[Linuxptp-users] Loss of sync with 25% of CPU load

[Diego G. P. <gpr...@un...>]

Hello everyone,

I have a query that maybe you can solve.

I use a switch as a GM and a board as a slave with ptp4l and phc2sys. 
They work perfectly with good offsets (-20ns,20ns) but when I force 
load, just 25%, they lose synchronization. Can you help me?


ptp4l[742.892]: rms    4 max    7 freq +16816 +/-   5 delay   235 +/-   0
phc2sys[743.863]: CLOCK_REALTIME phc offset        31 s2 freq +10667 
delay   4908
ptp4l[743.893]: rms    3 max    5 freq +16815 +/-   4 delay   235 +/-   0
phc2sys[744.864]: CLOCK_REALTIME phc offset         0 s2 freq +10645 
delay   4932
ptp4l[744.893]: rms    3 max    6 freq +16813 +/-   4 delay   234 +/-   0
phc2sys[745.864]: CLOCK_REALTIME phc offset       -62 s2 freq +10583 
delay   4896
ptp4l[745.894]: rms    3 max    5 freq +16814 +/-   4 delay   234 +/-   0
phc2sys[746.864]: CLOCK_REALTIME phc offset        47 s2 freq +10673 
delay   4896
ptp4l[746.895]: rms    3 max    6 freq +16812 +/-   4 delay   234 +/-   0
ptp4l[747.473]: clockcheck: clock jumped backward or running slower than 
expected!
ptp4l[747.473]: port 1 (enp3s0): SLAVE to UNCALIBRATED on 
SYNCHRONIZATION_FAULT
ptp4l[747.473]: port 1 (enp3s0): UNCALIBRATED to SLAVE on 
MASTER_CLOCK_SELECTED
phc2sys[748.022]: port aabbcc.fffe.00094e-1 changed state
phc2sys[748.022]: port aabbcc.fffe.00094e-1 changed state
ptp4l[748.438]: port 1 (enp3s0): SLAVE to MASTER on 
ANNOUNCE_RECEIPT_TIMEOUT_EXPIRES
ptp4l[748.438]: selected local clock aabbcc.fffe.00094e as best master
phc2sys[748.438]: port aabbcc.fffe.00094e-1 changed state
ptp4l[748.438]: port 1 (enp3s0): assuming the grand master role
phc2sys[748.438]: reconfiguring after port state change
ptp4l[748.438]: clockcheck: clock jumped backward or running slower than 
expected!
phc2sys[748.438]: selecting enp3s0 for synchronization
phc2sys[748.438]: selecting CLOCK_REALTIME as the master clock
phc2sys[748.438]: enp3s0 sys offset        49 s0 freq  +16813 delay   4943
ptp4l[748.645]: clockcheck: clock jumped forward or running faster than 
expected!
ptp4l[749.123]: selected best master clock 00049f.fffe.ef0808
ptp4l[749.123]: port 1 (enp3s0): MASTER to UNCALIBRATED on RS_SLAVE
phc2sys[749.438]: port aabbcc.fffe.00094e-1 changed state
phc2sys[749.439]: reconfiguring after port state change
phc2sys[749.439]: master clock not ready, waiting...
ptp4l[749.646]: port 1 (enp3s0): UNCALIBRATED to SLAVE on 
MASTER_CLOCK_SELECTED
ptp4l[749.772]: rms    3 max    6 freq +16802 +/-  22 delay   234 +/-   0
phc2sys[750.439]: port aabbcc.fffe.00094e-1 changed state
phc2sys[750.440]: reconfiguring after port state change
phc2sys[750.440]: selecting CLOCK_REALTIME for synchronization
phc2sys[750.440]: selecting enp3s0 as the master clock
phc2sys[750.440]: CLOCK_REALTIME phc offset      -211 s0 freq +10673 
delay   4896
ptp4l[750.772]: rms   18 max   24 freq +16796 +/-  13 delay   234 +/-   0
phc2sys[751.440]: CLOCK_REALTIME phc offset      -252 s2 freq +10600 
delay   4908
ptp4l[751.773]: rms   20 max   26 freq +16820 +/-   6 delay   234 +/-   0
phc2sys[752.850]: CLOCK_REALTIME phc offset      -219 s2 freq +10381 
delay   4872
ptp4l[752.850]: port 1 (enp3s0): SLAVE to MASTER on 
ANNOUNCE_RECEIPT_TIMEOUT_EXPIRES
ptp4l[752.850]: selected local clock aabbcc.fffe.00094e as best master
ptp4l[752.851]: port 1 (enp3s0): assuming the grand master role
ptp4l[752.852]: clockcheck: clock jumped backward or running slower than 
expected!
ptp4l[753.023]: clockcheck: clock jumped forward or running faster than 
expected!
ptp4l[753.123]: selected best master clock 00049f.fffe.ef0808
ptp4l[753.124]: port 1 (enp3s0): MASTER to UNCALIBRATED on RS_SLAVE
phc2sys[753.850]: port aabbcc.fffe.00094e-1 changed state
phc2sys[753.851]: port aabbcc.fffe.00094e-1 changed state
phc2sys[753.851]: reconfiguring after port state change
phc2sys[753.851]: master clock not ready, waiting...


In the log, as you can see, it works well until I run the CPU load. Here 
is the command I run with stress-ng: stress-ng --cpu 4 --cpu-load 25 
-sched fifo --sched-prio 99 --times

ptp4l and phc2sys are fifo priority 99 as well.


Do you know how to solve it? Ask me whatever you want.


Kind regards,

Diego