Re: [PATCH v3][RFC] add raw video stream support for Samsung SUR40

On 03/07/2015 08:52 PM, Florian Echtler wrote:

On 06.03.2015 12:47, Hans Verkuil wrote:

quoted

On 03/06/2015 12:24 PM, Florian Echtler wrote:

quoted

On 21.02.2015 11:22, Hans Verkuil wrote:

quoted

On 02/20/2015 10:46 PM, Florian Echtler wrote:

quoted

On 16.02.2015 12:40, Hans Verkuil wrote:

quoted

I prefer to dig into this a little bit more, as I don't really understand
it. Set the videobuf2-core debug level to 1 and see what the warnings are.
Since 'buf.qbuf' fails in v4l2-compliance, it's something in the VIDIOC_QBUF
sequence that returns an error, so you need to pinpoint that.

OK, I don't currently have access to the hardware, but I will try this
as soon as possible.

Finally got a chance to try again with videobuf2-core.debug=1. Same
result on 3.19 and 4.0-rc2, after running v4l2-compliance -s from
today's master (full log attached, but important part is below):

I'm not familiar enough with the inner workings of videobuf2 to make any
sense of it, any new insights from you guys?

Can you do:
echo 2 >/sys/class/video4linux/videoX/dev_debug
and run again?
That way I see the vb2 debug messages in related to the issued ioctls.

See attachment, this is the full syslog output from one v4l2-compliance
run on 4.0-rc2, with video0/dev_debug=2 and core.debug=1.

quoted

And if you can also supply the v4l2-compliance -s output, just for
reference?

Also attached for completeness, the important part is:

Streaming ioctls:
	test read/write: OK
	test MMAP: OK
		fail: v4l2-test-buffers.cpp(280): !g_timestamp().tv_sec &&
!g_timestamp().tv_usec

Hmm, I don't think I saw this before.

Anyway, looking at the code I think I see at least one thing that is dubious
and that needs to be changed:

In sur40_process_video() you check for buffers at the start:

	if (list_empty(&sur40->buf_list))
		return;

Replace this with:

	if (!vb2_start_streaming_called(&sur40->queue))
		return;

This will wait until start_streaming was called before it starts processing
video (and start_streaming is only called if at least 3 buffers have been
queued).

Right now the first buffer can be returned without STREAMON actually having
been called. That's certainly wrong.

Whether that is the cause of this bug I do not know, but fix this first.

If this doesn't solve it, then please do another run but this time use

echo 10 >/sys/class/video4linux/videoX/dev_debug

so I see the (D)QBUF ioctls as well. Otherwise use the same procedure as
before.

Thanks!

	Hans