Timestamping CQE bug in Bluefield-3 NICs

Hi,

I have a program that uses the CQE to record tx/rx hardware timestamps on Connect X-7s and Bluefield-3 NICs. I am using PTP to sync my clocks.

Occasionally, I see timestamps that make no sense (they are extremely far in the past or future). I know this bug existed in the Connect X-7 and Bluefield-3 NICs, and that this bug was fixed, as I see the bug fix recorded in the firmware history for both the Connect X-7 and Bluefield-3 firmware that says: fixed the cr_space (Configuration Registers Space) in port configuration to prevent wrong timestamp of cqes.

Updating the firmware for the Connect X-7s to the latest version allows my code to run as expected on the Connect X-7s. However, even though the bug fix for the Bluefield-3s was pushed in version 32.39.2048, and I am using a firmware version that was released much later (32.42.1000), I still have this timestamping issue when running the same code (that works on the Connect X-7s) on the Bluefield-3s.

Is there an issue with timestamping the cqes that I am unaware of, such as yet another bug in the Bluefield-3 firmware?

Thanks.

Hello,

Thanks for your question.
Based on the scenario you described, the further investigation may require collecting logs and other relevant data from the setup the timestamping issue is reproduced on.
In this case we would ask you to open a new support case in Nvidia portal, or just send an email to enterprisesupport@nvidia.com, and this case will be handled according to the entitlement.

Best Regards,
Anatoly