ScanDevices stucks in endless loop

[XzuluX]

The adapted ScanDevices function stucks in endless loop on my current configuration. I have to investigate further to check the root cause, but it seems commit 9ad66bb Wait for PREOP before accessing EEPROM broke something on ScanDevices logic. If I revert to commit b393404 all is working as expected ... any idea?

[Apollo3zehn]

So PREOP is never reached? The only changed lines are: https://github.com/Apollo3zehn/EtherCAT.NET/blob/master/native/SOEM_wrapper/soem_wrapper.c#L562-L566
I need to update these to throw a timeout error instead of looping forever. I´ll check also against my test slaves.

[XzuluX]

Yes PREOP is never reached, if I comment out these changed lines of code its working ;)

[Apollo3zehn]

I found the solution, it is required to check for EC_STATE_PRE_OP | EC_STATE_ACK instead of EC_STATE_PRE_OP alone. I´ll publish an update soon.

[Apollo3zehn]

Should be fixed now with commit 4c8a647 :-) Sorry for the confusion.

[XzuluX]

@Apollo3zehn I just recognized your fix increased the master start time significantly. I currently have four slaves and it now takes approx. 25s compared to 5s before.

If I change the code like this, master start time goes back to 5s. The differece is to explicitly call ecx_writestate(context, 0) and only check for EC_STATE_PRE_OP.

Do you think it's a proper solution ?

// wait for all slaves to reach PREOP state
int counter = 200;
context->slavelist[0].state = EC_STATE_PRE_OP;
ecx_writestate(context, 0);

do
{
    ecx_statecheck(context, 0, EC_STATE_PRE_OP, 5 * EC_TIMEOUTSTATE);
} while (counter-- && context->slavelist[0].state != EC_STATE_PRE_OP);
               
if (context->slavelist[0].state != EC_STATE_PRE_OP)
    return -0x0104;

[Apollo3zehn]

I have also noticed an increase in startup time but I thought it was caused by my network card driver. I´ll check that tomorrow.

[Apollo3zehn]

We could also skip the automatic state change in SOEM ( https://github.com/OpenEtherCATsociety/SOEM/blob/d9261e801d109eaf99b94907b6d24bac4a123968/soem/ethercatconfig.c#L593) and do it ourselves later. The thing is that I don't know why EC_STATE_ACK is required.

I have slaves devices available for testing on monday, unfortunately not tomorrow.

[Apollo3zehn]

I think I found the issue:

At the end of ecx_config_init (called from within ScanDevices), SOEM requests the PreOp state + EC_STATE_ACK flag (0x02 + 0x10) (https://github.com/OpenEtherCATsociety/SOEM/blob/v1.4.0/soem/ethercatconfig.c#L596-L600).

Back in ScanDevices we try to check if that state has been reached within the timeout using ecx_statecheck. The problem is that ecx_statecheck just checks for the actual state (0x02) and not the flag.

Opposed to that, slavelist[0].state contains both, the state and the ACK flag (0x02 + 0x10).

Therefore my previous solution could not work and instead it should probably look like I did it here: 2f1ee3c
With my test slaves, I could reduce the time to a few milliseconds for the INIT->PREOP transition. Does this solution also work for you?

[XzuluX]

Great, thank's for solving this issue, now the performance is much better! Yes this works for me.