Intel I210: Excessive Collisions on TX

[MKotu]

Hi.

We've met some issue with I210 controller.

I210 is used in linux environment (MSI-X, multiqueue).

RX queues seems to be working (we can receive packets).

But we've found that each and every transmitted TX descriptor contains E1000_TXD_STAT_EC (i.e 0x00000002, Excess Collisions) bit in status field.

So, what are the possible reasons of "Excess Collision" on every TX packet?

Thanks in advance.

[idata]

Hi mkotulla,

Thank you for the post. Just to double check:1) What is the board you used? 2) Driver version for the I210 ethernet controller? 3) What is the software you used for the transmission? (in case there is) 4) What is the switch model used?Rgds,wb

[MKotu]

Hi.

1) We are using HP E0X95AA (Intel Ethernet I210 - T1 GbE NIC) ethernet adapter via PCIe slot (2.5 GT/s, x1): MSI-X and 4 rx-tx queues are enabled.

Standard best-effort LAN traffic is directed to tx queue(3), leaving tx-queue(0) through tx-queue(2) on the Intel i210 available to be mapped into a user-space Intel igb lib and managed by the lib.

All receive packets go to the same queue shared with the best-effort LAN traffic.

2) Driver version is Intel igb_avb 5.3.2 driver within Linux 4.6 kernel and Intel user-space igb lib.

3) AVnu Open-AVB software tools (avtp_pipeline, standalone applications) are used to communicate with Intel igb lib which is used for transmission (the register space of the I210 is mapped into Intel igb lib).

4) We are using two testcases:

- two I210 adapters are connected via AVB switch;

- two I210 adapters are connected one-to-one without any switches;

In both cases behavior is the same - each and every TX descriptor transmitted via Intel igb lib contains E1000_TXD_STAT_EC (i.e 0x00000002, Excess Collisions) bit in status field.

Standard best-effort LAN traffic via rx-tx-queue(3) seems to be working (i.e. ping, file transfers work).

But we've also noticed a significant number of dropped standard RX packets reported by ifconfig. And it grows approx. by 1 dropped packet every 1-2 seconds.

And each transmission on tx-queue(0) through tx-queue(2) causes E1000_TXD_STAT_EC (i.e 0x00000002, Excess Collisions) status bit in TX descriptor and dropped packet.

If it will help to find the possible reason of Excess Collisions - I can send the I210 register dump (the output of ethtool -d).

Kind regards,Manfred Kotulla

[idata]

Hi Manfred Kotulla,

Thank you very much for the information. Can you help share the register dump? Thanks,wb

[MKotu]

Hi.

I've attached register dump from 'ethtool -d' command.

Please, see register_dump.log file in the attachments.

Thanks in advance,

Manfred Kotulla

[idata]

Hi Manfred Kotulla,

Thank you for the information provided. I will further check on this.rgds,wb

[idata]

Hi Manfred Kotulla,

Just to double check:Have you contacting HP* already as this is an HP OEM card, please make sure you are using the driver from HP* correctlyAre these card came with the server directly from HP* or just HP* OEM retail model?Please update me if HP* have provide any support in case you contacted them already.thanks,wb

[MKotu]

Hi,

I've tested with the original PCIe ETH adapter card from Intel: INTEL Eth Adptr I210-T1 (version: G86562-004).

And got the same issue:

RX queues seems to be working (we can receive packets).

But each and every transmitted TX descriptor contains E1000_TXD_STAT_EC (i.e 0x00000002, Excess Collisions) bit in status field.

I've attached register dump from 'ethtool -d' command for Intel ETH card.

Please, see INTEL_register_dump.log file in the attachments.

Thanks in advance,

Manfred Kotulla

[idata]

Hi Manfred Kotulla,

Thank you for the update and provided additional testing information. I will further check on this.rgdswb

[idata]

Hi Manfred Kotulla,

Good day. Just to double check if you have tried cloning the repository as below:git clone https://github/com/AVnu/Open-AVB.git ? (the link is available under "Clone or download" of https://github.com/AVnu/Open-AVB/tree/open-avb-next) Doing this will have access to both master and open-avb-next branches and will see commits way newer. Please provide the latest version of open-avb-next when you are done.Thanks,wb

[idata]

Hi Manfred Kotulla,

Please feel free to provide information requested.Thanks,wb