Latency, error finishing read, and IRQ affinity

sudo chmod 777 rt_setup.sh
sudo chmod a+x rt_setup.sh
I think ...

i changed the permissons on the file. i am refering to, in the file it states; "by default this program does nothing. change the exicution bits to run this file' or something along those lines. ill post an error report next time i am in front of it.

I should note that the rt_setup.sh depends on the host Ethernet port being setup in the interfaces file
and that the IP addresses in rt_setup.sh must match the host IP address setup there.

as far as permissions go, all that's needed is

chmod +x   rt_setup.sh
and then
sudo ./rt_setup.sh

i didn't get that pc working but i have a duo core laptop that i have it up and running on. i had it working on 2.8 at one point but my hardware configuration doesn't seem to like 2.9. i had issues with 2.9 when i first started researching linuxcnc so i switched back to 2.8 but i can't seem to get it working the way i had it. im going to remove and all the extra hardware from that pc and see if i can't install them one by one and pin point the issue. i get a journalctl error when i go through all the steps on the first page and when i run sudo ./rt_setup.sh, i get command not found. even thou i went through and setup the ethernet correctly and am able to ping the mesa 7i96s and get output from mesaflash. i created the file, made it exicutable and i still get the errors. the only pc i was able to get it work on was the duo core just to see if it was me or the machine. the service didn't run because it said i only had two cores but i didn't get the same error i get on two other four core machines. at this point, as long as it runs on the laptop i can work on the other machines at a later time and try to figure out my issues. thank you guys for all your help. it wont be the last time i need you, im sure. right now, i need some example hal and ini files to work with my drivers and servo as spindle. but ill start a new thread if my searching turns up empty

Is there any clear benefit or drawback to isolating extra cpus and then pinning the mesa ethernet to something other than the last one?

Also is there a way to configure which CPU LinuxCNC (I assume motmod?) pins to?

AFAIK LinuxCNC always pins itself to the highest numbered CPU

hachi wrote:

Is there any clear benefit or drawback to isolating extra cpus and then pinning the mesa ethernet to something other than the last one?

Also is there a way to configure which CPU LinuxCNC (I assume motmod?) pins to?

Probably eye candy because isolating more cores gives better looking  (faster) latency-histograms.
But with a Mesa card, the key paramater is the network latency, not the CPU latency.

PCW is right - LinuxCNC always pins itself to the highest numbered CPU. The only benefit to isolating other cores as far as I know is to isolate any cores that share cache with the highest numbered core. The script does this automatically.

I've gone through this thread and used the script, made the obvious mistakes of not having irqbalancer installed and having the wrong ip (10.10.10.1 instead of 10.10.10.11) in the script.  I've fixed all that, but I am still having a near immediate error when I start of Unexpected realtime delay on task 0 with period 1000000 followed by a hm2/hm2_7i76e.0: Smart Serial Port 0: Doit not cleared from previous servo thread....

I can move the machine around and run the spindle (modbus issues resolved with group admission change), but after a short while I get an error finishing read and the machine goes in to fault.

The hardware was all working with 2.8, so I don't think there is something that has died as it will all still work in 2.8, it's just this realtime delay issue.  I hate to bug anyone with this as I feel like I just haven't done enough due diligence, but this thread seems to be hitting all the high spots related to latency and I have not managed to resolve it.  Any help would be greatly appreciated.

Kind regards,

Sam

I've now gone through the BIOS and think I have everything set as it should be and there is no change. Also had it allocate a fixed RAM for the onboard graphics since it sort of felt like the error would happen more quickly if I resized the linuxcnc window, but I don't think that was anything more than coincidence. Ping looks fine. I think the latency seems fine.

ip setup seems similar to others with working setups and the correct IP address is being used in the script.

Again this all works fine with debian 10 and linuxcnc 2.8.  I am kind of at a loss at this point, though I feel like this is some simple config problem that I am overlooking.