@hobbyno1, no I have not. Last fall (2019) the land under my shop trailer and the storage yard company was sold out from under us (me and 400+ other tenants). I took a $250K personal write-off to save my credit history so that some guy next door could have a bigger parking lot - all my tools (including the Bridgeport S2 mill and tooling I had for a few weeks after trading up from the S1) are gone. Around the same time, Nvidia came out with its Preempt-RT patch for the L4T distribution on Jetson. If you think the storage situation is bad luck, I'm a white guy in his 40's who just bought a retired Ford Police Interceptor Utility (police "Ford Explorer") in February because I liked the car for technical and practical reasons, and I live in the S.F. Bay Area in the USA. I drive around with a target on my back in the current political climate, because of a few other people's poor choices. I'm up right now at 3:30am PDT because I can't sleep after getting into a row with my parents about the political climate.
I had pushed pretty much everyone I knew who was still there at Nvidia to escalate the Preempt-RT patch up in priority, and like you and numerous others, I have had success in applying the patch to a Jetson Nano, but that's as far as I got
To happier stuff...
@andypugh, the Nano uses an ARM A57, whereas the RPi 3 uses the A53 or the 4 uses the A72. The codebase may be compatible from a practical code-base sense, but it'll likely require some adaptation as libraries and dependencies are likely to vary.
To both of you, I'd recommend starting with a recent Nvidia L4T distribution, getting the P-RT patch applied (which it sounds like hobbyno1 has done), then start compiling LinuxCNC from there manually to see what if anything breaks. If you do find something that breaks, report it here and it can be worked through. There are likely some assumptions that have been made over the course of developing LinuxCNC that just by habit rather than malice are not compatible with various ARM implementations and will need to be rethought. It's just the nature of applying an existing code base to a new architecture. Since some work has been done for RaspberryPi already, I'm fairly certain that you'll find a lot of it "just works" with some tweaks to compilers/installers.
I'm willing to collaborate in this thread and help out where I can. My personal preference is to not lock it down to a specific OS image that gets flashed to a unit, but to have the LinuxCNC compilation be applicable to any suitable platform - this requires finding where build tools get "stuck".
I had gotten as far as purchasing some MESA hardware (I went to their shop in person to pick it up one lunch), but I had focussed on the RPi platform first. Got motion control running before the ground came out from under me last year. I was in the process of looking at adding a bunch of HW-level logic into the FPGA front-end of the MESA to do serial BISS absolute encoder correlation to A/B/Z encoder inputs, and to take a serial SpaceMouse Module as a 6-DOF jog-wheel... I had also been poking at Gecko about a closed-loop stepper/encoder version of their GR214V so that I could do 10000steps/per rev direct-drive into a "zero backlash" planetary roller-screw movement on the Bridgeport table. I was chasing 1uM per step.
Full disclosure before the next part (I think I've done this before), I worked at Nvidia for about 8 years in the Desktop Chip Solutions HW group (consumer video cards division) until a few years back. In my opinion, Nvidia has been doing an okay job getting a lot of this Linux stuff figured out - though it's still the bastard child of an OS to them in a lot of cases. Most of the time and effort they expend is for large commercial contracts - as you would expect, it pays the bills. I am also of a mind that in the service of the demands of large corporate customers, they have taken a triage approach to finding and fixing shortcomings rather than a structural approach to supporting the OS itself. Best advice I can give in light of that, is if you find something that isn't working and it's an Nvidia issue - get on their support forum and start making some noise.
