Re: [Linuxptp-users] Jump of ~UTC-TAI offset after taking the interface down and up

[FUSTE E. <emm...@th...>]

Le 21/08/2018 à 14:32, Richard Cochran a écrit :
> On Mon, Aug 20, 2018 at 01:28:25PM +0000, FUSTE Emmanuel wrote:
>> I think the big jump is because the PHC was stopped and completely reset
>> between a down/up sequence.
> Right.  The driver resets the time to the system time value (which is UTC).
>
>> Is it an expected behavior ?
> For ptp4l?  Yes.
>
> For the ifdown/ifup of the driver?  I would call it a driver bug.  It
> is completely illogical that that bringing the interface down and up
> again would change the stored time.
>
> I have complained about this on netdev before, but to no avail.  There
> drivers that do all of their bringup/teardown in their .open/close
> callbacks and *not* in the .probe/remove methods.  It may be that
> the HW forces implementation in some cases.
>
> But in at least one case, where *I* wrote a perfectly adequate PHC
> driver that sets the clock exactly once, in probe/remove, the $corp
> developers when ahead and moved that into open/close.  Brilliant.
>
>> Could we do something on the ptp4l side to
>> address this case ?
> Yes, you can use the 'step_threshold' option.
>
> Thanks,
> Richard
>
Thank you for the detailed explanation.

Emmanuel.