
+processed data. The media bus format on the resizer's source pad will be either
+RGB121212_1X36 or YUV10_1X30, depending on whether you want to capture RGB or
+YUV. The ISP's debayering block outputs RGB data natively, setting the source
+pad format to YUV10_1X30 enables the colour space conversion block.
+
+In this example we target RGB565 output, so select RGB121212_1X36 as the resizer
+source pad's format:
+
+.. code-block:: none
+
+    # Set formats on the TPG and ISP
+    media-ctl -V "'mali-c55 tpg':0[fmt:SRGGB16_1X16/1920x1080]"
+    media-ctl -V "'mali-c55 isp':0[fmt:SRGGB16_1X16/1920x1080]"
+    media-ctl -V "'mali-c55 isp':1[fmt:SRGGB16_1X16/1920x1080]"
+
+    # Set routing on the FR resizer
+    media-ctl -R "'mali-c55 resizer fr'[0/0->1/0[1],2/0->1/0[0]]"
+
+    # Set format on the resizer, must be done AFTER the routing.
+    media-ctl -V "'mali-c55 resizer fr':1[fmt:RGB121212_1X36/1920x1080]"
+
+The downscale output can also be used to stream data at the same time. In this
+case since only processed data can be captured through the downscale output no
+routing need be set:
+
+.. code-block:: none
+
+    # Set format on the resizer
+    media-ctl -V "'mali-c55 resizer ds':1[fmt:RGB121212_1X36/1920x1080]"
+
+Following which images can be captured from both the FR and DS output's video
+devices (simultaneously, if desired):
+
+.. code-block:: none
+
+    yavta -f RGB565 -s 1920x1080 -c10 /dev/video0
+    yavta -f RGB565 -s 1920x1080 -c10 /dev/video1
+
+Cropping the image
+~~~~~~~~~~~~~~~~~~
+
+Both the full resolution and downscale pipes can crop to a minimum resolution of
+640x480. To crop the image simply configure the resizer's sink pad's crop and
+compose rectangles and set the format on the video device:
+
+.. code-block:: none
+
+    media-ctl -V "'mali-c55 resizer fr':0[fmt:RGB121212_1X36/1920x1080 crop:(480,270)/640x480 compose:(0,0)/640x480]"
+    media-ctl -V "'mali-c55 resizer fr':1[fmt:RGB121212_1X36/640x480]"
+    yavta -f RGB565 -s 640x480 -c10 /dev/video0
+
+Downscaling the image
+~~~~~~~~~~~~~~~~~~~~~
+
+Both the full resolution and downscale pipes can downscale the image by up to 8x
+provided the minimum 640x480 resolution is adhered to. For the best image result
Maybe "minimum 640x480 output resolution".

+the scaling ratio for each dimension should be the same. To configure scaling we
s/dimension/direction/

+use the compose rectangle on the resizer's sink pad:
+
+.. code-block:: none
+
+    media-ctl -V "'mali-c55 resizer fr':0[fmt:RGB121212_1X36/1920x1080 crop:(0,0)/1920x1080 compose:(0,0)/640x480]"
+    media-ctl -V "'mali-c55 resizer fr':1[fmt:RGB121212_1X36/640x480]"
+    yavta -f RGB565 -s 640x480 -c10 /dev/video0
+
+Capturing images in YUV formats
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+If we need to output YUV data rather than RGB the color space conversion block
+needs to be active, which is achieved by setting MEDIA_BUS_FMT_YUV10_1X30 on the
+resizer's source pad. We can then configure a capture format like NV12 (here in
+its multi-planar variant)
+
+.. code-block:: none
+
+    media-ctl -V "'mali-c55 resizer fr':1[fmt:YUV10_1X30/1920x1080]"
+    yavta -f NV12M -s 1920x1080 -c10 /dev/video0
+
+Capturing RGB data from the source and processing it with the resizers
+----------------------------------------------------------------------
+
+The Mali-C55 ISP can work with sensors capable of outputting RGB data. In this
+case although none of the image quality blocks would be used it can still
+crop/scale the data in the usual way.
+
+To achieve this, the ISP's sink pad's format is set to
+MEDIA_BUS_FMT_RGB202020_1X60 - this reflects the format that data must be in to
+work with the ISP. Converting the camera sensor's output to that format is the
+responsibility of external hardware.
+
+In this example we ask the test pattern generator to give us RGB data instead of
+bayer.
+
+.. code-block:: none
+
+    media-ctl -V "'mali-c55 tpg':0[fmt:RGB202020_1X60/1920x1080]"
+    media-ctl -V "'mali-c55 isp':0[fmt:RGB202020_1X60/1920x1080]"
+
+Cropping or scaling the data can be done in exactly the same way as outlined
+earlier.
Do we use the ISP's output on pad 1 or pad 2 in this case ? The text
seems to imply that the ISP is bypassed, but the example doesn't mention
any routing change. You may want to clarify this.


Pad 1 for bypassed RGB data. I'll make it more clear.


+
+Capturing raw data from the source and outputting it unmodified
+-----------------------------------------------------------------
+
+The ISP can additionally capture raw data from the source and output it on the
+full resolution pipe only, completely unmodified. In this case the downscale
+pipe can still process the data normally and be used at the same time.
+
+To configure raw bypass the FR resizer's subdevice's routing table needs to be
+configured, followed by formats in the appropriate places:
+
+.. code-block:: none
+
+    # We need to configure the routing table for the resizer to use the bypass
+    # path along with set formats on the resizer's bypass sink pad. Doing this
+    # necessitates a single media-ctl command, as multiple calls to the program
+    # reset the routing table.
Really ?


Yeah

Does -V reset the routing table ? That surprises me.


It's not -V, it's the fact of opening a subdev which calls .init_state(), which resets the routing table...in a single process the fds are held open throughout so all is well, but if you run the program twice they're opened each time and the routing is reset.


+    media-ctl -R "'mali-c55 resizer fr'[0/0->1/0[0],2/0->1/0[1]]"\
+    -V "'mali-c55 isp':0[fmt:RGB202020_1X60/1920x1080],"\
+       "'mali-c55 resizer fr':2[fmt:RGB202020_1X60/1920x1080],"\
+       "'mali-c55 resizer fr':1[fmt:RGB202020_1X60/1920x1080]"
+
+    # Set format on the video device and stream
+    yavta -f RGB565 -s 1920x1080 -c10 /dev/video0
+
+References
+==========
+.. [1] https://git.linuxtv.org/v4l-utils.git/
+.. [2] https://git.ideasonboard.org/yavta.git
+.. [3] https://developer.arm.com/Processors/Mali-C55
diff --git a/Documentation/admin-guide/media/v4l-drivers.rst b/Documentation/admin-guide/media/v4l-drivers.rst
index 4120eded9a13..1d9485860d93 100644
--- a/Documentation/admin-guide/media/v4l-drivers.rst
+++ b/Documentation/admin-guide/media/v4l-drivers.rst
@@ -18,6 +18,7 @@ Video4Linux (V4L) driver-specific documentation
  	ipu3
  	ipu6-isys
  	ivtv
+	mali-c55
  	mgb4
  	omap3isp
  	omap4_camera




[Index of Archives]     [Device Tree Compilter]     [Device Tree Spec]     [Linux Driver Backports]     [Video for Linux]     [Linux USB Devel]     [Linux PCI Devel]     [Linux Audio Users]     [Linux Kernel]     [Linux SCSI]     [XFree86]     [Yosemite Backpacking]


  Powered by Linux