iMX8QM Ethernet IEEE_rx_macerr

[gilmoreat]

I was running some iperf3 tests between two iMX8QM boards and every once in a while, I would see a burst of retries when running at a gigabit. When I checked statistics with ethtool -S, I say that I was getting some IEEE_rx_macerr errrors. It looks like these error indicate that the rx buffer in the MAC is overflowing and dropping packets. Is this expected? Is there a way in linux to increase this buffer size (running "ethtool -g" returnned "operation not supported")?

I am running a yocto build base on kernel 6.1 on a custom board (design based on the iMX8QM MEK). 

[JosephAtNXP]

Hi,

Thank you for your interest in NXP Semiconductor products,

Could you share the options set in iperf3 test as the configuration made to the interfaces prior or confirm if the board is fresh booted?

Are you able to replicate on two 8QM MEK? Could you run that test in latest Yocto release?

I would like to add, how many queues are configured? Check ERR050395: ENET: Ethernet RX hang when receiving traffic through multiple queues in 8QM Errata.

Regards

[gilmoreat]

Also, how do you check the number of queues? The two applications that would be using the ethernet interface at the time of the test are iperf3 and ssh. Does this mean that we would be using 2 queus?

[gilmoreat]

I am running iperf3 from a fresh boot and using the default settings except for using the -t option to run the command for more than the default 10 seconds. Unfortunately we were not able to find a MEK board in stock in the past so I don't have one to test with. We also require a some patches to the yocto to run on our board so updating to a newer version would take a while. 

[gilmoreat]

Coming back to this ticket since we were finally able to get a set of the imx8qm MEK development boards. I ran the following test:

- Connect the ethernet ports of both MEK boards together.

- boot default image that came on provided SD card (4.14.62-imx_4.14.62_1.0.0_beta+g1907fe4
)

- create a script called "dummyLoad.sh"

#!/bin/bash
x=1

while [ true ]
do
  x=$x+1
done

 - Run six instances of dummyLoad.sh in the background.

- run iperf3 between the two boards for 120 seconds

In two out of three runs, iperf3 reported an instance or two of a batch of retries. 

Is anyone able to recreate this on the NXP side?

[JosephAtNXP]

Hi @gilmoreat,

Queues are not always equal to the number of processes, could you try to run the following command?

ethtool -L <interface> rx 1 tx 1

Regards

[gilmoreat]

Hi @JosephAtNXP,

I tried running that command and I get the following error:

netlink error: Operation not supported

I am running an image based on kernel 6.1.22. Are more options for ethtool supported in later versions of the kernel/ethernet driver?

[JosephAtNXP]

Hi,

Please share the output of:

# ifconfig -a
# ethtool -l <interface>

Regards