Re: [PATCH v3 05/17] mm/sparse-vmemmap: support section-based vmemmap accounting

[Muchun Song <muchun.song@linux.dev>]

On 2026/8/4 11:55, Muchun Song wrote:
> section_nr_vmemmap_pages() can account ordinary sections and DAX sections,
> but section-based vmemmap optimization keeps its compound order in struct
> mem_section and retains a different number of vmemmap pages.
>
> Teach section_nr_vmemmap_pages() to recognize section-based optimized
> sections and calculate their vmemmap page count from the section order
> and the HVO retained page count.
>
> Signed-off-by: Muchun Song <songmuchun@bytedance.com>
> ---
> v3:
> - Add vmemmap_optimizable_order() for order-based optimization checks
>
> v2:
> - Remove an unnecessary vmemmap_can_optimize() call to simplify the code
>    (suggested by Mike Rapoport).
> - Rewrite the commit message for better understanding.
> ---
>   include/linux/mmzone.h |  6 ++++--
>   mm/sparse-vmemmap.c    | 10 ++++++----
>   mm/sparse.h            | 16 ++++++++++++++++
>   3 files changed, 26 insertions(+), 6 deletions(-)
>
> diff --git a/include/linux/mmzone.h b/include/linux/mmzone.h
> index 81e16d71e1f0..663726dd6463 100644
> --- a/include/linux/mmzone.h
> +++ b/include/linux/mmzone.h
> @@ -107,8 +107,10 @@
>   	 is_power_of_2(sizeof(struct page)) ? \
>   	 MAX_FOLIO_NR_PAGES * sizeof(struct page) : 0)
>   
> -/* The number of struct pages covered by the retained vmemmap pages with HVO enabled. */
> -#define VMEMMAP_OPTIMIZATION_NR_STRUCT_PAGES	(PAGE_SIZE / sizeof(struct page))
> +/* The number of retained vmemmap pages with HVO enabled. */
> +#define VMEMMAP_OPTIMIZATION_PAGES		1
> +#define VMEMMAP_OPTIMIZATION_NR_STRUCT_PAGES	\
> +	(VMEMMAP_OPTIMIZATION_PAGES * PAGE_SIZE / sizeof(struct page))
>   #define VMEMMAP_OPTIMIZATION_MIN_ORDER		(ilog2(VMEMMAP_OPTIMIZATION_NR_STRUCT_PAGES) + 1)
>   
>   #define __VMEMMAP_OPTIMIZATION_NR_ORDERS	\
> diff --git a/mm/sparse-vmemmap.c b/mm/sparse-vmemmap.c
> index 107215cf8488..b7abc5494bb9 100644
> --- a/mm/sparse-vmemmap.c
> +++ b/mm/sparse-vmemmap.c
> @@ -649,24 +649,26 @@ void offline_mem_sections(unsigned long start_pfn, unsigned long end_pfn)
>   static int __meminit section_nr_vmemmap_pages(unsigned long pfn, unsigned long nr_pages,
>   		struct vmem_altmap *altmap, struct dev_pagemap *pgmap)
>   {
> -	const unsigned int order = pgmap ? pgmap->vmemmap_shift : 0;
> +	const struct mem_section *ms = __pfn_to_section(pfn);
> +	const int order = pgmap ? pgmap->vmemmap_shift : section_order(ms);
> +	const int vmemmap_pages = pgmap ? VMEMMAP_RESERVE_NR : VMEMMAP_OPTIMIZATION_PAGES;
>   	const unsigned long pages_per_compound = 1UL << order;
>   
>   	VM_WARN_ON_ONCE(!IS_ALIGNED(pfn | nr_pages, PAGES_PER_SUBSECTION));
>   	VM_WARN_ON_ONCE(nr_pages > PAGES_PER_SECTION);
>   
> -	if (!vmemmap_can_optimize(altmap, pgmap))
> +	if (!vmemmap_can_optimize(altmap, pgmap) && !section_vmemmap_optimizable(ms))
>   		return DIV_ROUND_UP(nr_pages * sizeof(struct page), PAGE_SIZE);

Sashiko said this conditional logic mix DAX and HVO optimization checks in
a way that severely corrupts vmemmap accounting for nr_memmap_pages.

This is a false positive because its premise is based on memory hotplug
support for HVO. However, HVO cannot be enabled through memory hotplug at
this time.

Muchun,
Thanks.
>   
>   	if (order < PFN_SECTION_SHIFT) {
>   		VM_WARN_ON_ONCE(!IS_ALIGNED(pfn | nr_pages, pages_per_compound));
> -		return VMEMMAP_RESERVE_NR * nr_pages / pages_per_compound;
> +		return vmemmap_pages * nr_pages / pages_per_compound;
>   	}
>   
>   	VM_WARN_ON_ONCE(!IS_ALIGNED(pfn | nr_pages, PAGES_PER_SECTION));
>   
>   	if (IS_ALIGNED(pfn, pages_per_compound))
> -		return VMEMMAP_RESERVE_NR;
> +		return vmemmap_pages;
>   
>   	return 0;
>   }
> diff --git a/mm/sparse.h b/mm/sparse.h
> index b9b6b47e85ce..6ad190ec48cf 100644
> --- a/mm/sparse.h
> +++ b/mm/sparse.h
> @@ -43,6 +43,17 @@ static inline bool pfn_vmemmap_optimizable(unsigned long pfn)
>   	return (pfn & (nr_pages - 1)) >= VMEMMAP_OPTIMIZATION_NR_STRUCT_PAGES;
>   }
>   
> +static inline bool vmemmap_optimizable_order(unsigned int order)
> +{
> +	if (!IS_ENABLED(CONFIG_HUGETLB_PAGE_OPTIMIZE_VMEMMAP))
> +		return false;
> +
> +	if (!is_power_of_2(sizeof(struct page)))
> +		return false;
> +
> +	return order >= VMEMMAP_OPTIMIZATION_MIN_ORDER;
> +}
> +
>   /*
>    * mm/sparse.c
>    */
> @@ -80,6 +91,11 @@ static inline void __section_mark_present(struct mem_section *ms,
>   
>   	ms->section_mem_map |= SECTION_MARKED_PRESENT;
>   }
> +
> +static inline bool section_vmemmap_optimizable(const struct mem_section *section)
> +{
> +	return vmemmap_optimizable_order(section_order(section));
> +}
>   #else
>   static inline void sparse_init(void) {}
>   #endif /* CONFIG_SPARSEMEM */