On 21/01/18 18:29, jacopo mondi wrote:
> Hi Hans,
>
> On Sun, Jan 21, 2018 at 11:23:12AM +0100, Hans Verkuil wrote:
>> On 21/01/18 11:21, Hans Verkuil wrote:
>>> On 21/01/18 10:53, jacopo mondi wrote:
>>>> Hi Hans,
>>>>
>>>> On Fri, Jan 19, 2018 at 12:20:19PM +0100, Hans Verkuil wrote:
>>>>> static int ov7670_g_parm(struct v4l2_subdev *sd, struct v4l2_streamparm *parms)
>>>>> {
>>>>> struct v4l2_captureparm *cp = &parms->parm.capture;
>>>>> struct ov7670_info *info = to_state(sd);
>>>>>
>>>>> if (parms->type != V4L2_BUF_TYPE_VIDEO_CAPTURE)
>>>>> return -EINVAL;
>>>>>
>>>>> And parms->type is CAPTURE_MPLANE. Just drop this test from the ov7670 driver
>>>>> in the g/s_parm functions. It shouldn't test for that since a subdev driver
>>>>> knows nothing about buffer types.
>>>>>
>>>>
>>>> I will drop that test in an additional patch part of next iteration of this series,
>>>
>>> Replace g/s_parm by g/s_frame_interval. Consider g/s_parm for subdev drivers as
>>> deprecated (I'm working on a patch series to replace all g/s_parm uses by
>>> g/s_frame_interval).
>>
>> Take a look here:
>>
>> https://git.linuxtv.org/hverkuil/media_tree.git/log/?h=parm
>>
>> You probably want to use the patch 'v4l2-common: add g/s_parm helper functions'
>> for the new ceu driver in your patch series. Feel free to add it.
>
> Thanks, I have now re-based my series on top of your 'parm' branch,
> and now I have silenced those errors on bad frame interval.
>
> CEU g/s_parm now look like this:
>
> static int ceu_g_parm(struct file *file, void *fh, struct v4l2_streamparm *a)
> {
> struct ceu_device *ceudev = video_drvdata(file);
> int ret;
>
> ret = v4l2_g_parm(V4L2_BUF_TYPE_VIDEO_CAPTURE_MPLANE,
> ceudev->sd->v4l2_sd, a);
> if (ret)
> return ret;
>
> a->parm.capture.readbuffers = 0;
>
> return 0;
> }
>
> Very similar to what you've done on other platform drivers in this
> commit:
> https://git.linuxtv.org/hverkuil/media_tree.git/commit/?h=parm&id=a58956ef45cebaa5ce43a5f740fe04517b24a853
>
> I have a question though (please bear with me a little more :)
> I had to manually set a->parm.capture.readbuffers to 0 to silence the following
> error in v4l2_compliance (which I have now updated to the most recent
> remote HEAD):
>
> fail: v4l2-test-formats.cpp(1114): cap->readbuffers
> test VIDIOC_G/S_PARM: FAIL
>
> fail_on_test(cap->readbuffers > VIDEO_MAX_FRAME);
> if (!(node->g_caps() & V4L2_CAP_READWRITE))
> fail_on_test(cap->readbuffers);
> else if (node->g_caps() & V4L2_CAP_STREAMING)
> fail_on_test(!cap->readbuffers);
>
> CEU does not support CAP_READWRITE, as it seems atmel-isc/isi do not, so
> v4l2-compliance wants to have readbuffers set to 0. I wonder why in
> the previously mentioned commit you didn't have to set readbuffers
> explicitly to 0 for atmel-isc/isi as I had to for CEU. Will v4l2-compliance
> fail if run on atmel-isc/isi with your commit, or am I missing something?
I've reworked the g/s_parm helper functions so they will now check for
the READWRITE capability and set readbuffers accordingly. I'll post a new
version later today.
Thanks for testing this, I missed that corner case.
Regards,
Hans
