CPSW issue on DM8148 SOC - ethernet hangs when ifup/ifdown with ping traffic

Hi Piero,

Can you reproduce this issue on the DM814x TI EVM? Can you send me the steps to reproduce?

Can you try with the latest ti81xx linux kernel (which has some net/cpsw patches over the latest EZSDK)? The latest kenel is available at:
arago-project.org/.../

We have few differences in ethernet between DM814x/AM387x and AM335x:
processors.wiki.ti.com/.../AM387x_To_AM335x_Hardware_Migration_Guide

The SDOCM00094019 is valid till AM335x SDK 6, then seems to be fixed. You can get some later version of the AM335x SDK and check the ethernet driver cpsw.c

Regards,
Pavel

Dear Pavel,

thanks for your answers.

Pavel Botev said:

Hi Piero,

Can you reproduce this issue on the DM814x TI EVM?

No, I'm sorry, but I don't have the EVM close at hand.

Pavel Botev said:

Can you send me the steps to reproduce?

This is quite easy, I see the problem simply pinging the eth0 interface from the PC, while running a script that launches ifconfig eth0 down/up in loop.

Pavel Botev said:

Can you try with the latest ti81xx linux kernel (which has some net/cpsw patches over the latest EZSDK)? The latest kenel is available at:
arago-project.org/.../

The two latest patches seem promising, thanks for pointing out.

Pavel Botev said:

We have few differences in ethernet between DM814x/AM387x and AM335x:
processors.wiki.ti.com/.../AM387x_To_AM335x_Hardware_Migration_Guide

The SDOCM00094019 is valid till AM335x SDK 6, then seems to be fixed. You can get some later version of the AM335x SDK and check the ethernet driver cpsw.c

Regards,
Pavel

Ok, perfect.

Thanks you very much for your help.

Best Regards,

Piero Pezzin

Piero,

Piero Pezzin said:

Pavel Botev

Can you send me the steps to reproduce?

This is quite easy, I see the problem simply pinging the eth0 interface from the PC, while running a script that launches ifconfig eth0 down/up in loop.

Can you share that script?

Regards,
Pavel

Pavel Botev said:

Can you share that script?

Regards,
Pavel

Yes, please find it below:

#!/bin/sh
I=0
while true; do
    let I=I+1
    echo "LOOP #: $I"
    ifconfig eth0 down 
    sleep 10
    ifconfig eth0 192.168.10.100 up
    sleep 10       
done

Dear Pavel,

forgot to mention:

- CPSW is configured in Dual Emac mode
- The interfaces are connected to different LANs and up
- The Ethernet failure happens faster when pinging both ports from a PC, with the ifup/ifdown loop targeting eth0

I've applied the patches from the ti81xx-master branch and quickly run the test, but the problem seems to be still here...

Moreover, I've checked the cpsw driver in the AM335x tree and to me it seems quite different from the DM81xx one..

Best Regards,

Piero Pezzin

Piero,

In the below location you can find net/ethernet/cpsw patches for newer than DM814x devices (like AM335x). You can have a look and check if you will find the patch you need. One example is:

git.omapzoom.org/

The location is:
git.omapzoom.org/

Regards,
Pavel

See also the below e2e thread:
e2e.ti.com/.../290526
e2e.ti.com/.../374766

This issue might be fixed in the AM335x SDK7 linux-3.12

Regards,
Pavel

Dear Pavel,

I've applied those patches suggest in e2e.ti.com/.../290526 to the DM814x kernel, but I still see problems (although behaviour is sometimes different - sometimes I get an "irq: xxx nobody cared" error message, then pings become very slow, but the ethernet subsystem doesn't hang).

Moreover, I've tried on a AM335x-based platform with 3.12 kernel from AMSDK 7 and CPSW in dual emac, and I see exactly the same behaviour: pinging both the interfaces from the PC, ifup/ifdown loop on eth1, after few minutes ethernet hangs.

Any help will be greatly appreciated.

Thanks and best regards,

Piero Pezzin

Piero,

I can suggest you to try with AM335x SDK8, SDK1.00.00.00, SDK1.00.00.03, SDK2.00.00.00, SDK2.00.01.07, to see if some of the new SDK version fix this.

Regards,
Pavel

I found this problem report for AM335x also in this forum:

e2e.ti.com/.../1766826

Can the TI team duplicate the problem?
The TI team has identified a solution?

Thanks a lot