Re: [PATCH v2 05/24] vmlinux.lds.h: Fix ALIGN(8) omission causing NULL ptr on i386

[sashiko-bot@kernel.org]

Thank you for your contribution! Sashiko AI review found 2 potential issue(s) to consider:
- [High] The implicit 8-byte alignment added to `BOUNDED_SECTION` macros corrupts the `.notes` ELF section by injecting padding at the start, breaking `PT_NOTE` parsing.
- [Low] The patch fails to apply the `_ALIGNED` variant to `.BTF_ids`, violating the commit message's stated refactoring rules.
--

commit 8b6f5abfb8521b62106505848b2663ec04bac566
Author: Jim Cromie <jim.cromie@gmail.com>
Subject: vmlinux.lds.h: Fix ALIGN(8) omission causing NULL ptr on i386

This commit explicitly adds 8-byte alignment to BOUNDED_SECTION macros to
prevent unaligned section pointers. It also introduces _ALIGNED variants
for manually aligned sections and updates existing macros to use them.

Link: https://lore.kernel.org/lkml/177402491426.6181.12855763650074831089.b4-review@b4/

> diff --git a/include/asm-generic/bounded_sections.lds.h b/include/asm-generic/bounded_sections.lds.h
> index 268cdc34389b0..8ff3e3420f60a 100644
> --- a/include/asm-generic/bounded_sections.lds.h
> +++ b/include/asm-generic/bounded_sections.lds.h
[ ... ]
> -#define BOUNDED_SECTION_PRE_LABEL(_sec_, _label_, _BEGIN_, _END_)	\
> +#define BOUNDED_SECTION_PRE_LABEL_ALIGNED(_sec_, _label_, _BEGIN_, _END_, _ALIGNED_) \
> +	. = ALIGN(_ALIGNED_);						\
>  	_BEGIN_##_label_ = .;						\
>  	KEEP(*(_sec_))							\
>  	_END_##_label_ = .;
>  
> -#define BOUNDED_SECTION_POST_LABEL(_sec_, _label_, _BEGIN_, _END_)	\
> +#define BOUNDED_SECTION_PRE_LABEL(_sec_, _label_, _BEGIN_, _END_)	\
> +	BOUNDED_SECTION_PRE_LABEL_ALIGNED(_sec_, _label_, _BEGIN_, _END_, 8)

[Severity: High]
Does this implicit 8-byte alignment break the .notes ELF section by
injecting padding?

The NOTES macro in vmlinux.lds.h defines the .notes output section, which
inherits a 4-byte alignment from its input *(.note.*) sections. It often
follows RO_EXCEPTION_TABLE, which can end on a 4-byte boundary.

If the output section VMA is not 8-byte aligned, this new explicit
. = ALIGN(8) inside the BOUNDED_SECTION_BY macro will inject 4 bytes of
zero padding at the beginning of the .notes section.

This padding can be misinterpreted as an empty note header by external
tools and the kernel, breaking PT_NOTE parsing. Should the NOTES macro be
updated to use BOUNDED_SECTION_BY_ALIGNED(.note.*, _notes, 4)?

> diff --git a/include/asm-generic/vmlinux.lds.h b/include/asm-generic/vmlinux.lds.h
> index 9c61dd083f26a..477333bdf3d35 100644
> --- a/include/asm-generic/vmlinux.lds.h
> +++ b/include/asm-generic/vmlinux.lds.h
[ ... ]
> @@ -650,7 +650,7 @@
>  #define BTF								\
>  	. = ALIGN(PAGE_SIZE);						\
>  	.BTF : AT(ADDR(.BTF) - LOAD_OFFSET) {				\
> -		BOUNDED_SECTION_BY(.BTF, _BTF)				\
> +		BOUNDED_SECTION_BY_ALIGNED(.BTF, _BTF, PAGE_SIZE)	\
>  	}								\
>  	. = ALIGN(PAGE_SIZE);						\
>  	.BTF_ids : AT(ADDR(.BTF_ids) - LOAD_OFFSET) {			\

[Severity: Low]
The commit message states "update BOUNDED_SECTION uses inside . = ALIGN(x)
stanzas to use _ALIGNED variants".

Was .BTF_ids intentionally omitted from this rule?

It sits right below .BTF and also has an outer . = ALIGN(PAGE_SIZE), but
still uses the base BOUNDED_SECTION_BY macro. This doesn't cause functional
issues since it is already page-aligned, but it leaves the macro usage
inconsistent with the stated refactoring rules.

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20260523-dd-maint-2-v2-0-b937312aa083@gmail.com?part=5