Ethercat drives not responding, mesa error finishing read

I can try it, but since ethercat at the command line (linuxcnc isn't even running) shows ~50% lost uploads, what's the point?

I don't think this is a linuxcnc problem, but rather some networking or drive problem.

So you built ethercat master from source? Rebuild with the option --disable-eoe or something that makes it exclude eoe.
The three lines you showed from the syslog says it want to establish eoe network interfaces. Disable that.

Since you have problems with the Realtek adapters, here is a tip.
Add an Intel PCIe network adapter. Of course you need a free PCIe slot. With an X4 slot you can use a board with multiple interfaces.
And if you only have an M.2 slot you can use that forum.linuxcnc.org/18-computer/55782-adding-a-m-2-ethernet-port
 

You mentioned you should have an Intel driver. In which case the R8168-dkms won't be installed or do anything. Start at the beginning

what doeslspci | grep -i ethernet

report?

I have the same situation as you. If I solve it, I will reply below

Having mostly put the machine together and found ModbusTCP unsuitable, I took another attempt at Ethercat.

1. I installed LinuxCNC 2.9.4 from the ISO on the nearest available beige box (which has a Realtek ethernet card)
2. I installed the r8168-dkms driver
3. I built ethercat from source to exclude EoE as follows:
   1. git clone gitlab.com/etherlab.org/ethercat.git ethercat-master
   2. cd ethercat-master/
   3. git checkout stable-1.6
   4. ./bootstrap (I didn't have autoconf or libtool installed: they are necessary)
   5. ./configure --sysconfdir=/etc/ --disable-8139too --enable-userlib --enable-generic --disable-eoe
   6. make all modules
   7. sudo su
   8. make modules_install install
   9. depmod
   10. exit
   11. sudo nano /etc/ethercat.conf
   12. sudo systemctl enable ethercat.service
   13. sudo systemctl start ethercat.service
   14. sudo systemctl status ethercat.service
   15. sudo chmod 666 /dev/EtherCAT0
   16. sudo nano /etc/udev/rules.d/99-ethercat.rules
   17. sudo udevadm control --reload-rules
4. I installed linuxcnc-ethercat from the repository
5. I installed the CIA402 module:
   1. git clone github.com/dbraun1981/hal-cia402
   2. sudo halcompile --install cia402.comp

And it works very nicely most of the time. See attached ethercat xml file.

I see some "PLL error" failures sometimes, which can only be "solved" by restarting the control process. I've tried adjusting refClockSyncCycles and sync0Shift in accordance with some other threads, but I don't understand what those do or what they should actually be.

PLL was documented not long ago. Wasn't aware it can stop linuxcnc, that's a bit excessive.
github.com/linuxcnc-ethercat/linuxcnc-et...istributed-clocks.md

The PLL fault doesn't freeze or kill the process: one drive will stop responding to cyclic position commands. The only reset function I've built into my code so far toggles cia402.x.enable and cia402.x.fault-reset, and that doesn't seem to be enough to reestablish communication with the drive.

Do you see pll-reset-count increasing?

What's the exact error message you get? I can search the code to see what's going on.

You can always try "ethercat rescan", sometimes it clears errors.
e-stop on when you do this.

Edit: I guess it is PLL error 0x32 error message you get.
Check for other messages in syslog, like missed datagrams.
Improved network latency is the answwer in that case.
Also the pll-reset-count as mentioned above.