[PATCH v2] mx6sabresd: Reduce U-Boot proper size to fix boot regression

[Tom Rini]

On Fri, Jul 04, 2025 at 01:40:40PM -0300, Fabio Estevam wrote:
> On Fri, Jul 4, 2025 at 12:56 PM Tom Rini <trini at konsulko.com> wrote:
> >
> > On Fri, Jul 04, 2025 at 11:40:22AM -0300, Fabio Estevam wrote:
> >
> > > The mx6sabresd U-Boot proper binary size has grown beyond the
> > > CONFIG_BOARD_SIZE_LIMIT.
> >
> > So, is CONFIG_BOARD_SIZE_LIMIT set incorrectly then and needs to be
> > lowered?
> 
> The current CONFIG_BOARD_SIZE_LIMIT is 715766.
> 
> This value is CONFIG_ENV_OFFSET (0xC0000) - the offset for U-Boot
> proper (69k) = 786432 − 69×1024 = 715776.
> 
> The calculation itself is correct.
> 
> The first issue: CI does not report a build failure. However, my local
> build reports:
> 
> u-boot.img exceeds file size limit:
>   limit:  0xaebf6 bytes
>   actual: 0xaed70 bytes
>   excess: 0x17a bytes
> make: *** [Makefile:1501: u-boot.img] Error 1
> make: *** Deleting file 'u-boot.img'
> make: *** Waiting for unfinished jobs....
> u-boot-dtb.img exceeds file size limit:
>   limit:  0xaebf6 bytes
>   actual: 0xaed70 bytes
>   excess: 0x17a bytes
> make: *** [Makefile:1501: u-boot-dtb.img] Error 1
> make: *** Deleting file 'u-boot-dtb.img'
> 
> The second issue is that even if I remove the
> CONFIG_HAS_BOARD_SIZE_LIMIT/CONFIG_BOARD_SIZE_LIMIT, it builds, but
> fails to boot like this:

Ah, so toolchain mismatch problems. Newer GCC optimizes better.

> U-Boot SPL 2025.07-rc5-00012-g7027b445cc0b-dirty (Jul 04 2025 - 13:22:16 -0300)
> Trying to boot from MMC1
> 
> This boot failure is related to the initcall series.
> 
> To avoid the mx6abresd boot regression and play safe for the upcoming
> 2025.07, I'm more inclined to go with the previous patch version that
> increases CONFIG_BOARD_SIZE_LIMIT.
> 
> Jerome said he is working on a better fix, which reduces the size by
> 1702 bytes for sabresd.
> 
> Having CONFIG_BOARD_SIZE_LIMIT with too low a headroom may cause a
> problem where, on some build systems, such as U-Boot CI, the build
> succeeds, while others, like my local build, fail.
> 
> Let's provide more CONFIG_BOARD_SIZE_LIMIT headroom to avoid this
> situation from happening again.
> 
> If you agree, feel free to pick v1.

I think Jerome has the wrong idea here. The previous problem was on
armv5 CPUs, armv7 CPUs are thumb2 and aren't tripping this problem. I
suspect that if you used gcc-14.2.0 from kernel.org to build, current
top of tree mainline would work. I think we should still take this patch
as it's both correct and saves much more space so that we aren't quite
so close and so dependent on gcc optimizations.

-- 
Tom
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 659 bytes
Desc: not available
URL: <https://lists.denx.de/pipermail/u-boot/attachments/20250704/742ffed0/attachment.sig>