Re: [PATCH v7 4/6] mm/sparse-vmemmap: Fix DAX vmemmap accounting with optimization

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

> On Apr 28, 2026, at 15:00, David Hildenbrand (Arm) <david@kernel.org> wrote:
> 
> On 4/28/26 04:21, Muchun Song wrote:
>> 
>> 
>>> On Apr 27, 2026, at 18:17, David Hildenbrand (Arm) <david@kernel.org> wrote:
>>> 
>>> On 4/26/26 11:26, Muchun Song wrote:
>>>> When vmemmap optimization is enabled for DAX, the nr_memmap_pages
>>>> counter in /proc/vmstat is incorrect. The current code always accounts
>>>> for the full, non-optimized vmemmap size, but vmemmap optimization
>>>> reduces the actual number of vmemmap pages by reusing tail pages. This
>>>> causes the system to overcount vmemmap usage, leading to inaccurate
>>>> page statistics in /proc/vmstat.
>>>> 
>>>> Fix this by introducing section_nr_vmemmap_pages(), which returns the exact
>>>> vmemmap page count for a given pfn range based on whether optimization
>>>> is in effect.
>>>> 
>>>> Fixes: 15995a352474 ("mm: report per-page metadata information")
>>>> Cc: stable@vger.kernel.org
>>>> Signed-off-by: Muchun Song <songmuchun@bytedance.com>
>>>> Acked-by: Mike Rapoport (Microsoft) <rppt@kernel.org>
>>>> Acked-by: Oscar Salvador <osalvador@suse.de>
>>>> ---
>>>> v6 -> v7:
>>>> - Refine the alignment assertions in section_nr_vmemmap_pages().
>>>> ---
>>>> mm/sparse-vmemmap.c | 34 ++++++++++++++++++++++++++++++----
>>>> 1 file changed, 30 insertions(+), 4 deletions(-)
>>>> 
>>>> diff --git a/mm/sparse-vmemmap.c b/mm/sparse-vmemmap.c
>>>> index 3340f6d30b01..01f448607bad 100644
>>>> --- a/mm/sparse-vmemmap.c
>>>> +++ b/mm/sparse-vmemmap.c
>>>> @@ -652,6 +652,31 @@ 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 unsigned long pages_per_compound = 1UL << order;
>>>> +
>>>> +  VM_WARN_ON_ONCE(!IS_ALIGNED(pfn | nr_pages, PAGES_PER_SUBSECTION));
>>>> +
>>>> +  if (!vmemmap_can_optimize(altmap, pgmap))
>>>> +  return DIV_ROUND_UP(nr_pages * sizeof(struct page), PAGE_SIZE);
>>>> +
>>>> +  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;
>>>> +  }
>>>> +
>>>> +  VM_WARN_ON_ONCE(!IS_ALIGNED(pfn | nr_pages, PAGES_PER_SECTION));
>>>> +  VM_WARN_ON_ONCE(nr_pages > PAGES_PER_SECTION);
>>> 
>>> I would just have done that at the very top, as this check applies to all cases.
>> 
>> My initial reasoning was that the current formula holds for compound pages smaller
>> than the section size, and we only need to impose limits when the page size exceeds
>> it. While the current callers of section_nr_vmemmap_pages() don't pass sizes larger
>> than a section, this will change in the future (see [1]).
> 
> A function that is called *section_* will get a range that exceeds a section?
> 
> That sounds conceptually wrong, no?

It does seem a bit ambiguous. I will rename it to something more appropriate if I
expand its functionality in the future. For this series, I will update a v8 to move
VM_WARN_ON_ONCE(nr_pages > PAGES_PER_SECTION); to the top of this function.

Thanks.

> 
> 
> -- 
> Cheers,
> 
> David