[etherlab-users] Reference clocks and empty configurations

[Gavin Lambert]

Hi,

I've recently been trying to test a bunch of unusual configurations to see
how things behave.  One of these is a configuration where slaves are present
on the network, but none have been added to any domains.  (It seems like a
reasonable interim bring-up state.)

In this case, if I call ecrt_master_activate and run a realtime loop
(dispatching internal packets but not actually processing any domains) then
the first slave on the network will log a "Sync manager watchdog" error
frequently (about twice per second, at least with my test setup), because
the master keeps putting it into OP but is not actually sending any output
data to it.

This appears to be a collision between automatically selecting the first
DC-capable slave on the network as the reference clock and
ec_master_request_op requiring the reference slave to be brought to OP even
if it's not configured.  Is this expected behaviour?  In particular,

1. Why does the reference clock need to be forced to OP?  DC sync registers
etc still work perfectly fine in any state, AFAIK.  I suppose there might be
something special needed for some special kinds of clocks (eg. one that has
external syncing) but presumably the master app would config and/or select
them explicitly in that case.

2. Why does the auto-selection select a slave that is not configured?  Since
unconfigured devices may not be able to go to OP at all, this seems like a
conflict with the prior requirement.  Shouldn't it pick the first present
*and configured* slave that supports DC?

3. Why does it want to have a reference clock at all in a network with no
configured slaves, even when slaves are present?

Regards,
Gavin Lambert