Machinekit/RPi3b+: isolcpus with latency-test

[Thomas Treichl]

Hi everybody,

  a few weeks ago I bought a Raspberry Pi 3b+ and since then I'm making experiences with both: RPi and Machinekit. However, I've come to a point to not get any further anymore, having one minor and one major problem. With this email I describe the minor problem, I think it will take a lot more time for the major one. Thanks for any advice in advance helping me further! 

  I'm successfully running a Linux RT kernel on RPi 3b+, I isolated cores 2,3 of my CPU according to http://www.machinekit.io/docs/hal/threads-and-latency/, I'm successfully running the cgreb service. Here is some further information resp. output of my system (if I can provide more information about something then please let me know):

pi@raspberrypi:~ $ uname -a && cat /etc/os-release | grep PRETTY

-> Linux raspberrypi 4.19.71-rt24-v7-rt+ 

-> #1 SMP PREEMPT RT Sun Jun 28 07:57:30 CEST 2020 armv7l GNU/Linux

-> PRETTY_NAME="Raspbian GNU/Linux 9 (stretch)"

pi@raspberrypi:~ $ lscgroup cpuset:/

-> cpuset:/

-> cpuset:/rt

pi@raspberrypi:~ $ dpkg -l machinekit-rt-preempt machinekit

-> ||/ Name           Version      Architecture

-> +++-==============-============-============

-> ii  machinekit     0.1.15705286 armhf       

-> ii  machinekit-rt- 0.1.15705286 armhf       

pi@raspberrypi:~ $ DEBUG=5 CGNAME=/rt latency-test

pi@raspberrypi:~ $ ps -Leo pid,tid,class,rtprio,stat,wchan,psr,comm,args | grep `pidof rtapi:0`

-> 1801  1801 TS       - SLsl -        0 rtapi:0         rtapi:0

-> 1801  1805 TS       - SLsl -        1 rtapi:0         rtapi:0

-> 1801  1806 TS       - SLsl -        1 rtapi:0         rtapi:0

-> 1801  1810 FF      98 SLsl -        2 slow:0          rtapi:0

   -> From the DEBUG messages of latency-test I get

   -> rtapicmd {

   ->   instance: 0

   ->   threadname: "slow"

   ->   threadperiod: 1000000

   ->   use_fp: false

   ->   cpu: -1

   ->   flags: 0

   ->   cgname: "/rt"

   -> }

  Success, the slow thread is started on isolated core #2 of the CPU. But, if I call latency-test with two more arguments then never any of the two threads starts on an isolated core. The DEBUG messages tell me "cpu: -1" for both threads and they are always started on core #0 of the the CPU, eg.

pi@raspberrypi:~ $ DEBUG=5 CGNAME=/rt latency-test 100us 1ms

pi@raspberrypi:~ $ ps -Leo pid,tid,class,rtprio,stat,wchan,psr,comm,args | grep `pidof rtapi:0`

-> 3005  3005 TS       - SLsl -        1 rtapi:0         rtapi:0

-> 3005  3009 TS       - SLsl -        0 rtapi:0         rtapi:0

-> 3005  3010 TS       - SLsl -        0 rtapi:0         rtapi:0

-> 3005  3014 FF      98 SLsl -        0 fast:0          rtapi:0

-> 3005  3016 FF      97 SLsl -        0 slow:0          rtapi:0

   -> From the DEBUG messages of latency-test I get

   -> equest: type: MT_RTAPI_APP_NEWTHREAD

   -> rtapicmd {

   ->   instance: 0

   ->   threadname: "fast"

   ->   threadperiod: 200000

   ->   use_fp: false

   ->   cpu: -1

   ->   flags: 0

   ->   cgname: ""

   -> }

  Next, if I switch machinekit packages on the same system to 

pi@raspberrypi:~ $ dpkg -l machinekit-*

-> ||/ Name                            Version              Architecture

-> +++-===============================-====================-============

-> ii  machinekit-cnc-rt-preempt       0.1.1569926353.git8e armhf       

-> ii  machinekit-hal-rt-preempt       0.2.1573130755.git12 armhf       

-> ii  machinekit-manual-pages         1.10340              all         

  then neither with the command "CGNAME=/rt latency-test" nor with the command "CGNAME=/rt latency-test 100us 1ms" any of the threads is started on an isolated core. I traced back this problem to here because I actually wanted to "loadrt motmod" within "halrun" and assign the two created threads to the two isolated cores, without success. 

  It looks like I'm missing something, maybe somebody can help me.

Thanks,

  Thomas