5i25/7i76

My Mesa stuff has been running great for the last six months on our old gantry machine. Was running the 2.5 git stuff from January 2012. Tried a build from today, and I've got some weirdness. Sorry, I gotta get back to the shop and post some configs... it won't detect the 7i76 field IO and the stepgens go insane as soon as the hm2_pci is loaded by hal_cmd. No EMC running at this point, just halcmd. If you fire up EMC and the AXIS UI, you get a nice visual representation of the stepgens taking themselves for a drive, as AXIS plots out their position feedback. Glad the e-stop was handy this morning...

Now I went and cp'd the hostmot2 and hm2_pci modules from my old build, and I can operate with them fine (Jan modules compiled against Jan build, loaded by the July build). But if I load the July built modules, bad news. Any known regressions?

But all this aside, here's what I'm really worried about. We just built a new 5x10 plasma table, dual rack and pinion stepper drives on the Y. I tried gantrykins, but I get following errors starting at 700ipm, about half of what this machine can do with the drives slaved electrically (one stepgen, feeding two gecko drives). Then I tried running 3 logical axes with trivkins, but 4 stepgens. On the Y, having the motion module feed it's commanded position to two stepgens. But only taking it's feedback from one. I get this creepy, noticable lag where it appears to move both motors togethers, but on closer look, one of them definitely lags behind the other.

Now, maybe that's what's supposed to happen, if you connect the signals in HAL like that, but some reassurance would be nice. And I'd like to get this July build working right. Gotta say though, this was the cleanest build I've ever had for EMC. Hardly any tweaks needed to get it to compile on Fedora Core. Usually it's a multi-day Makefile slashing event

As far as the trivikins setup goes I cant see why there would be a delay (more than a few uSec)
between the stepgens if they are setup identically

Can you post your HAL/INI files?

Tomorrow I'll have my condensed configs, RTAI debug output, and the works. Tonight I have an easy evening shift at work, figured I could chase the what-if's on the Internet in the mean time.

I have them absolutely identical, step timing, scale, the works. Both ends are plugged in, I promise. Although usually that's what everyone says... in my case I have a buddy looking over my shoulder as I do these things, just waiting to pounce on a silly mistake. For the glory, of course.

I quadrupled the rate for my servo base timing.. shooting in the dark. Shouldn't matter thought, right? I imagine on every beat of the servo thread, both stepgens get serviced, and then are written to the 5i25 all within a very very very short period of each other. Very contrary to what I'm seeing.

I think I'm going to have to take a break from my 13 years of Red Hat loyalty and try the Ubuntu Live CD. You would not believe how tough it can be to build LinuxCNC and RTAI on Fedora Core. I have a problem with the built-in GCC soft FPU functions not being included when compiling the hostmot2 driver. So I have a copy of the GCC sources handy, and I statically link in the needed object files. Only for the hostmot2 driver. Maybe there is some floating point weirdness taking place. The last build has been very problem free though.

When gantrykins trips out on a following error, you can see the commanded positions in hal-config for the two stepgens are just different enough to cause the error. Why why why are the values different. What do you guys use for debugging tools, in the fast pace world of real time kernel mode code?

*hangs head in shame* I had to come back and report, as to not create a wild goose chase.

Configuration problem. That's with one axis position command, feeding two stepgens. It works fine now.

Gantrykins is still cranky. And the current hostmot2 drivers aren't working for me. But it's all good enough for now.

I'll tell you though, I'm pretty excited about the preempt_rt work. It'd be nice if by next year, LinuxCNC was just another program in the repositories. I'd gladly trade some latency, to not have to keep build notes for myself.

Anyways, thanks.