DDR Training of polarfire soc discovery kit reference design

[Tokeidai2]

Hello,

[What I did]
I programed following job file to the PolarFire Discovery Kit.
2024.06 latest
MPFS_DISCOVERY.zip
https://github.com/polarfire-soc/polarfire-soc-discovery-kit-reference-design/releases/tag/v2024.06

[Result]
In Terminal, DDR Training was stopped at 80%

DDR training ...
80% [ ..........]

[Request]
If the file is wrong, I would like to ask Microchip to correct the file.

[For reference]
Others also stop DDR training at 80%.
https://forum.beagleboard.org/t/anyone-tried-the-discovery-kit-from-microchip/38144/45

Thank you for reading.

[FilkerZero]

I have seen this a number of times, but not consistently, on my MPFS-DISCO-KIT using the 2025.03 release of the reference design.

In my case it seems to happen only when the board has been on for a little while, so I was assuming that its caused by a thermal issue with the DDR controller, but I have not done anything to test that conclusion.

[trajcenikolovnick]

@Tokeidai2

We have seen this as well since months ago and we found it might help if you brute-force the cycling over all the allowed values for the reference clock offsets (CK/CA additive offset) , 0 - 7 per the datasheet, to find a stable timing window for the DDR interface..

As a hint you can try this in
File: hart-software-services/baremetal/polarfire-soc-bare-metal-library/src/platform/mpfs_hal/common/nwc/mss_ddr.c
Line: (about 5100)
Function: static uint8_t ddr_manual_addcmd_refclk_offset(DDR_TYPE ddr_type, uint8_t * refclk_sweep_index)

You might also want to do a small research of what these values means, and to arrange/sort these values in some more meaningful fashion that will help the DDR training pass even faster.

Here is also a link that might be related to the issue, we've found it as a possible justification to our solely empirical findings thru debugging:

https://support.microchip.com/s/article/The-CK--CA-additive-offset-value-is-not-propagated-inside-the-DDR-Controller-Sub-System-of-PolarFire

Would be nice if you post back your results if these hints were helpfull with some benchmarks

Cheers,
Nick