Do it only if a foreign master is changed to a new one,
do not do it if this is the first foreign master, i.e.
we were not synchronized to any PTP master and the clockIdentity
in the parent DS was our own. In other words, the flag
newGrandmaster is meant to detect change between foreign
masters, and not connection to a foreign master when
before there was none). The "if" was added to prevent
unnecessary trashing between SLAVE and UNCALIBRATED
state (without the if, the UNCALIBRATED would be forced
unnecesserily when connecting the link).

Set the parentDetect flag to NOT_WR_PARENT only if we think
that the parent is WR_PARENT, i.e. thsi flag is not set in
case that parentDetection=PD_NO_DETECTION because that means
we already know whether we are speaking with WR guy or not.
Previously, this condition was parentDetection!=PD_NOT_WR_PARENT
and there was some trashing when WR slave port was speaking
to nonWR port. This is because parentDetection=PD_NO_DETECTION
was set and then parentDetection=PD_NOT_WR_PARENT was re-set
which cause the slave to go back to UNCALIBRATED from SLAVE.

When link goes down, the servo_reset function is called, this is
the perfect place to force the restart of WR link Setup. Yet, this
function is called also when WR link setup times out. In this case,
we do not want to force its re-running. Thus, the doRestart flag is
set only on link down detection.

This reverts commit a3b0cbac.

It was a bad idea in the case when a slave WR port speaks with a
non-WR port. In such case, the servo is not initialized and we
have trashing of states. Next commit will fix the problem in a
different way

If slave link goes down for short time (e.g. executing ifconfig wrX down;
ifconfig wrX up), there is no lost Announce message and there is no
new Parent, however the link down was detected and servo cleaned
(e.g. set to WRH_UNINITIALIZED). Such situation needs to be detected
and "WR link setup" needs to be performed. This change covers this
corner case.

The function servo_reset() depends (at least in WR ext) on variable
ppi->extState that is cleaned in pp_leave_current_state(). Thus,
the servo_reset() function was never called.

This calculation is done using the timestamps but without any correction.

The delayMS in PS was not set in the WR servo.

When a transition of a interface form up->down->up occured, the Slave port was restarted sometimes with the extension OFF.
It has been fixed by re-enabling the extension when a transition state from SLAVE to UNCALIBRATED is detected.

Fix problem related to code optimization

The conversion has been fixed now for negative values.

The detection of a new parent has been moved in a call back of the S1
treatment. The previous implementation had some side effect due to the
fact that the detection was coming too late in some cases (pb with
erbest calculation)

The doRestart called by the servo is now extension specific. It help to
make a more cleaner restart.

The calculation of the erbest is no longer called if
ExternalPortConfiguration is enabled. The erbest is set when a new
foreign master is added.

The S1 treatment must be called only if the foreign master is
accepted/registered by the BMCA.

- In case of handshake failure due to the reception of an unexpected
message, execute the same procedure executed in case of time-out. The
behavior will be more coherent.
- better diagnostics

This version of ppsi, currently submodule of proposed_master of
wr-switch-sw, is not building because of INT_MAX, UINT_MAX and the like.

This includes <limits.h> in all files where this is needed.

Obviously, <limits.h> will be missing in wrpc-sw and bare builds, but
this version does not build for most architectures anyways (including
arch-unix).
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

Bug description : When a MASTER port is at a lower port number than a
SLAVE port, the SLAVE port is jumping all the time between UNCALIBRATED
and SLAVE state.

The 2 octets were displayed as as int16 so they can be swapped depending
of the architecture.

When the PPS generation is forced by configuration, the PPS was disabled
when an instance was leaving the UNCALIB/SLAVE states to another state.

- When a new GM has been detected by the BMCA, the SLAVE state must
transition to UNCALIBRATED. If the state is already set to UNCALIBRATED
, nothing must be done.

- All announce messages not received from erbest are not been taken into
account. erbest is the best qualified foreign master.
- Comments added

Introduction of an other way to calculate the best foreign master. It
has been changed to fix bugs found during ISPC20119 (IOL test). Tests
showed unstable behavior in the computation of foreign master
qualification in some specific circumstances (time window of 8s with
announce messages every 5s)

Issue detected by IOL test in ICALEPS2019
The selection of the ErBest was not taking account all the time of the
qualification.

- New mechanism for clock degradation (degrade clock class, clock
accuracy, clock variance, timeSource, ptpTimeScale, frequencyTraceable,
timeTraceable)
- timeSource can be configured now
- If clock accuracy, clock variance and timeSource are not set in
ppsi.conf, default values are used depending of the clock class.