Version 2.8.2 Debian Buster PREEMPT-RT ISO Freezes

Maybe you have same problem with mine. I have pc with 1 GB ram and it freezes after some time of using LCNC.
I watched the ram consumption and it slowly came up during some period.
can you confirm you have enough ram.

At the moment I've 2 GB of ram. I had 4 GB but took out half of it in the course of testing to see whether the freezes were due to a memory problem.

It seems rather unlikely that LinuxCNC has some form of memory leak. If it did everybody would ultimately have their systems freeze.

I'm coming to think that the problem has to be related to ethernet communication with the Mesa card. When LinuxCNC is just sitting there, it can't be doing much other than periodically exchanging packets with the card. A timer goes off, LinuxCNC receives and sends packets, and then LinuxCNC goes back to sleep. Perhaps the communication involves polling with interrupts disabled, and somewhat the awaited event never occurs? This would explain how the whole system freezes.

I did try monitoring the CPU temperature, and it never goes above 68C, even with LinuxCNC running.

I've just been playing around with things, and think that I may be getting somewhere.

Running a browser at the same time as LinuxCNC causes things to fall apart pretty dramatically.  I had an 7i76e error message ("Dolt not cleared from previous thread") show up, and found myself in a situation where LinuxCNC was responding but any attempt to move any of the axes produced no movement and a "Joint n following error",  Also there was that immediate failure last night when, because I'd been looking at the forum, I just happened to have a browser running.

At the moment I've LinuxCNC running with the Intenet Ethernet cable disconnected.  I've had a latency error message and a repeat of the 7i76e message (perhaps 5.10 has made things worse in this respect - I've never seen this message before) but at least the system has run for an unusually long time without freezing.

At the moment I have the Internet connected to LAN connector #1 (the "default" as reported by the network manager) and the Mesa card connected to LAN connector #2.  Perhaps Connector #1 has some form of priority, any activity on it interferes with activity on Connector #2, and I should try swapping things around?

 

It seems that I have at least managed to convert my problem into something more definitive.  LinuxCNC now survives for no more than a few seconds before running into a very definite and repeatable error.

The machine has two LAN ports: enpls0 ("wired connection 1") and ens1 ("wired connection 2"). Originally the internet was connected to enpls0, the Mesa card was connected to ens1, and the network manager showed enpls0 as the "default".

I've now swapped things around. The Mesa card is now connected to enpls0, the internet to ens1, and at some point in the process the network manager began showing ens1 as the "default".

I can ping the Mesa card. The very first round trip always seems to take a tad longer, but after that round trip times look pretty reasonable (e.g: min 0.099ms, avg 0.111ms, max 0.124 ms). And I can use mesaflash to extract information from the card. So Mesa card communications would appear to be working properly.

And I'm able to browse the Internet.

But when I start up LInuxCNC and try to do anything I almost immediately get

hm2/hm2.7i76e.0: error finishing read, iter = 58860

The iter value is typically somewhat smaller.

Once this has happened any attempt to move any axis produces "joint n following error" (where "n" is the axis number) and no movement at all.

And just to top things off, I am now seeing the latency warning message on a fairly regular basis. When I run the latency test, jitter for both threads is about 50000us. Not great, and very marginal for software stepping, but perhaps more than fine for a Mesa card?  I've no idea as to why the values are so high and how they might be improved. The only thing that I can think of is isocpus and I'm not sure whether this might be helpful.  I've done all I can with the BIOS - unfortunately there are very few things than can be disabled.

One strange thing I noticed in configuring the Mesa network connection is that things work if one simply omits a "gateway" value, but that if one actually enters the suggested value (10.10.10.1) then LinuxCNC fails on startup and produces an error file.  Can anybody explain what is going on here?

And can anybody explain how I'm getting  "finishing reads" errors when communications would appear to be functional, round trip times look reasonable, and latency values seems to be within reason (for a Mesa card)?

I really am at my wits end, and I am beginning to think that another PC may be my only option.
 

The servo period is 1,000,000.

When ping is allowed to run for a while, the max round trip time goes up to about 0.6 ms. Which is getting close to the servo period of 1ms. What might make some communications take so long when the average remains not far above 0.1ms?

I've no ethtool and when I tried to install it I ran into an issue:

john@Taig:~$ sudo apt install ethtool
Reading package lists... Done
Building dependency tree
Reading state information... Done
You might want to run 'apt --fix-broken install' to correct these.
The following packages have unmet dependencies:
linux-image-5.10.0-13-rt-amd64 : Breaks: wireless-regdb (< 2019.06.03-1~) but 2016.06.10-1 is to be installed

I've only the vaguest idea what this means, and I've no idea how to proceed....

When I try "apt --fix broken iinstall" the suggestion is to remove linux 5.10.

Presumably this would get me back to where I was before, and this might not be so bad an idea given that 5.10 doesn't seemed to have solved anything. It would be interesting to see whether I get the same read errors with the original version of linux and the LAN ports swapped. This would indicate an issue with enpls0.

But unfortunately I'd be no closer to explaining or eliminating the freeze ups.