Laurent, > While I agree with you, we have a problem here. The V4L2 specification > mandates the use of gettimeofday() to fill the v4l2_buffer::timestamp field: > > http://www.linuxtv.org/downloads/video4linux/API/V4L2_API/spec- > single/v4l2.html#v4l2-buffer > > I'll discuss this with core V4L2 developers. >
Thank you, btw I've tested uvc v4l2 buffer timestamps with gettimeofday and with ktime_get, and taking the first frame as reference, ktime_get proved much more steady. While for gettimeofday (at 30 fps), I got anything from 28 ms to 40 ms between frames in a totally random sequence, for ktime_get, I allways got a steady 32 ms, 32ms, 36ms, 32ms, 32ms, 36ms. The only explanation I could find is that my quad cores have the TSC out of sync, since in a single core machine both methods seemed equally steady. Best regards, Paulo _______________________________________________ Linux-uvc-devel mailing list [email protected] https://lists.berlios.de/mailman/listinfo/linux-uvc-devel
