Mesa 7i92 Read Error

That actually sounds like a different type of error (bad KSZ8851?) Since the OP
had the same read error with a different 7I92 card. I Also ran his original card for 72 hours without a dropped packet and did tests at -51C(Freeze mist) and 100C without
issue which tend to rule out a marginal chip

I thought all the boards were testing before shipping and the defective chip would be detected. I noticed that the error appears either immediately after the activation of the interface or after a sufficiently long time. There may be some incompatibility of interfaces.

Yes, all cards are initialized and loopback tested over Ethernet but subtle problems
may be undetected (even by factory testing of the chips)

A good example of a similar issue is flash EEPROM chips, occasionally you get ones that
leak and forget their programming after a couple months, manufacturers do not test for
this flaw on a chip by chip basis and we see these every once in a while.

I FOUND THE PROBLEM!!

I was never sure where the communication problem came from - the specific computer I was using, the Linux installation and configuration, LinuxCNC itself, the hardware in the controller box, electromagnetic interference, power supplies, etc, etc.

With the help provided, I was finally able to track it down to the ethernet cable. Specifically, it was a short cable that ran between a panel mounted female RJ-45 to the 7i92 inside of the controller enclosure. I replaced the main ethernet cable with a shielded version a couple of years ago. I didn't replace the short cable, because it was very hard to get to, and many other cables had connectors that were too long to work, due to the enclosure layout. I was also thinking that I might be getting electromagnetic interference outside the metal enclosure, but that the inside cable was already shielded. This short cable had a sharp bend in it right at the 7i92 card. The bend must have resulted in a broken wire that was making a fairly good but not a great connection, and would fail after literally millions of reads over an extended period. That's what made it so hard to find.

I pulled the short cable out of the enclosure, but still included it in the middle of a string of three ethernet cables. It worked with no errors over a 24 hour period when it was not bent. I found that if I bent the cable right at the connection, I could cause the read error immediately.

I've ordered another short cable, but with a built-in 90 degree plug. I'm thinking this will permanently fix my problem.

Because of the extremely intermittent nature of the problem, it took several years and countless hours to find it. It is probably the hardest problem I've ever had to debug. On the good side, I now have extensive knowledge about LInux and LinuxCNC.

Many thanks to all who posted things to try over the years, and special thanks to PCW.

i did also face multiple times problems on the 7i92
that disapeard on realy good cable use not the cheep "china" stuff you can get for 1USD

real good cables from the next Telecomunication or TV-store
there are double shielded cat-6 out
en.wikipedia.org/wiki/Category:Ethernet_cables

Four years later, here's an update on where I am.
After replacing the short cable described above, I continued to get 7i92 read errors, usually, but not always, at very long intervals. It was mostly annoying, although when it happened during threading operations, it would cut a straight ring around my thread.

I spent a lot of time over the years trying to find the problem. I tried a lot of software fixes - changes to boot parameters, updated versions of Debian Linux, LinuxCNC and Gmoccapy, changes to the INI file - with no avail. Another area I was focused on was network adapters, and I tried multiple manufacturers. I also tried adding both large and small capacitors to my power supplies in case I was getting excessive ripple. Looking at the supplies with an oscilloscope didn't reveal much.

My last test was to try a second 7i92 connected to a wall wart power supply plus a new network patch cable in place of my CNC control unit. This completely removed any potential for problems in the CNC control unit.

To my surprise, I got no errors over long intervals! This was even with no Debian Grub boot parameters.

The good thing about this is that I could now focus on the CNC control unit and not any software issues. First, I used the original shielded network patch cable and the stand alone 7i92 to thoroughly check the cable. No errors. Then I added the short network panel mount cable inside the control box to the circuit. Errors! As described above, this shielded panel mount cable was a purchased to replace an unshielded panel mount cable I had trouble with before. I then ran the good patch cable directly to the CNC control box, and bypassed the short panel mount cable. Again, no errors!

It was another bad network patch cable that caused the problems. I have a network cable tester, and it tests OK. No broken wires. I flexed it while testing, and it was still OK.

I ordered a new panel mount cable from a different manufacturer. It came in a week ago, and I installed it. I just completed a 24 hour test, with no errors.

So the lesson here is that the Mesa 7i92 is very sensitive to network patch cables, even when they test OK on other devices. If you're getting read errors after very long intervals, the problem could be the patch cable.

I think it could also be that there is an intermittent connector issue.
We have never seen any cable sensitivity before.