From: Vincent Cheng <vincent.cheng...@renesas.com> When clock stepping is unable to happen instantaneously the subsequent timestamps after a clock step does not reflect the step result and undesired clock freq and clock steps would occur.
When using ts2phc to synchronize timestamping clock using external 1 PPS, it could take up to 1 second for the timestamps to reflect the clock step. step_window, when set, indicates the number of Sync events after a clock step in which the clock servo will not do any frequency or step adjustments. Below example illustrates the problem for 16 PPS when clock step does not occur before the next set of timestamps are received. Debug statements were added to show T1 and T2 timestamps and the freq and step requests at clock_sychronize() for SERVO_JUMP. ptp4l[255352.651]: selected best master clock 00b0ae.fffe.02e810 ptp4l[255352.651]: port 1: LISTENING to UNCALIBRATED on RS_SLAVE ptp4l[255352.717]: debug: T1: 1611934788908411436 T2: 10341336582 ... ptp4l[255352.904]: master offset -1611934778567080326 s0 freq -128 path delay 5436 ptp4l[255352.967]: debug: T1: 1611934789158411436 T2: 10591336566 ptp4l[255352.967]: debug: adj freq -159.999967232 ptp4l[255352.971]: debug: step 1611934778567080308 ... ptp4l[255353.217]: debug: T1: 1611934789408411436 T2: 10841336502 ptp4l[255353.217]: debug: adj freq 0.000027648 ptp4l[255353.221]: debug: step 1611934778567080368 At 16 PPS, the packet interval is 0.0625 seconds. The first step occurs at [255352.971], T2 is around 10 seconds. The next step occurs at [255353.221], T2 is still around 10 seconds. In an ideal setup, the clock step would be reflected instantaneously and the correct T2 should be around 1611934799 seconds. Below shows result of adding step_window. Clock step occurs at s1, SERVO_JUMP. s2 is SERVO_LOCKED. The setup is using ts2phc to synhronize the network PHC to external time stamp signal. The progation delay of the 1 PPS signals have a worst case of 2 seconds. step_window is set to 0 (default 0), so retains original behavior, ie. will use subsequent timestamps to calculate next clock adjustments. logSyncInterval -4 logMinDelayReqInterval -4 first_step_threshold 0.001000000 step_threshold 0.000001000 step_window 0 ptp4l[3831.568]: port 1: LISTENING to UNCALIBRATED on RS_SLAVE ptp4l[3831.634]: master offset -1613279867776376096 s0 freq +0 path delay 7711 ptp4l[3831.698]: master offset -1613279867776376100 s1 freq -64 path delay 7707 ptp4l[3831.822]: master offset -1613279867776376124 s0 freq -64 path delay 7711 ptp4l[3831.884]: master offset -1613279867776376140 s0 freq -64 path delay 7711 ptp4l[3831.948]: master offset -1613279867776376124 s1 freq +192 path delay 7703 ... ptp4l[3832.447]: master offset -1613279867776376180 s0 freq -128 path delay 7707 ptp4l[3832.511]: master offset -1613279867776376166 s1 freq +96 path delay 7705 ptp4l[3833.572]: rms 5875888855636859904 max 6453119470999989248 freq +67785 +/- 119252 delay 8448 +/- 2020 ptp4l[3834.634]: rms 6453119470999931904 max 6453119470999931904 freq -139 +/- 360 delay 7711 +/- 21 ptp4l[3835.697]: rms 7237815975126660096 max 8136258558135193600 freq +75045 +/- 113951 delay -137763073020509120 +/- 496711823640969536 ptp4l[3836.696]: rms 6068654064854322176 max 8136258558135192576 freq +69808 +/- 117623 delay 8492 +/- 1500 ptp4l[3837.759]: rms 2174226957844811520 max 2174226957844814848 freq -3409 +/- 6863 delay 7697 +/- 43 ptp4l[3838.821]: rms 1348149014385676288 max 2174226957844810752 freq -85 +/- 39 delay -113134119417210480 +/- 265323028344133024 ptp4l[3839.821]: rms 703483405420974336 max 703483405420974336 freq -199 +/- 52 delay 7712 +/- 11 The multiple clock steps (s1 - [3831.698], [3831.948], etc.) overlap each other and causes convergence propblems for higher packet rates like 16 packets per second. Setting step_window to 32 (2 seconds at 16 PPS): logSyncInterval -4 logMinDelayReqInterval -4 first_step_threshold 0.001000000 step_threshold 0.000001000 step_window 32 ptp4l[4051.547]: port 1: LISTENING to UNCALIBRATED on RS_SLAVE ptp4l[4051.676]: master offset -1613280090261264012 s0 freq +0 path delay 7723 ptp4l[4051.740]: master offset -1613280090261263998 s1 freq +224 path delay 7701 ptp4l[4051.926]: master offset -1613280090261264042 s0 freq -256 path delay 7709 ptp4l[4052.051]: master offset -261264088 s0 freq +1000000000 path delay 7727 ptp4l[4052.176]: master offset -261264116 s0 freq -384 path delay 7703 .. ptp4l[4053.551]: master offset -261264516 s0 freq -192 path delay 7699 ptp4l[4053.676]: master offset -261264558 s0 freq -192 path delay 7717 ptp4l[4053.801]: master offset -598 s2 freq -374 path delay 7721 ptp4l[4053.802]: port 1: UNCALIBRATED to SLAVE on MASTER_CLOCK_SELECTED ptp4l[4053.864]: master offset -580 s2 freq -535 path delay 7703 ptp4l[4053.926]: master offset -538 s2 freq -667 path delay 7689 ptp4l[4053.989]: master offset -506 s2 freq -797 path delay 7689 ptp4l[4054.051]: master offset -444 s2 freq -887 path delay 7695 ptp4l[4054.114]: master offset -420 s2 freq -996 path delay 7695 ptp4l[4054.176]: master offset -332 s2 freq -1034 path delay 7687 ptp4l[4054.239]: master offset -276 s2 freq -1077 path delay 7691 ptp4l[4054.301]: master offset -224 s2 freq -1108 path delay 7695 ptp4l[4054.364]: master offset -146 s2 freq -1097 path delay 7681 ptp4l[4054.426]: master offset -114 s2 freq -1109 path delay 7681 ptp4l[4054.489]: master offset -34 s2 freq -1063 path delay 7705 ptp4l[4054.551]: master offset 2 s2 freq -1038 path delay 7705 The first clock step at[4051.740] is given a chance to complete. After 2 seconds (ie. 32 Sync packets at 16 PPS rate), the master offset at [4053.801] properly reflects the results of the first clock step. v2: Suggested by: Richard Cochran <richardcoch...@gmail.com> - Avoid using a timer with asynchronous signal. - Use step_window in terms of number of Sync events to skip. - Simplifies implementation to a simple counter. Vincent Cheng (1): clock: Introduce step_window to free run x Sync events after a clock step. clock.c | 17 +++++++++++++++++ config.c | 1 + ptp4l.8 | 8 ++++++++ 3 files changed, 26 insertions(+) -- 2.7.4 _______________________________________________ Linuxptp-devel mailing list Linuxptp-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/linuxptp-devel