Configure ConnectX with the mlxreg/mcra Utility

[5hubh4m]

Hello, I have two questions that are somewhat related.

I was studying the script disable_wqe_checks.sh. It uses mcra to set a bunch of registers. Is there any resource available to understand what these registers are and what the values correspond to. Perhaps there is a way to translate the address given to mcra to registers reported by mlxreg.

Secondly, I am trying to use the mlxreg utility to configure the endianness of the fetch-add verb (I want the output to be big-endian). I found the register HOST_ENDIANNESS that seemed like it would configure that. However, when I try to change it’s value, I get the error: ME_ICMD_OPERATIONAL_ERROR.

$> sudo mlxreg -d /dev/mst/mt4115_pciconf0 --reg_name HOST_ENDIANNESS -s "he=0x1"
You are about to send access register: HOST_ENDIANNESS with the following data:
Field Name | Data    
===================
he    | 0x00000001
===================

 Do you want to continue ? (y/n) [n] : y
 Sending access register...
-E- Failed to send access register: ME_ICMD_OPERATIONAL_ERROR

I have tried it on ConnectX-5 and 6 as well. Does anyone know a solution to any of these two problems?

[ssimcoejr]

Hello 5hubh4m,

Thank you for posting your inquiry to the NVIDIA Developer Forums.

Is there any resource available to understand what these registers are…

Unfortunately, this is proprietary information, and cannot be shared.

Secondly, I am trying to use the mlxreg utility to configure the endianness of the fetch-add verb (I want the output to be big-endian).

Unfortunately, this is a non-standard configuration, and we do not support this.
Use of the mcra utility to modify adapter register values should only be done under direct instruction from NVIDIA engineering.

If you require our products to operate in a capacity that they are not currently designed to operate, this is a customization consideration which we can only assist with through our solutions engineering team.

You can reach out to our solutions engineering team via a request to our sales team. They will be able to assist you with developing a solution that fits your unique business needs.

Thanks, and best regards,
NVIDIA Enterprise Support