#linuxcnc-devel Logs

Sep 02 2018

#linuxcnc-devel Calendar

07:04 AM Jin|away is now known as Jin^eLD
07:06 AM Jin^eLD: seb_kuzminsky: btw I kind of figured out the pin name length mystery, because it still did not add up even counting the component name, it seems that it appens .changepin, so the full name results in compname.pinname.changepin
07:57 AM Jin^eLD: seb_kuzminsky: shall I assign the PR to you since we discussed it a while ago or will someone just pick it up?
09:08 AM Jin^eLD: thx
02:10 PM rmu|w: pcw_home: i think you mentioned experimenting with the lepotato cc (amlogic S905 SOC), do you have any realtime kernel for that board? does mainline rt preempt work there?
05:41 PM pcw_home: rmu|w: I built a kernel but like most ARM system it requires a rather bizzare boot sequence so I'm stuck there ATM
05:42 PM pcw_home: I expect to also run into lack of hardware OpenGL support...
05:45 PM pcw_home: supposedly one of Libre computers claims to fame is their attempt to get all kernel patches mainlined so I was hopeful a 4.18.5 kernel might have the right stuff
05:46 PM jthornton: is the possibility to connect ist via usb/rs485 interface but the mesa interface would be nicer
05:46 PM jthornton: 06:02 AM valentin_s: i got a 7i96 fpga card
05:46 PM jthornton: that was this morning
05:46 PM jthornton: i would like to connect my huanyang vfd via rs485 directly to the serial interface of my mesa card.
05:46 PM jthornton: missed a bit
05:47 PM pcw_home: Unfortunately that requires someone to connect the modbus (mb2hal maybe?) interface to a low level hardware UART interface
05:48 PM jthornton: ok, I was just wondering
05:49 PM pcw_home: it would be nice (and probably better real time control that the userland modbus interface if the VFD modbus implementation was fast)
05:49 PM pcw_home: than the
05:49 PM pcw_home: good for plasma current setting also
05:50 PM pcw_home: I do have possible help with this so I will look into it now
05:52 PM pcw_home: rmu|w: something pretty weird with the RPI is that if you enable hardware OpenGL support you need isolcpus=1,2,3 to get a reliable 1 KHz servo thread running the SPI hardware
05:54 PM pcw_home: this of course make the RT stuff run like molasses in Siberia in January
05:54 PM pcw_home: sorry non-RT
05:58 PM pcw_home: but if you are only running LinuxCNC and no other large programs it seems fine (other than the fact that if you have a large and complex gremlin backplot and you try some fancy manipulations you can run out of memory...)
08:44 PM jepler: I'll go ahead and spin a new revision of the stretch live CD, try to get it uploaded tonight. seb has probably fixed a number of headaches by removing modemmanager...
08:44 PM jepler: not to mention all the updates from debian, new kernels and all that jazz
08:45 PM jepler: ugh that's weird, it has two kernels somehow
08:51 PM jepler: it installs the one from strech/main first (the one in the last debian point release)
08:51 PM TurBoss: jepler: hello, what is missing with the rtai kernel on stretch?
08:51 PM jepler: then it installs the one from strech/updates/main. at that point I would have expected the old kernel to become autoremovable
08:51 PM jepler: TurBoss: I don't work with rtai kernels anymore, so I'm out of date
08:51 PM TurBoss: ok thanks
08:52 PM jepler: the other day, hours after you left, I suggested that more information about your error might be in dmesg, such as the specific signal. I don't know if you saw that or tried looking in dmesg.
08:52 PM jepler: er specific symbol
08:52 PM jepler: TurBoss:
08:52 PM TurBoss: oh sorry didn't notice :(
08:54 PM TurBoss: recopiling
08:58 PM TurBoss: jepler: dmesg output without loading rtai modules
08:59 PM TurBoss: http://dpaste.com/1VNRT6K
08:59 PM TurBoss: now If I load the modules..
08:59 PM TurBoss: sek
09:00 PM TurBoss: insmod: ERROR: could not insert module /usr/realtime/modules/rtai_hal.ko: File exists
09:02 PM jepler: when properly configured, linuxcnc takes care of loading/unloading the modules.
09:03 PM TurBoss: then I don't need to load them firts? ok
09:03 PM jepler: rtf_create is a standard rtai function. https://www.rtai.org/userfiles/documentation/magma/html/api/group__fifos__ipc.html so on first glance it's unusual for it not to be available
09:04 PM jepler: but beyond that I don't have any other advice offhand
09:05 PM TurBoss: maybe because i pick rtai 5.1?
09:05 PM TurBoss: I should pick an older version?
09:05 PM jepler: hm more roadblocks on the way to uploading fresh isos :-/ I figured this was simple
09:05 PM TurBoss: :)
09:05 PM jepler: TurBoss: it's always a superhuman effort to get rtai going, that's why it is approximately 100 years between linuxcnc releasing CDs based on RTAI and why I'd tell you to use preempt-rt instead
09:06 PM TurBoss: oh
09:07 PM TurBoss: I need some hardaware
09:11 PM pcw_home: Preempt-RT has comparable latency to RTAI on some (Usually fast) hardware
09:12 PM pcw_home: http://freeby.mesanet.com/8300-4.18.png
09:15 PM TurBoss: still experimenting
09:16 PM pcw_home: I have not found any low power (fanless) CPUs that have really good latency on Preempt-RT though, decent for servo thread only but not good for software stepping
09:19 PM TurBoss: I saw the humingboard thing
09:26 PM TurBoss: https://wiki.solid-run.com/doku.php?id=products:imx6:hummingboard:hbpro has pci-e
09:27 PM TurBoss: havent tested
09:29 PM jepler: OK, upload of "-r11" stretch rt-preempt images is in progress, should finish within the hour.. I won't stick around and shepherd it though
09:32 PM pcw_home: I can see a possible advantage of a $30/35 RPI/Rock64/Potato but an X86 PC is likely cheaper and faster than the hummingbird
09:33 PM pcw_home: a $30 rock64 and a couple 3 PowerStep01s on a board might be a decent low cost system
09:35 PM TurBoss: true
09:35 PM pcw_home: doable in china for <$100 including 3 10A 80V step drives
09:39 PM pcw_home: (probably need stick-on heatsinks to actually get 10A but the PowerStep01s are quite decent (16 mOhm MOSFETs))
09:40 PM jepler: yeah I was just thinking -- it is sure to work just fine without a heatsink since it's shown without one in this aliexpress listing. https://www.aliexpress.com/store/product/-/1795334_32707808105.html
09:40 PM pcw_home: :-)
09:41 PM pcw_home: yeah the fact that MOSFETs have strong positive on resistance tempcos will bite
09:44 PM pcw_home: neglecting switching losses and using 25C on resistance its still >3W at 10A (but 10A is higher than most common step motor currents)
09:45 PM jepler: certainly for the people where it's reasonable to think about $100 controllers
09:45 PM jepler: but people always want more than they pay for
09:46 PM pcw_home: isolate the SPI and daisy chain it through all 3/4 drives and LinuxCNC can drive it with a 1 KHz servo thread
09:47 PM pcw_home: Its interesting that the PowerStep01 has a "VFD" like voltage drive mode (no current control)
09:48 PM pcw_home: but voltage scaled to velocity based on motor resistance&inductance model
09:49 PM pcw_home: might have slightly better damping than current control
09:53 PM jepler: have you worked with this particular chip? I guess it should work in servo-ish mode for linuxcnc, send it "run(DIR,SPD)" and query back "ABS_POS" and "EL_POS". Looks like there's no way to read the latched ABS_POS + EL_POS at once (EL_POS is the microstep position)
09:54 PM jepler: you get 2 overlapping bits between ABS_POS and EL_POS I guess, but at any modest speed that is just not gonna work for much
09:57 PM pcw_home: I have not but know people that are using it in this way
09:58 PM pcw_home: I think all you need is ABS_POS
10:01 PM pcw_home: (which AFAICR is in units of the current ustep)
10:03 PM jepler: "the stored value unit is equal to the selected step mode"
10:03 PM pcw_home: yeah
10:03 PM jepler: I read it as being whole steps only
10:03 PM jepler: it should be fine then
10:04 PM jepler: afk, goodnight
10:04 PM pcw_home: 'nite