Hi Benjamin,

quoted

In addition, it would make perfect sense to extend the validity of the
hardware time with the event time for longer intervals. The relative
error only makes a difference on milisecond intervals.

A patch that seamlessly extends the validity of the hardware time,
ideally using the event time, seems like a viable solution.

Just to be sure:
When I receive scan_time, I increment timestamp with the device value.
When not, I find out the number of counter wrap I missed with the
kernel timer (jiffies) to get the actual device time.

Is that ok for you?

If the device has no scan time, then no scan time is reported, of
course.  If the scan time delta _and_ the actual time delta are both
below logical_max (say), then increase the counter with the scan time
delta. Else, if the actual time delta is below 2^31, increase the
counter with the actual time delta. Else set the counter to zero.

This ought to give us the sought-after accuracy for small times, the
sought-after extension for intermediate times, and a clear definition
of unknown time.

Cheers,
Henrik