Thread (24 messages) 24 messages, 7 authors, 2021-05-25

Re: [PATCH 00/23] media: imx: imx7-mipi-csis: Add i.MX8MM support / imx8mq support

From: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Date: 2021-05-25 11:25:32
Also in: linux-media

Hi Martin,

On Tue, May 25, 2021 at 09:32:54AM +0200, Martin Kepplinger wrote:
Am Freitag, dem 21.05.2021 um 16:36 +0300 schrieb Laurent Pinchart:
quoted
On Fri, May 21, 2021 at 01:02:30PM +0200, Martin Kepplinger wrote:
quoted
Am Freitag, dem 21.05.2021 um 12:43 +0300 schrieb Laurent Pinchart:
quoted
On Fri, May 21, 2021 at 11:25:20AM +0200, Martin Kepplinger wrote:
[snip]
quoted
quoted
quoted
quoted
media-ctl --set-v4l2 "'csi':0 [fmt:SGBRG10/640x480 colorspace:raw]"
media-ctl --set-v4l2 "'imx8mq-mipi-csi2.0':0 [fmt:SGBRG10/640x480 colorspace:raw]"
media-ctl -l "'hi846 2-0020':0 -> 'imx8mq-mipi-csi2.0':0 [1]"

which gets me:

Device topology
- entity 1: csi (2 pads, 2 links)
            type V4L2 subdev subtype Unknown flags 0
            device node name /dev/v4l-subdev0
        pad0: Sink
                [fmt:SGBRG10_1X10/640x480 field:none colorspace:raw xfer:none ycbcr:601 quantization:full-range]
                <- "imx8mq-mipi-csi2.0":1 [ENABLED,IMMUTABLE]
        pad1: Source
                [fmt:SGBRG10_1X10/640x480 field:none colorspace:raw xfer:none ycbcr:601 quantization:full-range]
                -> "csi capture":0 [ENABLED,IMMUTABLE]

- entity 4: csi capture (1 pad, 1 link)
            type Node subtype V4L flags 0
            device node name /dev/video0
        pad0: Sink
                <- "csi":1 [ENABLED,IMMUTABLE]

- entity 10: imx8mq-mipi-csi2.0 (2 pads, 2 links)
             type V4L2 subdev subtype Unknown flags 0
             device node name /dev/v4l-subdev1
        pad0: Sink
                [fmt:SGBRG10_1X10/640x480]
                <- "hi846 2-0020":0 [ENABLED]
        pad1: Source
                [fmt:SGBRG10_1X10/640x480]
                -> "csi":0 [ENABLED,IMMUTABLE]

- entity 15: hi846 2-0020 (1 pad, 1 link)
             type V4L2 subdev subtype Sensor flags 0
             device node name /dev/v4l-subdev2
        pad0: Source
                [fmt:SGBRG10_1X10/640x480 field:none colorspace:raw]
                -> "imx8mq-mipi-csi2.0":0 [ENABLED]
This looks better.
quoted
but streaming still fails with:

[  352.255129] imx7-csi 30a90000.csi1_bridge: media bus code not
compatible with the pixel format set on the video node: 1 != 0
What is the capture command line ? Can you trace this (I assume the
message is printed by capture_validate_fmt(), it's not present in
mainline so I don't know what 1 and 0 correspond to, even though I
suspect they would be IPUV3_COLORSPACE_* values) to see why it fails
?
capture command:

v4l2-ctl -d "/dev/v4l/by-path/platform-30a90000.csi1_bridge-video-
index0" --set-fmt-video=width=640,height=480 --stream-mmap
--stream-to=test.raw --stream-count=1

I'll have to continue after the weekend, but let's share some logs.
Yes, "1 != 0" is from capture_validate_fmt():

priv->vdev.cc->cs != cc->cs

When I print the format the imx_media_find_mbus_format() finds and
do:

media-ctl --set-v4l2 "'csi':0 [fmt:SGBRG10/640x480 colorspace:raw]"

I see:

[  184.251144] mc: media_release: Media Release
[  184.254397] selected specific mbus code 0 for list nr 0 (fourcc 0x59565955)
[  184.264564] selected specific mbus code 0 for list nr 0 (fourcc 0x59565955)
[  184.274763] selected specific mbus code 0 for list nr 21 (fourcc 0x36314247)
[  184.285102] selected specific mbus code 0 for list nr 21 (fourcc 0x36314247)
[  184.295383] selected specific mbus code 0 for list nr 21 (fourcc 0x36314247)
[  184.305752] selected specific mbus code 0 for list nr 21 (fourcc 0x36314247)

21 is the correct bayer format I want, but there's 0, so
"MEDIA_BUS_FMT_UYVY8_2X8" found the first 2 times. That is
IPUV3_COLORSPACE_YUV (1) while the correct Bayer format 21 is
IPUV3_COLORSPACE_RGB (0).

so some format settings not yet correct.
You need to specify the capture pixel format to v4l2-ctl. The driver
defaults to YUYV (I think) otherwise. The CSI bridge will pad data with
0's on the right, so you need SGBRG16 (if I recall correctly, try
SGRBG10 if it doesn't work).
right, now with:

v4l2-ctl -d "/dev/v4l/by-path/platform-30a90000.csi1_bridge-video-
index0" --set-fmt-video=width=640,height=480,pixelformat='GB16'
--stream-mmap --stream-to=test.raw --stream-count=1

indeed the colorspace mismatch is gone.

__media_pipeline_start returns broken pipe because of:
link validation failed for 'hi846 2-0020':0 -> 'imx8mq-mipi-csi2.0':0, error -32:

imx7-csi 30a90000.csi1_bridge: walk: returning entity 'hi846 2-0020'
imx7-csi 30a90000.csi1_bridge: walk: returning entity 'imx8mq-mipi-csi2.0'
hi846 2-0020: hi846: starting hi846_get_format
hi846 2-0020: Get format w=640 h=480 code=0x300e colorspace=0xb
imx7-csi 30a90000.csi1_bridge: v4l2_subdev_link_validate_default: field does not match (source 1, sink 0)
imx7-csi 30a90000.csi1_bridge: v4l2_subdev_link_validate_default: link was "hi846 2-0020":0 -> "imx8mq-mipi-csi2.0":0
imx7-csi 30a90000.csi1_bridge: link validation failed for 'hi846 2-0020':0 -> 'imx8mq-mipi-csi2.0':0, error -32
...
imx7-csi 30a90000.csi1_bridge: pipeline start failed with -32


then I added the init_cfg() callback (that sets sink field to NONE) and
I could stream frames.
Nice :-)
So the v4l2/mc side of things looks ok, despite the one thing:

Also, now when saying "--list-formats-ext" on the video node, it lists
all formats that the mipi driver supports (since it doesn't ask the
sensor anymore about that). That seems to be wrong.
That's expected. You can restrict the enumerated formats to the ones
matching a given media bus code by setting the mbus_code field of struct
v4l2_fmtdesc ([1]).

[1] https://linuxtv.org/downloads/v4l-dvb-apis/userspace-api/v4l/vidioc-enum-fmt.html#c.V4L.v4l2_fmtdesc
Also, the frames are all black, so I'll need to debug the driver side
of things.
Indeed :-)
But I again append the current driver in case you want to have a look.
I'm afraid I don't have time right now, but I'll try to review it when
you'll post the patches.

It'd be curious to try the libcamera simple pipeline handler based on
that code.
thanks so much for walking through this with me :) I won't forget the
dt properties and upstreaming - after I get correct frames.
-- 
Regards,

Laurent Pinchart
Keyboard shortcuts
hback out one level
jnext message in thread
kprevious message in thread
ldrill in
Escclose help / fold thread tree
?toggle this help