On Thu, 8 Aug 2024 15:00:52 +0200 Alexander Lobakin wrote: > > Technically, neither ART nor TSC are directly related to the PTP cross > > timestamp. It's just the implementation on Intel NICs, where those > > NICs use x86 ART to crosstimestamp. > > > > For cross timestamp on ARM, it's also HW specific and depends on which > > timer the HW uses for timestamping. I'm not really sure what's the HW > > protocol in this case and if e.g. E830 can latch other timers than > > x86 ART in its ART_TIME registers. > > > > get_device_system_crosststamp() supports multiple clock sources defined > > in enum clocksource_ids. Maybe instead of checking ART flag, the driver > > could get clocksources and if CSID_X86_ART is available, it would assign > > the pointer to crosststamp function, but I'm not convinced. > > I mean, I'm fine with the arch-specific definitions in the driver as > long as the netdev maintainers are fine. Or maybe they could propose > some generic solution.
I don't like it either, FWIW, but it seems like this is what everyone is doing. Please do CC tglx / the time maintainers on the next version and net-next submission. I get the feeling they will wake up in a year telling us we did it all wrong, but hey, all we can do now is CC them..
