User space EtherCAT master shows better performance.

For some time, I have been testing an EtherCAT master that runs in user space.
My test data shows that this solution is faster, results in less variation in execution time, 
and significantly reduces the number of cycles with unusually long execution times.



The code is available here if you are interested: github.com/MetalMusings/linuxcnc-ethercat
I run both the IgH master and the user-space EtherCAT master, though not simultaneously.

This work stands on the shoulders of giants. Sasha Ittner (sittner, github.com/stratuMAK/ethercat )
was responsible for porting the EtherCAT code to user space, and Claude was a great help.
I simply guided Claude in the process of integrating this into Linuxcnc/lcec.

This is very interesting, I did try Sasha's uspace and did not work well for me, I may give it another go, thanks for sharing!

Should we join forces to improve the linuxcnc-ethercat repo ?

XDP-native is the main factor in the improvements, but not running in the kernel is better even without xdp-native.

Apart from the performance improvements I think its a plus to reduce the IgH ethercat eco-system down to a 
libethercat.so library and the ethercat utility. Anyways, it always nice to see new things.

To test, start with my linuxcnc-ethercat and follow the instructions. Better yet, let Claude follow the instructions.

I took a break from linuxcnc about a year ago. I am not updated on what's going on.
I have this old version of linuxcnc-ethercat. Made some improvements after Claude compared it with ETG documents.
I am not sure what I can contribute with these days.

I asked for improvement areas:

"
The improvements below would help users the most:

- A generic CiA 402 driver driven by data: read the supported modes (object 0x6502) and the PDO mapping from the drive itself. Vendor quirks go in a data table instead of new C files. Many of the existing drivers are basically CiA 402 with small variations.
- ESI files as a data source: read the vendor's ESI XML for default PDOs, SDO init commands and DC modes. A slave that lcec doesn't know yet then works from its ESI file, with no new code.
- A smarter configgen: scan the bus, match each slave against ESI files and known profiles, and generate both the XML configuration and a draft HAL file, for example linking each joint.N to its CiA 402 axis. Also add a check that compares the configuration with the actual bus before LinuxCNC starts.
- A simpler configuration schema: describe an axis as "drive X at position N, CSP mode, 10,000 counts per mm" instead of listing PDO details.
- Clearer errors: for example "slave 3 refused SAFEOP, AL status 0x001E, probably a wrong PDO mapping" instead of a silent timeout.
"