C6678 EVM Ethernet: all outgoing packets experience TX Carrier Loss errors

Dmitry Korenev

Hello,

In our project we boot the board over BOOTP+TFTP (using IBL) and then we communicate with the outside world via Ethernet using NDK+NIMU. Generally it works quite well, i.e. we have acceptable performance and reliability.

Unfortunatelly, sometimes the board goes silent. Investigating the problem we found that the board boots up successfully, NDK and NIMU are initialized, NDK successfully receives packets and sends packets via NIMU. It seems that CPSW also does it job. The problem is that in STATSB we have TxGoodFrames = 0 and TxCarrierSenseErrors equals to the number of outgoing packets. In the SGMII status register we have 0x3d that is SerDes PLL is locked, Auto-Negotiation is complete, and link is up. We've tried to restart auto-negotiation, extract and re-insert te cable, attach to another machine etc. but with no luck.

It seems that the problem arises whenever the board have incoming packets short after booting.

I have no idea what to do with it and kindly ask for your help.

Regards,

Dmitry

NDK: 2.22.03.20

PDK: 1.1.2.6

over 11 years ago

0 Pubesh over 11 years ago

TI__Mastermind 20790 points

Dmitry,

Please make sure, this project is running successfully when load and run through No Boot mode like using CCS, JTAG and gel file.

0 Dmitry Korenev over 11 years ago in reply to Pubesh

Intellectual 370 points

Pubesh,

I've managed to reproduce it in No Boot mode as well. GbE subsystem is initialized by the standard GEL script which comes with the EVM.

It looks like a race condition somewhere as it happens to work 9 times but not the 10th.

Dmitry

0 Pubesh over 11 years ago in reply to Dmitry Korenev

TI__Mastermind 20790 points

Dmitry,

So we can conform this issue occures while running the application through CCS.
Is that your own code or example code from MCSDK?
How you are sending the packets from board, TCP or UDP socket?
Can you able to ping the board IP address after not working case
Can you try the NDK example code avilable at: mcsdk_2_01_02_06\examples\ndk

0 Dmitry Korenev over 11 years ago in reply to Pubesh

Intellectual 370 points

Pubesh,

>> So we can conform this issue occures while running the application through CCS.

It seems so

>> Is that your own code or example code from MCSDK?

It's our own code

>> How you are sending the packets from board, TCP or UDP socket?

We use both TCP and UDP but I believe it's not relevant (see below)

>> Can you able to ping the board IP address after not working case

No. In order to ping the board we should have ARPed it but ARP replies are not transmitted via SGMII. We've traced ARP request through the stack and NIMU and it works as expected

>> Can you try the NDK example code avilable at: mcsdk_2_01_02_06\examples\ndk

Ok, I'm going to try the 'client' example. if I reproduce the problem, have you an idea on what will be the next step?

Regards,

Dmitry

0 Dmitry Korenev over 11 years ago in reply to Dmitry Korenev

Intellectual 370 points

Pubesh,

It took a huge amount of time but finally I've done it after 100+ attempts.

Is it possible to narrow down the issue? I mean, how can I check if PHY/SGMII/SERDES is responsible for the behaviour observed?

Thank you,

Dmitry

0 Pubesh over 11 years ago in reply to Dmitry Korenev

TI__Mastermind 20790 points

Dmitry,

You can read the relevant registers through CCS, in the way can debug the issue.

0 Dmitry Korenev over 11 years ago in reply to Pubesh

Intellectual 370 points

Thank you for excellent support...

0 Pubesh over 11 years ago in reply to Dmitry Korenev

TI__Mastermind 20790 points

Dmitry,

However this discussion of thread can helpful for future readers, If you verified this thread.

Processors

Processors forum

C6678 EVM Ethernet: all outgoing packets experience TX Carrier Loss errors