Homing problem with Gecko stepper drivers

I'm having the weirdest problem with homing. I have a Sherline that I originally built with Linisteppers, then moved to some stepper drivers based off TH6601's (I think that's right) and finally got a set of Gecko 251x's. I have one limit switch per axis, wired in series, and had homing working fine with the linisteppers and the TH6601's. When I moved to the gecko drivers I used the same basic configuration file, with my search velocity about 1/4 my default velocity, and my latch velocity about 1/4 my search velocity: same as I'd used before.

It doesn't work on the geckos at all. But it's weirder than just not working. When I initiate a homing command, the axes (z, then x, then y) make noises, blip, sometimes move a little bit in the right direction, maybe half a turn, usually just a few degrees, until I get annoyed and trigger a limit switch by hand. At that point, once I've pseudo-homed all three axes, the geckos work perfectly. I can cut stuff with excellent accuracy over multi hour runs.

I've posted about this before, but have spent more time playing with it tonight so I have more details. My default jog velocity is about 0.1 machine units. I've set the search velocity to every value from 10,000,000 down to 0.000001, in increments of 10x. Every axis behaves exactly the same at every setting, with the exception that at the very low velocity settings (below 0.001) when the axis does turn briefly it seems to do so at somewhat of a lower speed. (But not 10x or 100x or 1000x lower.)

I put a scope on the x axis step pin tonight. With the search velocity set to the same as the default jog velocity (0.1, in what the manual says are machine units) the scope shows that the homing steps have both the same amplitude and period as the steps when the machine is running normally. When I zoom in on an individual pulse, the waveform of the pulse appears to be identical between the two. The direction signal also appears to be the same between the two (although I need to look at that more: it appears to have more noise than I would expect for logic high.)

Using stepconf, I chose the Gecko 203 defaults for timing, and found for my machine values of 0.2 in/sec for velocity and 1 in/sec for acceleration provide reliable operation. (As in: any higher on either and the steppers start losing steps.)

I'm including my hal and ini files in hopes someone will have a guess as to what's going on. The .ini file has three wildly different values for search_velocity, as part of my try-every-possible-value-one-must-work experiment.

I'd love any advice anyone has.

yoiur ini is WRONG

you got 40000 steps per inch
i dont think this is right

HOME is the location that the mashine ends up after triggering the switch
Home_offset is the Location of the Switch to the Axis eighter max or min Usaly zero
OR as Max the max length of the axis

Sorry im a Metric as im in germany somone shoudt jump in with inch