On Thu, Jan 22, 2026 at 03:47:52PM -0300, Fabio Estevam wrote: > Hi Tom and Marek, > > On Thu, Jan 22, 2026 at 11:23 AM Tom Rini <[email protected]> wrote: > > > We did that, yes. I'm not sure what Fabio is running in to here.. > > Let me share more information with you. > > The board I am working on is based on Rockchip RV1103 (ARMv7) and will > submit it upstream soon.
Can you please share it with my off-list maybe? I'm still confused as to
where the "sram" and "sdram" parts are coming from.
> The board was booting fin last week. Yesterday, after rebasing it
> against the top of the tree U-Boot, the board no longer boots.
>
> The boot error is:
>
> Missing DTB
> spl_early_init() failed: -2
> ### ERROR ### Please RESET the board ###
>
> After running "git bisect", I found Tom's commit 8b0ebe054bb3 ("arm:
> Update linker scripts to ensure appended device tree is aligned") to
> be the guilty one.
>
> If I partially revert the arm/cpu/u-boot-spl.lds change, as shown
> below, the board boots again:
>
> --- a/arch/arm/cpu/u-boot-spl.lds
> +++ b/arch/arm/cpu/u-boot-spl.lds
> @@ -51,9 +51,9 @@ SECTIONS
> __rel_dyn_start = .;
> *(.rel*)
> __rel_dyn_end = .;
> - . = ALIGN(8);
> }
>
> + . = ALIGN(8);
> _image_binary_end = .;
> _end = .;
>
> With Tom's commit applied (failing case), the spl/u-boot-spl.map looks
> like this:
>
> .rel.dyn 0x0000000000015cac 0x4
> 0x0000000000015cac __rel_dyn_start = .
> *(.rel*)
> .rel.iplt 0x0000000000015cac 0x0 arch/arm/cpu/armv7/start.o
> 0x0000000000015cac __rel_dyn_end = .
> 0x0000000000015cb0 . = ALIGN (0x8)
> *fill* 0x0000000000015cac 0x4
> 0x0000000000015cb0 _image_binary_end = .
> 0x0000000000015cb0 _end = .
>
> With the partial revert above (booting case):
>
> .rel.dyn 0x0000000000015cac 0x0
> 0x0000000000015cac __rel_dyn_start = .
> *(.rel*)
> .rel.iplt 0x0000000000015cac 0x0 arch/arm/cpu/armv7/start.o
> 0x0000000000015cac __rel_dyn_end = .
> 0x0000000000015cb0 . = ALIGN (0x8)
> 0x0000000000015cb0 _image_binary_end = .
> 0x0000000000015cb0 _end = .
>
> Basically, in the good case, .rel.dyn is zero. In the failing case:
> .rel.dyn is 0x4 and *fill* is 0x4 as well.
>
> This change messes with dtb being found in SPL.
>
> What would be a proper fix?
Well, where is the dtb ending up in the resulting binary? The addresses
are aligned correctly, but the dtb isn't where it's supposed to be,
where is it instead? A challenge here is that I could only check the
linker script used by about half of the 32bit ARM boards
(arch/arm/mach-omap2/u-boot-spl.lds) and not the one used by the other
half (arch/arm/cpu/u-boot-spl.lds). The initial error you reported makes
me wonder if we can somehow unify these two afterall, which would be
good in general to do.
--
Tom
signature.asc
Description: PGP signature
