On Sat September 15 2012 22:16:24 Sylwester Nawrocki wrote:
> On 09/15/2012 02:35 PM, Hans Verkuil wrote:
> >>>> If we switch all existing drivers to monotonic timestamps in kernel
> >>>> release
> >>>> 3.x, v4l2-compliance can just use the version it gets from
> >>>> VIDIOC_QUERYCAP and
> >>>> enforce monotonic timestamps verification if the version is>= 3.x. This
> >>>> isn't
> >>>> more difficult for apps to check than a dedicated flag (although it's
> >>>> less
> >>>> explicit).
> >>>
> >>> I think that checking for the driver (kernel) version is a very poor
> >>> substitute
> >>> for testing against a proper flag.
> >>
> >> That flag should be the default in this case. The flag should be set by
> >> the framework instead giving every driver the job of setting it.
> >>
> >>> One alternative might be to use a v4l2_buffer flag instead. That does
> >>> have the
> >>> advantage that in the future we can add additional flags should we need to
> >>> support different clocks. Should we ever add support to switch clocks
> >>> dynamically,
> >>> then a buffer flag is more suitable than a driver capability. In that
> >>> scenario
> >>> it does make real sense to have a flag (or really mask).
> >>>
> >>> Say something like this:
> >>>
> >>> /* Clock Mask */
> >>> V4L2_BUF_FLAG_CLOCK_MASK 0xf000
> >>> /* Possible Clocks */
> >>> V4L2_BUF_FLAG_CLOCK_SYSTEM 0x0000
> >
> > I realized that this should be called:
> >
> > V4L2_BUF_FLAG_CLOCK_UNKNOWN 0x0000
> >
> > With a comment saying that is clock is either the system clock or a
> > monotonic
> > clock. That reflects the current situation correctly.
> >
> >>> V4L2_BUF_FLAG_CLOCK_MONOTONIC 0x1000
>
> There is already lots of overhead related to the buffers management, could
> we perhaps have the most common option defined in a way that drivers don't
> need to update each buffer's flags before dequeuing, only to indicate the
> timestamp type (other than flags being modified in videobuf) ?
Well, if all vb2 drivers use the monotonic clock, then you could do it in
__fill_v4l2_buffer: instead of clearing just the state flags you'd clear
state + clock flags, and you OR in the monotonic flag in the case statement
below (adding just a single b->flags |= line in the DEQUEUED case).
So that wouldn't add any overhead. Not that I think setting a flag will add
any measurable overhead in any case.
> This buffer flags idea sounds to me worse than the capability flag. After
> all the drivers should use monotonic clock timestamps, shouldn't they ?
Yes. But you have monotonic and raw monotonic clocks at the moment, and perhaps
others will be added in the future. You can't change clocks if you put this
in the querycap capabilities.
> Have anyone has ever come with a use case for switching timestamps clock
> type, can anyone give an example of it ? How likely is we will ever need
> that ?
Well, ALSA allows you to switch between gettimeofday and monotonic. So in
theory at least if an app selects gettimeofday for alsa, that app might also
want to select gettimeofday for v4l2.
I'd really like to keep this door open. My experience is that if something is
possible, then someone somewhere will want to use it.
Regards,
Hans
--
To unsubscribe from this list: send the line "unsubscribe linux-media" in
the body of a message to majord...@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
