#linuxcnc-devel Logs

Jul 27 2018

#linuxcnc-devel Calendar

02:01 AM rmu: just re-testet OpenGL (Full KMS), OpenGL (Fake KMS) and Legacy-settings on the pi3b:
02:04 AM rmu: only legacy seems to work. thread-time with Full KMS and Fake KMS is awful, sserial on the 7i90 refuses to start
02:17 AM rmu: thread-time of the servo thread mostly stays at 500k +- 100k which i think is strange for a 1000000ns period
03:00 AM rene-dev_: rmu
03:01 AM rene-dev_: rmu I think pcw reported that those times are in clock cycles not ns
03:05 AM rmu: 1,2ghz clock
03:05 AM rmu: even more confusing
03:07 AM rene-dev_: Im working on the tooltable cleanup again... in some places its still reporting the index instead of the pocket number
03:07 AM rmu: something strange is going on. running cyclic-test with Full-KMS is unimpressed by multiple glxgears, webbrowser and mathematica
03:07 AM rene-dev_: and, I found even more bugs, that are even there without my bug
03:07 AM rmu: but linuxcnc bombs out
03:07 AM rene-dev_: patch
03:09 AM rmu: very good ;-) i will set up a pi3b+ test system with mesa 7i90 and spi and try to get to the bottom of that latency stuff. i wouldn't be too surprised if it turns out that SPI is somehow affected by openGL
03:09 AM rene-dev_: it probably is. doesnt it use a closed source driver?
03:11 AM rmu: there are some binary blobs involved, in case of "Fake KMS" some translator codes translates opengl to proprietary broadcom api, in case of "Full KMS" i was under the impression that everything is in Mesa (the OpenGL Mesa) and Kernel KMS.
03:12 AM rmu: but of course still videocore needs those blobs to boot, setup up everything and so on
03:14 AM rene-dev_: https://github.com/LinuxCNC/linuxcnc/blob/master/tests/toolchanger/reload-tool/non-random/test-ui.py#L113
03:14 AM rene-dev_: this test passes, but it should not
03:14 AM rene-dev_: if I report the pocket number instead of index at that place, everything breaks
03:14 AM rene-dev_: and T remaps are still broken for some reason
03:29 AM rmu: without gui, just starting hostmot, motmod and servothread from halrun, everything works. servo-thread.time stays at half of expected value.
03:30 AM rmu: running cyclictest, multiple glxgears, and mathematica in parallel
07:07 AM jepler: only x86/x86_64 have the "some times are CPU cycles" quirk. elsewhere, we use microseconds for "clocks" https://github.com/LinuxCNC/linuxcnc/blob/master/src/rtapi/uspace_common.h#L263
07:07 AM jepler: er nanoseconds
08:41 AM pcw_home: jepler: ok so that particular RPI mystery solved...
10:55 AM hazzy-lab: seb_kuzminsky: Somewhat related to the pos.txt file issue, if the VAR file becomes corrupt the backplot becomes drunk, but LCNC seems to run normally
10:55 AM hazzy-lab: https://i.imgur.com/4OyevIP.png
10:56 AM hazzy-lab: Is this maybe because the back-plot used the corrupt var file, but LCNC uses the var,bak file if the var file is corrupt?
12:27 PM seb_kuzminsky: hazzy-lab: whoa, weird, please open an issue for that!