hverkuil: hi, I am trying to rivive this patch: https://patchwork.linuxtv.org/patch/60878/
you wrote in the comment that userspace should add a flag V4L2_BUF_FLAG_USE_BOOTTIME to tell the driver to use the boottime clock, but the docs says that the clock should be constant for all buffers in the streaming
also, this will require to change the code of all drivers that are currently default to MONOTONIC?
True. I think you should take a look at the proposed V4L2_FLAG_MEMORY_NON_CONSISTENT flag (Sergey's 'Implement V4L2_BUF_FLAG_NO_CACHE_* flags' series): that adds support to VIDIOC_REQBUFS/CREATE_BUFS to specify flags when creating the buffers (with the restriction that if you use CREATE_BUFS then the flags must be identical to the preceding REQBUFS or CREATE_BUFS).
So if you want to use the boottime clock, then that can be selected there. And v4l2_buffer can then set V4L2_BUF_FLAG_TIMESTAMP_BOOTTIME for all ioctls that use v4l2_buffer.
You'd also need a V4L2_BUF_CAP_SUPPORTS_BOOTTIME to signal support for this.
Main drawback is that Sergey's series isn't in yet, although I hope it can go in for 5.8.
I suggest you add this on top of his v4 series for now.
hverkuil: i'm preparing v7 for rkvdec h264.
grabbed your pull-request, removed your SOB, and addressed feedback from Mauro.
while here, did some minor cleanups and fixed compliance issues.
compliance now passes and the driver looks a bit better, tbh.
(just heads-up)
Great! I hope I have more time to go through posted patch series from tomorrow onwards. It was busy at work this week, but the worst seems to be over (fingers crossed).
No worries.
If this can't make it to 5.7, then I don't see any rush :)
If anything can still make it, well, that'd be nice.
But no worries, really.
ezequielg: regarding "hantro: set of small cleanups and fixes": this is stuff that should go to 5.7, right? I.e. it contains bug fixes.
(was planning to review that tomorrow)
if possible...
specially there's one patch for the core.
https://patchwork.linuxtv.org/patch/62297/
got it.
stevel2: hi :-)
pinchartl: hi! still here?
hverkuil: dafna2: I don't think the boottime patches are future proof, my take is that we should default/support MONOTONIC, cause it's in the spec, but then work are way to expose the clock id, so that any options is possible, DRM expose clock-id, but it's driver decision, and was like this from the start
my feeling is that there is no real use case for per driver, it's more like a system thing