Re: [PATCH v1 6/8] configs: cyclone5: Enable SPL FAT support

[Jan Kiszka <jan.kiszka@siemens.com>]

On 27.11.25 08:20, Sune Brian wrote:
>> I shared our wks file [1]: We create a raw partition in a legacy msdos
>> table, set its system ID to 0xa2 (via [2] when we use a hybrid partition
>> table, or classically [3] before that), and put u-boot-with-spl.sfp into
>> it. SPL is picked up by the romloader, but SPL in upstream completely
>> fails to find u-boot proper in that partition. And that is first of all
>> because SYS_MMCSD_RAW_MODE_U_BOOT_USE_SECTOR is enabled by default which
>> ignores any partitioning and only uses a hard-coded sector. If one
>> switches to SYS_MMCSD_RAW_MODE_U_BOOT_USE_PARTITION_TYPE mode, other
>> bugs after 2a00d73d081a surface like no longer built code or wrong
>> offsets inside the partition, and you also end up without u-boot proper
>> being found. Whatever other partitions or partition tables exist on the
>> SD card does not matter for SPL loading u-boot proper.
> 
> Ha Now I see what you encounter.
> No wonder. Because it is not supposed to run like that.
> (i am no software experts)
> 
> I might be wrong, no shame on that I do simple software only.
> 
> So what T.F. had mentioned all aligned will all my build and
> no additional patch is required from first place.
> 
> All there TYPE, #, OFFSET can work.
> because it never uses the old method of combining.
> 
> A2 <- only SPL
> FAT (boot) <- u-boot.img
> 
> Actually I mentioned this yesterday in our discussion. Maybe you
> skipped that important information.
> 
> All board defconfig after 2025.05 is outdated; no one syncs those.
> 
> That's why I comment that the file "u-boot-with-spl.sfp" is misleading.
> You have to turn on "FAT filesystems" change MMC boot partition
> accordingly.

This is not how upstream is currently configured. And upstream is the
reference because it not only sees vendor-related changes and decisions
but also other refactorings. And all those need to be kept in stable
state, irrespective of some random downstream trees. Actually, some of
the breakages I'm describing and trying to fix are not socfpga-only but
are conceptually affecting all boards that may use raw mmc loading today
or decide to do so in the future.

> 
> This is why I say this patch is not a must and combining SPL+UBOOT
> to A2 is not really needed but only for backward build flow compatibility
> concerns.

It is ok to work around current issues in upstream, but we still need to
fix raw mode loading - or drop it from the tree if it has no value
anymore (not my decision).

Jan

-- 
Siemens AG, Foundational Technologies
Linux Expert Center