<!-- Some styling for better description lists --><style type='text/css'>dt { font-weight: bold;float: left;display:inline;margin-right: 1em} dd { display:block; margin-left: 2em}</style>

   gnurou: <u>hverkuil</u>: I am testing v4l2-compliance with codec devices - is there a way to enforce the order in which formats are set on the output and capture queues? Codec API requires the output queue to be set first, which is the opposite of the current behavior
   <br> s/v4l2-compliance/v4l2-ctl
   hverkuil: <u>gnurou</u>: I'll try to answer later today. It's a bit busy this morning.
   gnurou: no problem!
   hverkuil: In general support for codecs is limited in both v4l2-ctl and v4l2-compliance. It's hard(er) to test and I never had good codec hardware to test with.
   gnurou: I may be interested in improving support, if you think that would make a worthy contribution
   ***: jeremymc has quit IRC (Read error: Connection reset by peer)
   <br> jeremymc1 has quit IRC (Read error: Connection reset by peer)
   svarbanov: <u>gnurou</u>: I have some ideas to improve v4l2-compliance in regards to codecs and streaming test, so I'm interested too :)
   gnurou: <u>svarbanov</u>: sounds great!
   <br> basically I'd like to make sure we can ensure compliance against the codec API
   <br> (and later, add request API tests for stateless codecs)
   hverkuil: <u>svarbanov</u>: gnurou: I have no plans to work on this, at least not for the near future. But I can give pointers/advice of course.
   svarbanov: <u>gnurou</u>: but we have to have codec API in media documentation first :)
   <br> <u>gnurou</u>: hverkuil will be happy to review documentation ;)
   hverkuil: My experience that it is best to do documentation and writing tests at the same time. It is very effective since you make a test for a nasty corner case and then realize that the spec doesn't mention what should happen in that case.
   neg: <u>hverkuil</u>: I have a question about v4l-utils commit ef45319c1686088a ("v4l2-compliance: VIDIOC_S_EDID and invalid pointers"). Is there any driver I can look at wich correctly handles the test for edid.edid = (__u8 *)0xdeadbeef?
   hverkuil: This check is done in v4l2-ioctl.c. So drivers shouldn't have to care about this. This test is meant to exercise the core code, not the drivers.
   neg: <u>hverkuil</u>: OK thanks, is that check recently added? I hit this on v4.16-rc1
   hverkuil: No, it's around for some time. How does it fail for you?
   neg: <u>fail</u>: v4l2-test-io-config.cpp(593): ret != EFAULT
   <br> <u>fail</u>: v4l2-test-io-config.cpp(629): EDID check failed for input 0.
   <br> test VIDIOC_G/S_EDID: FAIL
   <br> <u>hverkuil</u>: or rather on v4.16-rc1 I hit the edid-&gt;edid = 0 check, but if I handle that in the driver I hit the 0xdeadbeef check
   hverkuil: <u>neg</u>: can you run with the -T option and see what the returned error code is?
   <br> <u>neg</u>: which subdev driver handles the EDID ioctl?
   neg: VIDIOC_S_EDID returned 0 (Success)
   <br> adv7604 (I do the tests on Koelsch)
   hverkuil: Is this test done from a v4l-subdevX node or from a video node?
   <br> hmm, should not matter actually.
   neg: <u>hverkuil</u>: video node, v4l2-compliance -T -d /dev/video25
   hverkuil: video_usercopy() in v4l2-ioctl.c should generate the EFAULT.
   <br> See line 2895-7:
   <br>                err = -EFAULT;
   <br>                if (copy_from_user(mbuf, user_ptr, array_size))
   <br>                        goto out_array_args;
   <br> If you can debug this a bit, then that would be great.
   neg: I will, thanks for your help
   ***: ivanspa has left "Leaving"
   neg: <u>hverkuil</u>: OK so I found the issue but not sure how to best resolve it
   <br> <u>hverkuil</u>: problem is tha v4l2-compability calls VIDIOC_S_EDID with edid.edid = 0xdeadbeef and edid.blocks = 0
   <br> edid.blocks = 0 inflicts the check in check_array_args() for VIDIOC_S_EDID to tell video_usercopy() that the user have not provided arguments so copy_from_user() is not called and the error slips thrue
   <br> Accordig to the v4l2 documentation it's valid to call S_EDID with blocks = 0, 'When you set the EDID and blocks is 0, then the EDID is disabled or erased.'
   <br> So I'm not sure how to best solve this, make sure v4l2-compliance have blocks &gt; 0 or try to add more checks to v4l2, what do you think is the correct solution?
   <br> And you where correct if I in v4l2-complicance set edid.blocks = 1 the v4l2 framework catches and handles both the errors I hit erlier today (edid.edid = NULL and edid.edidi = 0xdeadbeef)
   <br> And one last note, if I program the EDID prior to running v4l2-compliance (v4l2-ctl -d .. --set-edid=type=hdmi) the compliance tests works
   ***: benjiG has left