Configure Memory Down

The term of Memory Down is used to describe when memory components such as DRAM devices are physically soldered onto a Printed Circuit Board (PCB). This is an alternative to using mechanical connectors to attach memory modules (DIMM) onto a system. Memory down configuration is often found in the embedded platforms due to the variety of constraints and usage models.

SBL supports memory down configuration via FSP UPDs. This guide provides some basic steps to configure SBL before calling FSP-M to initialize system memory.

Step 1 - Read schematic and understand DDR memory layout and model from datasheet.

Step 2 - Configure the following memory parameters in SBL

Start with <platform>\CfgData\CfgData_Memory.yaml to understand the possible values for memory parameters

For UP2, open CfgData_Ext_Up2.dlt and customize values that match the actual memory parameters. Given an example for 8GB LPDDR4 memory:

...

PLATFORMID_CFG_DATA.PlatformId           | 0x000E <-- Match actual Board ID

MEMORY_CFG_DATA.DualRankSupportEnable    | 0x1
MEMORY_CFG_DATA.RmtMode                  | 0x0
MEMORY_CFG_DATA.MemorySizeLimit          | 0x0
MEMORY_CFG_DATA.Ch0_RankEnable           | 0x3  <-- bit masks for dual Rank. 0x1 for single rank; 0x0 for no rank (no memory chip)
MEMORY_CFG_DATA.Ch0_DramDensity          | 0x2  <-- 8Gb (check datasheet)
MEMORY_CFG_DATA.Ch1_RankEnable           | 0x3
MEMORY_CFG_DATA.Ch1_DramDensity          | 0x2
MEMORY_CFG_DATA.Ch2_RankEnable           | 0x3
MEMORY_CFG_DATA.Ch2_DramDensity          | 0x2
MEMORY_CFG_DATA.Ch3_RankEnable           | 0x3
MEMORY_CFG_DATA.Ch3_DramDensity          | 0x2
MEMORY_CFG_DATA.RmtCheckRun              | 0x3
...

Optionally, you can use Configuration Editor to graphically view and modify configurations. See https://slimbootloader.github.io/developer-guides/configuration.html for details.

Note

If FSP UPD implementation supports hardcoded SPD table, simply replacing the SPD binary file under <platform>\CfgData. The file name of a SPD binary can be found in <platform>\CfgData\CfgData_MemSpd.yaml. If a platform supports more than one type of memory configuration, the MEMORY_CFG_DATA.SpdDataSel must be carefully set. Take Tiger Lake as an example: Spd_Ddrlp4.bin and Spd_Ddrlp5.bin are for LPDDR4 and LPDDR5 respectivly. The index of Spd_Ddrlp4.bin is 1, and it is 2 for Spd_Ddrlp5.bin.

In a board dlt file:

MEMORY_CFG_DATA.SpdAddressTable  | {..., 0, ... } <-- when 0, FSP will use hardcoded SPD for the memory slot (mem controller X, channel Y, Dimm Z). Otherwise, FSP reads its SPD from the smbus address.

MEMORY_CFG_DATA.SpdDataSelXYZ    | 1  <-- using LPDDR4 hardcoded SPD for memory slot XYZ. Or

MEMORY_CFG_DATA.SpdDataSelXYZ    | 2  <-- using LPDDR5

Step 3 - Build, stitch and test

If the memory configuration is correct, SBL should boot all the way to the shell. Verify the memory map information from the log to ensure the memory size matches with hardware.

Otherwise, SBL may hang after calling FSP-M and never returns. In this case, you have two troubleshooting options:

  1. Re-examine the configuration values, make changes and repeat.

  2. Object FSP debug build and get debugging output message during memory training flow.

Note

Please contact Intel representatives for how to obtain FSP source code.