[PATCH 2/3] media: coda: Add driver for Coda video codec.

[Philipp Zabel]

Am Montag, den 09.07.2012, 10:07 +0200 schrieb javier Martin:
[...]
> >> +static int vidioc_s_parm(struct file *file, void *priv, struct v4l2_streamparm *a)
> >> +{
> >> +     struct coda_ctx *ctx = fh_to_ctx(priv);
> >> +
> >> +     if (a->type == V4L2_BUF_TYPE_VIDEO_OUTPUT) {
> >> +             if (a->parm.output.timeperframe.numerator != 1) {
> >> +                     v4l2_err(&ctx->dev->v4l2_dev,
> >> +                              "FPS numerator must be 1\n");
> >> +                     return -EINVAL;
> >> +             }
> >> +             ctx->params.framerate =
> >> +                                     a->parm.output.timeperframe.denominator;
> >> +     } else {
> >> +             v4l2_err(&ctx->dev->v4l2_dev,
> >> +                      "Setting FPS is only possible for the output queue\n");
> >> +             return -EINVAL;
> >
> > Why disallow setting timeperframe on the capture side? Shouldn't it
> > rather succeed without setting anything and return the current context's
> > frame rate instead?
> >
> >> +     }
> >> +     return 0;
> >> +}
> >> +
> >> +static int vidioc_g_parm(struct file *file, void *priv, struct v4l2_streamparm *a)
> >> +{
> >> +     struct coda_ctx *ctx = fh_to_ctx(priv);
> >> +
> >> +     if (a->type == V4L2_BUF_TYPE_VIDEO_OUTPUT) {
> >> +             a->parm.output.timeperframe.denominator =
> >> +                                     ctx->params.framerate;
> >> +             a->parm.output.timeperframe.numerator = 1;
> >> +     } else {
> >> +             v4l2_err(&ctx->dev->v4l2_dev,
> >> +                      "Getting FPS is only possible for the output queue\n");
> >
> > The nominal capture side timeperframe should match that of the output
> > side.
> >
> > Actually, I'm not sure if this needs to be implemented at all, since
> > V4L2_CAP_TIMEPERFRAME is not set and capture frame dropping / output
> > frame duplication is not supported.
> 
> I am just following the steps of Samsung here:
> 
> http://lxr.linux.no/#linux+v3.4.4/drivers/media/video/s5p-mfc/s5p_mfc_enc.c#L1439
> http://lxr.linux.no/#linux+v3.4.4/drivers/media/video/s5p-mfc/s5p_mfc_opr.c#L775
> 

I don't think this is completely correct either. According to the V4L2
spec, setting timeperframe from an application is meant to make the
driver skip or duplicate frames to save bandwidth:

http://v4l2spec.bytesex.org/spec-single/v4l2.html#VIDIOC-G-PARM

So returning -EINVAL is not necessarily incorrect, as we can choose not
to support this ioctl - but claiming support for the output side (and
then doing nothing) but returning an error on the capture side seems a
bit inconsistent to me.

regards
Philipp