[c-nsp] Ethernet linkdown detection

[Saku Ytti]

How does exactly does ethernet linkdown work?

I've noticed that if autonego is turned on, you will even notice
unidirectional links and go down in both ends.

Standard is very difficult to follow, many things are not mandatory to
implement and there would be several ways to be standard compliant and
signal this. So not only would you need to be quite familiar in how to read
the standard, you'd also need to know what in practice is done.
Digging authoritative answer without sufficient background seems
impossible.

For sake of simplicity lets assume 10GBase-LR and common practical
implementation, say Cisco 7600, WS-X6704-10GE.

Now here goes what I think is happening, from skimming standard (section4
appears to be the one focusing on 10GE).

a) A side notices signal level dropping below threshold
b) A side asserts LOS, this is communicated over 10PASS-TS (45.2.1.39.5) (A linkdown)
c) 10PASS-TS periodically sends symbols which carry this bit
d) B side sees the 10PASS-TS symbols and asserts flos (45.2.1.40.5) (B linkdown)

Why this does not work without autonego is because 10PASS-TS is not
enabled, unless negotiated?

Now sometimes when you order connectivity from 3rd party this does not
work, why not? Is it because 10PASS-TS is 'peered' instead of 'tunneled'
and signalling information is lost between A-B?

This is probably quite far off from the truth. 10PASS-TS does not exactly
yell '10GE', I don't even get why it is in 45, I would expect this stuff to
be under 52 for 10GBase-LR. Or do all 10G implement 45?
Extra points if specific section in the standard is pointed out for the
correct answers. Would be nice to communicate it when this does not work,
to say 'please work as 42.fix.it.now in IEEE standard says'.

-- 
  ++ytti