Hi, On Wed, 2018-12-12 at 13:38 +0100, hverkuil-cisco@xxxxxxxxx wrote: > From: Hans Verkuil <hverkuil-cisco@xxxxxxxxx> > > As was discussed here (among other places): > > https://lkml.org/lkml/2018/10/19/440 > > using capture queue buffer indices to refer to reference frames is > not a good idea. > > Instead, after a long irc discussion: > > https://linuxtv.org/irc/irclogger_log/v4l?date=2018-12-12,Wed > > it was decided to use the timestamp in v4l2_buffer for this. > > However, struct timeval cannot be used in a compound control since > the size of struct timeval differs between 32 and 64 bit architectures, > and there are also changes upcoming for y2038 support. > > But internally the kernel converts the timeval to a u64 (nsecs since > boot). So we provide a helper function in videodev2.h that converts > the timeval to a u64, and that u64 can be used inside compound controls. > > In the not too distant future we want to create a new struct v4l2_buffer, > and then we'll use u64 from the start, so in that case the helper function > would no longer be needed. > > The first three patches add a new m2m helper function to correctly copy > the relevant data from an output buffer to a capture buffer. This will > simplify m2m drivers (in fact, many m2m drivers do not do this quite > right, so a helper function was really needed). > > The fourth patch clears up messy timecode documentation that I came > across while working on this. > > Patch 5 adds the new v4l2_timeval_to_ns helper function to videodev2.h. > The next patch adds the vb2_find_timestamp() function to find buffers > with a specific u64 timestamp. > > Finally the cedrus driver and documentation are updated to use a > timestamp as buffer identifier. > > I also removed the 'pad' fields from the mpeg2 control structs (it > should never been added in the first place) and aligned the structs > to a u32 boundary. > > Please note that this patch series will have to be updated one more > time when pending 4.20 fixes are merged back into our master since > those patches will move the cedrus mpeg controls to a different header. I hit the same build issue that Jonas reported already. With the appropriate fix, this works fine with the cedrus driver. So with the related fix included, this is: Tested-by: Paul Kocialkowski <paul.kocialkowski@xxxxxxxxxxx> Cheers, Paul > Regards, > > Hans > > Hans Verkuil (8): > v4l2-mem2mem: add v4l2_m2m_buf_copy_data helper function > vim2m: use v4l2_m2m_buf_copy_data > vicodec: use v4l2_m2m_buf_copy_data > buffer.rst: clean up timecode documentation > videodev2.h: add v4l2_timeval_to_ns inline function > vb2: add vb2_find_timestamp() > cedrus: identify buffers by timestamp > extended-controls.rst: update the mpeg2 compound controls > > Documentation/media/uapi/v4l/buffer.rst | 11 ++++---- > .../media/uapi/v4l/extended-controls.rst | 28 +++++++++++-------- > .../media/common/videobuf2/videobuf2-v4l2.c | 22 +++++++++++++-- > drivers/media/platform/vicodec/vicodec-core.c | 12 +------- > drivers/media/platform/vim2m.c | 12 +------- > drivers/media/v4l2-core/v4l2-ctrls.c | 9 ------ > drivers/media/v4l2-core/v4l2-mem2mem.c | 20 +++++++++++++ > drivers/staging/media/sunxi/cedrus/cedrus.h | 9 ++++-- > .../staging/media/sunxi/cedrus/cedrus_dec.c | 2 ++ > .../staging/media/sunxi/cedrus/cedrus_mpeg2.c | 21 ++++++-------- > include/media/v4l2-mem2mem.h | 20 +++++++++++++ > include/media/videobuf2-v4l2.h | 19 ++++++++++++- > include/uapi/linux/v4l2-controls.h | 14 ++++------ > include/uapi/linux/videodev2.h | 12 ++++++++ > 14 files changed, 136 insertions(+), 75 deletions(-) > -- Paul Kocialkowski, Bootlin (formerly Free Electrons) Embedded Linux and kernel engineering https://bootlin.com
