Thanks for ideas.

It gets worse and I've been pondering the next step.: "The machine is unusable; the axes shake and make a lot of noise. I tried rebooting and changing the refclock settings from -1 to values ​​of 1 or 1000, but to no avail. The PC has an Intel NIC, but, as I said, the hardware and software are identical to Debian 10. At this point, I don't know what to do. The problem is EtherCAT."
The customer would like to try Ethercat 1.52 from his Debian 10 days but I don't think it will run on later kernels 
The current Debian 10 Ethercat version on their repos is 1.62, Debian 12 and Debian 13 is 1.69 with no other earlier versions published. Latency is OK but not great for a 4 core machine
 

The only thing I can think of is to roll back to Debian 12. Any other ideas? 

 

rodw wrote:

Thought I'd ask here. User has upgraded from Debian 10 to latest Debian 13.3 and 2.10pre RIP config. Now Running grandixximos Ethercat driver but have not set syncToRefClock yet.

He reports:
With the same xml file, i/o and drive configuration, Debian 10 kernel 4.19 with ethercat 1.5.2 works fine, it has no latency at boot and hasn't given any problems. The current system has kernel 6.12 with ethercat 1.6.9 and has synchronization problems at boot, sometimes it doesn't see some drives on the bus, I'm forced to turn it off and on again. Would it be possible to use the current 6.12 kernel and linuxcnc 2.10, with ethercat 1.5.2?

I copied the lcec_conf and lcec.so files into the respective bin and rtailib folders from the Debian 10 version that I have and which works well on my machines, with linuxcnc 2.9.0 pre1 and etherrcat 1.5.2. At boot time, the drives are now seen within 1-2 seconds of booting and are always found on the bus.

This has 5 axis servos and a few Beckhoff modules attached to EK1100 

Next step is to add syncToRefClock

It seems odd to me that today's Ethercat hal driver is not performing but rolling back to an older version of it works. It seems to me his problem is the driver, not the version of Ethercat.

Any ideas or advice?

 

Ive experience of this non recognizing slaves on the bus during some experimenting at my debian 12 + 6.xx kernel ... but everything was solved by using non generic drivers and restarting ethercat service ifself... back that time I have benn using the 1.5.2 ethercat ... 

but it is sometimes odd till todays that with some hardware at bus ethercat is self-restarting and with some hardware is not ... and therefore bus has no slave active 

Ive not found solution at this till now...

It would be helpful to see the messages in syslog from when there was a problem.

from 5 April as an example. And cut to a reasonable length.

Hello PCW, I am having the same trouble.
Do you have or know where I can get a copy of 5i25_prob_rfx2.bit that includes a DPLL?
Thanks Robert?

Dear 

I think there are some opportunities in absolute encoders field with low information output as the one that I have.
 with only encoder counts in a multi-turn format s32 and encoder spins counter as well as multi-turn format s32.

I make it works but I need to spend some time testing it. but G33 and G76 are working with ethercat absolute encoder as spindle speed and position feedback , and some spontaneous axis following error to debug. ( jitter ? i do not know ).

Thank you.

Yes sorry, because your post was a bit all over the place.

about ethercat master, 1.6.9 is pretty new, so I was asking to double check with 1.6.8

But I agree linuxcnc-ethercat seems to be the issue, so I was asking about the code for lcec aka linuxcnc-ethercat
You moved the lcec_conf and lcec.so 
But do you have the code for the linuxcnc-ethercat that works? or can you at least share approximately where this working version could have come from? Without that, it is really hard to understand what is happening, and get to the bottom of it.

Before this, I tried a component and attempted to read the current X & Y position and the prior x,y position. This allowed me to calculate the radius which I wanted but it never worked accurately. So I went to the state tags approach and Andy one of the devs convinced me to add the heading while I was at it.

This is something that is missing in Linuxcnc. This is actually quite complex. I had a go at it some years ago but my use case faded away.
The pull request still persists.  github.com/LinuxCNC/linuxcnc/pull/900
I got it working but it needed a bit of cleanup once the devs reviewed. It really needs to be completed.

This gets pretty deep into the Linuxcnc internals. The interpreter and motion are totally seperate entities in Linuxcnc. The interpreter tokenises the gcode so motion only sees the waypoints.

The interpreter knows or can calculate the centre of the G2/G3 arc and its radius. Motion does not. The change which enabled this was the release of state tags. State Tags passes the interpreter state to motion in real time. 
github.com/LinuxCNC/linuxcnc/blob/master...c/motion/state_tag.h
I extended the tags to include
    GM_FIELD_FLOAT_ARC_RADIUS,
    GM_FIELD_FLOAT_ARC_CENTER_X,
    GM_FIELD_FLOAT_ARC_CENTER_Y,
    GM_FIELD_FLOAT_STRAIGHT_HEADING,
The last field is the heading you need for your knife.
I published all of these state tags as interp.* pins including the heading so interp.heading is exactly what you need!

I was thinking recently now I am retired I should revisit this project. Maybe you can help and come along for the ride.
 

1.69 was released two weeks ago. Its what you get from their repos with a current version of Debian Trixie.
gitlab.com/groups/etherlab.org/-/packages

The code should be here.
gitlab.com/etherlab.org/ethercat/-/tree/1.5.2?ref_type=tags

From 2013. wow! that's old! From that, I would think it was built from source from Sourceforge. But well before my involvement.
Just checking Etherlab's gitlab releases, the first version on gitlab repo was 1.53 in 2024 so must be from Sourceforge

From what I can see, its not an Ethercat issue because the Debian 10 version of  linuxcnc-ethercat (with Ethercat 1,52) works with no problems  on Debian 13.3 with Ethercat 1.69.

Ethercat 1.6.8 is the previous stable release, have you tried that? 1.6.9 is a bit new I think, maybe they changed something?

Do you have the code for lcec version that works on your debian 10? Can you share the .so ? 

dear friends , i am new imegrant from MACH3 to LINUXCNC, and hope to get engaged to the Real-Time environomrnt and its cabapilties to lead a tangential knife to my application. that MACH3 has this feature with suffecient performance, 
i am intending to develop same logic HALComponent to direct A-AXIS and now i am doing the math and trying to apply it to real-time C languge progrmming , i hope if there any help to guide and cooperate to accomplish it
here is the first approach 


component tangent_knife;
// inputs
pin in float vel_x;
pin in float vel_y;
pin in float acc_x ;
pin in float acc_y ;

pin in float current_angle;
pin in float threshold;
pin in float lookahead_time ;

// outputs
pin out float command_angle ;
pin out bit intervention_required;


function _;
license "GPL";
;;

#include <rtapi_math.h>

FUNCTION(_) {
    double target_angle;
    double future_vel_x, future_vel_y;
    double angle_diff;

    // 1.Simple Lookahead
    //  V_future = V_current + (Acceleration * Time)
    future_vel_x = vel_x + (acc_x * lookahead_time);
    future_vel_y = vel_y + (acc_y * lookahead_time);

    // 2. alculate ange
    if (fabs(future_vel_x) > 1e-6 || fabs(future_vel_y) > 1e-6) {
        target_angle = atan2(future_vel_y, future_vel_x);
    } else {
        target_angle = current_angle;
    }

    // calculating (Shortest Path)
    angle_diff = target_angle - current_angle;
    
    // normalization  -PI و PI
    while (angle_diff > PM_PI) angle_diff -= 2.0 * PM_PI;
    while (angle_diff < -PM_PI) angle_diff += 2.0 * PM_PI;

    if (fabs(angle_diff) > threshold) {
        intervention_required = 1;
   } else {
        intervention_required = 0;
    }

    command_angle = target_angle;
}

Thought I'd ask here. User has upgraded from Debian 10 to latest Debian 13.3 and 2.10pre RIP config. Now Running grandixximos Ethercat driver but have not set syncToRefClock yet.

He reports:
With the same xml file, i/o and drive configuration, Debian 10 kernel 4.19 with ethercat 1.5.2 works fine, it has no latency at boot and hasn't given any problems. The current system has kernel 6.12 with ethercat 1.6.9 and has synchronization problems at boot, sometimes it doesn't see some drives on the bus, I'm forced to turn it off and on again. Would it be possible to use the current 6.12 kernel and linuxcnc 2.10, with ethercat 1.5.2?

I copied the lcec_conf and lcec.so files into the respective bin and rtailib folders from the Debian 10 version that I have and which works well on my machines, with linuxcnc 2.9.0 pre1 and etherrcat 1.5.2. At boot time, the drives are now seen within 1-2 seconds of booting and are always found on the bus.

This has 5 axis servos and a few Beckhoff modules attached to EK1100 

Next step is to add syncToRefClock

It seems odd to me that today's Ethercat hal driver is not performing but rolling back to an older version of it works. It seems to me his problem is the driver, not the version of Ethercat.

Any ideas or advice?

 

Unless you changed the firmware, I doubt that that's the issue.
 (all cards are tested and shipped with standard firmware)

You think it's worth it to give it another shot with flashing another firmware before returning it or do you advice against that?

The hal and ini file look sane so I would guess its some kind of hardware issue.
 So the card should be returned for evaluation