Bug#855834: linuxptp: default startup argument "-i eth0" should be removed

[Tino Mettler]

On Wed, Feb 22, 2017 at 09:51:25 -0300, Felipe Sateler wrote:

[...]

> The readme suggests ptp4l can detect appropiate devices by itself. If
> that is true, then there is no problem to be solved :).

Hi,

do you mean this?

---------------------------------------------------------------------
   If the ethtool ioctl is available, then the ptp4l program will use
   it in order to discover the proper PHC device.
---------------------------------------------------------------------

This means that ptp4l can find the proper PHC device (/dev/ptpX) that
belongs to a certain ethernet interface (like eth0).  Before that, the
user had to specify both the ethernet interface (-i) and the PHC device
(-p) to use.

The ethernet interface still has to be specified either on the command
line (-i option) or in the config file.

>From the manual page:

--------------------------------------------------------------------
       -i interface
              Specify  a PTP port, it may be used multiple times. At
              least one port must be specified by this option or in
              the configuration file.
--------------------------------------------------------------------


> 
> If that is not true, I suggest the following:
> 
> 1. Convert ptp4l into a template unit, ptp4l at .service
> 2. Change the device to be the instance:
> ExecStart=ExecStart=/usr/sbin/ptp4l -f /etc/linuxptp/ptp4l.conf -i %i
> 3. Do not enable the unit.
> 4. Add a udev rule that starts the (properly instanced) service when
> it is detected.
> 
> You can see the ifupdown package for a similar approach: there is
> ifup at .service, a udev rule, and a helper program (ifupdown-hotplug)
> that determines if an instance should be started for the given device.

Thanks. I don't fully understand 4. What exactly should be detected,
and how? 

> While I looked at the service files, I noticed you order them after
> chrony, ntpdate and other time services. Systemd defines a standard
> place for that, so you can replace all those alternatives with
> `time-sync.target`.

Thanks for the suggestions. Currently, I just look how Fedora sets up
the services and do the same, so I'm always open for suggestions how to
improve them.

Regards,
Tino