Re: [PATCH V4 4/4] arm64/mm: Use pgdp_get() for PGD accesses

[sashiko-bot@kernel.org]

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
- [High] arm64/mm: KASAN fault in p4d_offset_phys() during kasan_early_init()
--

commit e9bf68a7a555b3c58c2a0a8736f337dde8136ab9
Author: Anshuman Khandual <anshuman.khandual@arm.com>

arm64/mm: Use pgdp_get() for PGD accesses

This commit replaces READ_ONCE() with pgdp_get() for PGD accesses and moves
p4d_offset() and p4d_offset_phys() from arch/arm64/include/asm/pgtable.h
into arch/arm64/mm/mmu.c to avoid header include dependencies.

> diff --git a/arch/arm64/mm/mmu.c b/arch/arm64/mm/mmu.c
> index 5b5a85a0527d6..11015e1deb32f 100644
> --- a/arch/arm64/mm/mmu.c
> +++ b/arch/arm64/mm/mmu.c
[ ... ]
> @@ -2463,3 +2463,17 @@ pud_t *pud_offset(p4d_t *p4dp, unsigned long addr)
>  	return pud_offset_lockless(p4dp, p4dp_get(p4dp), addr);
>  }
>  #endif /* CONFIG_PGTABLE_LEVELS > 3 */
> +
> +#if CONFIG_PGTABLE_LEVELS > 4
> +phys_addr_t p4d_offset_phys(pgd_t *pgdp, unsigned long addr)
> +{
> +	VM_WARN_ON_ONCE(!pgtable_l5_enabled());
> +
> +	return pgd_page_paddr(pgdp_get(pgdp)) + p4d_index(addr) * sizeof(p4d_t);
> +}

[Severity: High]
Does moving p4d_offset_phys() (and pud_offset_phys() in the preceding patch)
out-of-line into arch/arm64/mm/mmu.c cause a translation fault during early
boot when KASAN is enabled with 5-level (or 4-level) page tables?

During early boot, __primary_switched calls kasan_early_init() before the
KASAN shadow region is mapped. Because any KASAN shadow check will fault
until kasan_early_init() finishes populating the early shadow page tables,
arch/arm64/mm/Makefile disables KASAN sanitization for kasan_init.o:

    KASAN_SANITIZE_kasan_init.o := n

and set_swapper_pgd() in arch/arm64/mm/mmu.c is marked noinstr for the same
reason.

When kasan_early_init() populates the shadow page tables, it calls into
p4d_offset_phys() and pud_offset_phys():

    kasan_early_init()
      kasan_pgd_populate(..., early=true)
        kasan_p4d_populate(..., early=true)
          kasan_p4d_offset(..., early=true)
            p4d_offset_kimg()
              p4d_offset_phys()
          kasan_pud_populate(..., early=true)
            kasan_pud_offset(..., early=true)
              pud_offset_kimg()
                pud_offset_phys()

Previously, p4d_offset_phys() and pud_offset_phys() were static inline
functions in arch/arm64/include/asm/pgtable.h, so calls from kasan_init.c
were inlined into kasan_init.o and compiled without KASAN instrumentation.

Now that both functions live in arch/arm64/mm/mmu.c, which is KASAN-sanitized,
the compiler emits KASAN shadow checks for their memory accesses (such as
pgdp_get(pgdp), p4dp_get(p4dp), and vabits_actual) before the early shadow
memory is mapped.

Should p4d_offset_phys() and pud_offset_phys() be marked noinstr or
__no_sanitize_address?

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20261105043403.3326445-1-anshuman.khandual@arm.com?part=4