On Tue, Oct 17 2023 at 18:27, Thomas Gleixner wrote:
> static bool pps_arm_next_pulse(struct pps_tio *tio, ktime_t expires)
> {
> u64 art;
>
> if (!ktime_real_to_base_clock(expires, CSID_X86_ART, &art))
> return false;
>
> pps_compv_write(tio, art - ART_HW_DELAY_CYCLES);
> return true;
> }
>
> ktime_real_to_base_clock() does not exist, but that's the function you
> really want to have.
It just occured to me that CLOCK_REALTIME might not really the best
clock to base this on.
It's obvious why this can't be based on CLOCK_MONOTONIC, but I rather
would base this on CLOCK_TAI which at least does not have the issue of
leap seconds and the related nightmares.
Thanks,
tglx
