#linuxcnc-devel Logs

Apr 04 2023

07:29 AM rmu: what could be the cause of a ~75ms periodic slowdown for ~5ms
07:37 AM rmu: duration is not very consistent and period also varies a little bit
07:37 AM rmu: but I see hm2_7i97.0.0.write.time go up from 30000 to 70000 and read.time go up from 200000 to 600000
07:39 AM rmu: DRAM refresh?
07:55 AM rmu: nxmachine remote control software
08:10 AM rmu: testing with sysbench --test-memory suggests it is really the memory bandwidth thats the problem. this phenomenon seems to be worse under newer kernels.
01:01 PM pcw--home: This may be the general issue with post 4.19 kernels (cache policy related?)
01:22 PM linuxcnc-build2: Worker `checkin` is missing. It was seen last on Tue Apr 4 11:21:10 2023.
01:26 PM rmu: without nx remote display connected, values seem to be better on newer kernels, as long as you don't connect, then it will provoke missing read
01:26 PM seb_kuzminsky: looks like i missed some excitement while i was off on vacation :-/
01:27 PM seb_kuzminsky: https://github.com/LinuxCNC/linuxcnc/issues/2420
01:28 PM rmu: interrupt handling looks completely different on that machine between 4.19 and 6.1, same with cpu frequency, 6.1 reports 1500MHz, on 4.19, it seems CPU is constantly in "turbo" mode @ 2400
01:32 PM pcw--home: I always wondered about the wide time values for even idle LinuxCNC processes, and whether is was interrupt related or resource contention
01:38 PM rmu: this remote display software does h.264 encoding in software i suppose, highly optimized code capturing the display, and the "time" spikes would be consistent with 15fps. i can't provoke similar effects with video playback, glxgears or webgl stuff I tried.
01:41 PM rmu: on x86_64 those numbers are cpu cycles?
02:54 PM pcw--home: Yes clock cycles
06:43 PM andypugh_: pcw--home: I finally spotted a major part of why I wasn’t getting very far with the Modbus / PktUART reads:
06:43 PM andypugh_: https://github.com/LinuxCNC/linuxcnc/blob/master/src/hal/drivers/mesa-hostmot2/pktuart.c#L192