On 2/17/20 11:06 AM, Miroslav Lichvar wrote:
On Fri, Feb 14, 2020 at 01:23:44PM +0200, Ville Heikkinen wrote:
In case there is no connection to the server, notify the kernel
that the clock is currently unsynchronized.

Is the idea to use the flag for monitoring phc2sys?

The idea is to monitor clock synchronization status: there could be processes running in a container with a limited knowledge of the system. Setting the flag would provide an easy way to those processes achieve this.

I'm not sure if that's how the UNSYNC flag is supposed to be used.
When the clock stops being updated by phc2sys, it doesn't mean it's
"unsynchronized", does it? It will slowly drift away and the kernel
will set the flag automatically when the maximum error (rising at 500
ppm) reaches 16 seconds.

It's true that the kernel sets the flag after a while, but it seems that in some scenarios this takes too long.

It's not that clear that how the interfaces should be used, and the specification is 25 years old. In any case, the interface is specified in RFC 1589. It doesn't give much info how the interface should be used. However, the author of the RFC, Mills, D.L., has written article "Unix kernel modifications for precision time synchronization" (https://www.eecis.udel.edu/~mills/database/reports/kern/kernb.pdf). In there, it's specified that "STA_UNSYNC set/cleared by the caller to indicate clock unsynchronized (e.g., when no peers are reachable)". So, this is the reasoning why the flag could be used like this.

Ville Heikkinen

Linuxptp-devel mailing list

Reply via email to