camera not working in olpc3

[dsd]

from the wiki:

gst-launch-0.10 v4l2src ! ffmpegcolorspace ! pngenc ! filesink location=foo.png

The above works in update1 build 708, but not in OLPC3-19.

Firstly, you can only run it as root. Secondly, this error appears (when root):

ERROR: from element /pipeline0/v4l2src0: Device '/dev/video0' cannot capture at 32768x32768
Additional debug info:
v4l2src_calls.c(1227): gst_v4l2src_set_capture (): /pipeline0/v4l2src0:
Call to S_FMT failed for YUYV @ 32768x32768: Invalid argument

We also tried to set the resolution as 160x120, and got a similar error message.

[dsd]

camera works under olpc3 using v4l2 directly, tested using http://www.video4linux.org/browser/v4l2-apps/test/pixfmt-test.c modified to use V4L2_FIELD_NONE (progressive)

[dsd]

gstreamer log

[dsd]

we need to apply this patch (equivalent of what we did for f7)

[dsd]

and /dev/video0 permissions need to be fixed ala #7302

then it sort-of-works (one frame can be captured)

[dgilmore]

Is this patch appropriate for all users? If so then we need to get the patch included upstream. if not then we need to get the patch included upstream with a special configure flag for the use cases where it is appropriate.

So step one is get it upstream. step two is grab from upstream. step three is profit

[dsd]

I'm working on this. I think it may be a kernel bug so I sent a patch to the v4l list, we'll see what happens...

[dsd]

resent the kernel patch, http://marc.info/?l=linux-video&m=121485948506869&w=2

The record activity is broken with the new gstreamer. A gstreamer developer suggested that the pipelines being used were invalid so I'm working on clearing things up. I'm almost done...

[erikb]

Record worked well on fc7.

What is the nature of the pipeline problem? Specifically, which pipeline is broken? Can you post it?

[dsd]

It worked well on F7 on an old gstreamer, but I think it breaks even if you update to F7's latest gstreamer. Let's use #6850 for the record issues, and this ticket for the kernel/gstreamer issue.

[dsd]

As discussed on IRC, due to time constraints we should include the fix (the kernel patch) right now even though we aren't 100% sure it is correct. If we get further insight before the release timeframe is up, we can hopefully fix this properly in time for the release.

Deepak, please include http://marc.info/?l=linux-video&m=121485948506869&w=2 in our kernel. Jon Corbet acked it, but the v4l maintainer (Mauro) has not replied to either of my mails on the topic.

[dsd]

I put the patch in the master and testing branches of my olpc-2.6 git repo: git://dev.laptop.org/users/dsd/olpc-2.6.git

Please pull when convenient!

[dsaxena]

Replying to dsd:

I put the patch in the master and testing branches of my olpc-2.6 git repo: git://dev.laptop.org/users/dsd/olpc-2.6.git Please pull when convenient!

Pulled into olpc-2.6.

[dsd]

I caught Mauro on IRC, who says that the existing ov7670 behaviour is correct, but my patch is correct as well (i.e. drivers can legally work either way). This means that gstreamer should be fixed to be tolerant of this. I'll work on that...

[dsd]

gstreamer patch sent upstream: http://bugzilla.gnome.org/show_bug.cgi?id=541956

[dsd]

gstreamer patch accepted upstream, now for fedora inclusion: https://bugzilla.redhat.com/show_bug.cgi?id=454534

[dsd]

gstreamer-plugins-good.i386 0:0.10.8-8.fc9 includes the userspace fix, so we're all done here.