Re: [PATCH] mm/sparse: fix BUILD_BUG_ON check for section map alignment

[David Hildenbrand (Arm)]

On 3/31/26 13:30, Muchun Song wrote:
> The comment in mmzone.h states that the alignment requirement
> is the minimum of PAGE_SHIFT and PFN_SECTION_SHIFT. However, the
> pointer arithmetic (mem_map - section_nr_to_pfn()) results in
> a byte offset scaled by sizeof(struct page). Thus, the actual
> alignment provided by the second term is PFN_SECTION_SHIFT +
> __ffs(sizeof(struct page)).
> 
> Update the compile-time check and the mmzone.h comment to
> accurately reflect this mathematically guaranteed alignment by
> taking the minimum of PAGE_SHIFT and PFN_SECTION_SHIFT +
> __ffs(sizeof(struct page)). This avoids the issue of the check
> being overly restrictive on architectures like powerpc where
> PFN_SECTION_SHIFT alone is very small (e.g., 6).
> 
> Also, remove the exhaustive per-architecture bit-width list from the
> comment; such details risk falling out of date over time and may
> inadvertently be left un-updated, while the existing BUILD_BUG_ON
> provides sufficient compile-time verification of the constraint.
> 
> No runtime impact so far: SECTION_MAP_LAST_BIT happens to fit within
> the smaller limit on all existing architectures.
> 
> Fixes: def9b71ee651 ("include/linux/mmzone.h: fix explanation of lower bits in the SPARSEMEM mem_map pointer")
> Signed-off-by: Muchun Song <songmuchun@xxxxxxxxxxxxx>
> ---
>  include/linux/mmzone.h | 24 +++++++++---------------
>  mm/sparse.c            |  3 ++-
>  2 files changed, 11 insertions(+), 16 deletions(-)
> 
> diff --git a/include/linux/mmzone.h b/include/linux/mmzone.h
> index 7bd0134c241c..584fa598ad75 100644
> --- a/include/linux/mmzone.h
> +++ b/include/linux/mmzone.h
> @@ -2073,21 +2073,15 @@ static inline struct mem_section *__nr_to_section(unsigned long nr)
>  extern size_t mem_section_usage_size(void);
>  
>  /*
> - * We use the lower bits of the mem_map pointer to store
> - * a little bit of information.  The pointer is calculated
> - * as mem_map - section_nr_to_pfn(pnum).  The result is
> - * aligned to the minimum alignment of the two values:
> - *   1. All mem_map arrays are page-aligned.
> - *   2. section_nr_to_pfn() always clears PFN_SECTION_SHIFT
> - *      lowest bits.  PFN_SECTION_SHIFT is arch-specific
> - *      (equal SECTION_SIZE_BITS - PAGE_SHIFT), and the
> - *      worst combination is powerpc with 256k pages,
> - *      which results in PFN_SECTION_SHIFT equal 6.
> - * To sum it up, at least 6 bits are available on all architectures.
> - * However, we can exceed 6 bits on some other architectures except
> - * powerpc (e.g. 15 bits are available on x86_64, 13 bits are available
> - * with the worst case of 64K pages on arm64) if we make sure the
> - * exceeded bit is not applicable to powerpc.
> + * We use the lower bits of the mem_map pointer to store a little bit of
> + * information. The pointer is calculated as mem_map - section_nr_to_pfn().
> + * The result is aligned to the minimum alignment of the two values:
> + *
> + * 1. All mem_map arrays are page-aligned.
> + * 2. section_nr_to_pfn() always clears PFN_SECTION_SHIFT lowest bits. Because
> + *    it is subtracted from a struct page pointer, the offset is scaled by
> + *    sizeof(struct page). This provides an alignment of PFN_SECTION_SHIFT +
> + *    __ffs(sizeof(struct page)).
>   */
>  enum {
>  	SECTION_MARKED_PRESENT_BIT,
> diff --git a/mm/sparse.c b/mm/sparse.c
> index dfabe554adf8..c2eb36bfb86d 100644
> --- a/mm/sparse.c
> +++ b/mm/sparse.c
> @@ -269,7 +269,8 @@ static unsigned long sparse_encode_mem_map(struct page *mem_map, unsigned long p
>  {
>  	unsigned long coded_mem_map =
>  		(unsigned long)(mem_map - (section_nr_to_pfn(pnum)));
> -	BUILD_BUG_ON(SECTION_MAP_LAST_BIT > PFN_SECTION_SHIFT);
> +	BUILD_BUG_ON(SECTION_MAP_LAST_BIT > min(PFN_SECTION_SHIFT + __ffs(sizeof(struct page)),
> +						PAGE_SHIFT));

If that would trigger, wouldn't the memmap of a memory section be
smaller than a single page?

Is this really something we should be concerned about? :)

-- 
Cheers,

David