memmap pages  can be allocated either from the memblock (boot) allocator
during early boot or from the buddy allocator.

When these memmap pages are removed via arch_remove_memory(), the
deallocation path depends on their source:

* For pages from the buddy allocator, depopulate_section_memmap() is
  called, which also decrements the count of nr_memmap_pages.

* For pages from the boot allocator, free_map_bootmem() is called. But
  it currently does not adjust the nr_memmap_boot_pages.

To fix this inconsistency, update free_map_bootmem() to also decrement
the nr_memmap_boot_pages count by invoking memmap_boot_pages_add(),
mirroring how free_vmemmap_page() handles this for boot-allocated pages.

This ensures correct tracking of memmap pages regardless of allocation
source.

Cc: stable@vger.kernel.org
Fixes: 15995a352474 ("mm: report per-page metadata information")
Signed-off-by: Sumanth Korikkar <sumanthk@linux.ibm.com>
---
 mm/sparse.c | 1 +
 1 file changed, 1 insertion(+)

diff --git a/mm/sparse.c b/mm/sparse.c
index 3c012cf83cc2..d7c128015397 100644
--- a/mm/sparse.c
+++ b/mm/sparse.c
@@ -688,6 +688,7 @@ static void free_map_bootmem(struct page *memmap)
 	unsigned long start = (unsigned long)memmap;
 	unsigned long end = (unsigned long)(memmap + PAGES_PER_SECTION);
 
+	memmap_boot_pages_add(-1L * (DIV_ROUND_UP(end - start, PAGE_SIZE)));
 	vmemmap_free(start, end, NULL);
 }
 
-- 
2.48.1

On 04.08.25 11:08, Sumanth Korikkar wrote:
> memmap pages  can be allocated either from the memblock (boot) allocator
> during early boot or from the buddy allocator.
> 
> When these memmap pages are removed via arch_remove_memory(), the
> deallocation path depends on their source:
> 
> * For pages from the buddy allocator, depopulate_section_memmap() is
>    called, which also decrements the count of nr_memmap_pages.
> 
> * For pages from the boot allocator, free_map_bootmem() is called. But
>    it currently does not adjust the nr_memmap_boot_pages.
> 
> To fix this inconsistency, update free_map_bootmem() to also decrement
> the nr_memmap_boot_pages count by invoking memmap_boot_pages_add(),
> mirroring how free_vmemmap_page() handles this for boot-allocated pages.
> 
> This ensures correct tracking of memmap pages regardless of allocation
> source.
> 
> Cc: stable@vger.kernel.org
> Fixes: 15995a352474 ("mm: report per-page metadata information")
> Signed-off-by: Sumanth Korikkar <sumanthk@linux.ibm.com>
> ---
>   mm/sparse.c | 1 +
>   1 file changed, 1 insertion(+)
> 
> diff --git a/mm/sparse.c b/mm/sparse.c
> index 3c012cf83cc2..d7c128015397 100644
> --- a/mm/sparse.c
> +++ b/mm/sparse.c
> @@ -688,6 +688,7 @@ static void free_map_bootmem(struct page *memmap)
>   	unsigned long start = (unsigned long)memmap;
>   	unsigned long end = (unsigned long)(memmap + PAGES_PER_SECTION);
>   
> +	memmap_boot_pages_add(-1L * (DIV_ROUND_UP(end - start, PAGE_SIZE)));
>   	vmemmap_free(start, end, NULL);
>   }
>   

Looks good to me. But now I wonder about !CONFIG_SPARSEMEM_VMEMMAP, 
where neither depopulate_section_memmap() nor free_map_bootmem() adjust 
anything?

Which makes me wonder whether we should be moving that to 
section_deactivate().

-- 
Cheers,

David / dhildenb

On Mon, Aug 04, 2025 at 02:27:20PM +0200, David Hildenbrand wrote:
> On 04.08.25 11:08, Sumanth Korikkar wrote:
> > memmap pages  can be allocated either from the memblock (boot) allocator
> > during early boot or from the buddy allocator.
> > 
> > When these memmap pages are removed via arch_remove_memory(), the
> > deallocation path depends on their source:
> > 
> > * For pages from the buddy allocator, depopulate_section_memmap() is
> >    called, which also decrements the count of nr_memmap_pages.
> > 
> > * For pages from the boot allocator, free_map_bootmem() is called. But
> >    it currently does not adjust the nr_memmap_boot_pages.
> > 
> > To fix this inconsistency, update free_map_bootmem() to also decrement
> > the nr_memmap_boot_pages count by invoking memmap_boot_pages_add(),
> > mirroring how free_vmemmap_page() handles this for boot-allocated pages.
> > 
> > This ensures correct tracking of memmap pages regardless of allocation
> > source.
> > 
> > Cc: stable@vger.kernel.org
> > Fixes: 15995a352474 ("mm: report per-page metadata information")
> > Signed-off-by: Sumanth Korikkar <sumanthk@linux.ibm.com>
> > ---
> >   mm/sparse.c | 1 +
> >   1 file changed, 1 insertion(+)
> > 
> > diff --git a/mm/sparse.c b/mm/sparse.c
> > index 3c012cf83cc2..d7c128015397 100644
> > --- a/mm/sparse.c
> > +++ b/mm/sparse.c
> > @@ -688,6 +688,7 @@ static void free_map_bootmem(struct page *memmap)
> >   	unsigned long start = (unsigned long)memmap;
> >   	unsigned long end = (unsigned long)(memmap + PAGES_PER_SECTION);
> > +	memmap_boot_pages_add(-1L * (DIV_ROUND_UP(end - start, PAGE_SIZE)));
> >   	vmemmap_free(start, end, NULL);
> >   }
> 
> Looks good to me. But now I wonder about !CONFIG_SPARSEMEM_VMEMMAP, where
> neither depopulate_section_memmap() nor free_map_bootmem() adjust anything?
> 
> Which makes me wonder whether we should be moving that to
> section_deactivate().

Agree. I will move accounting to section_deactivate() then.

Thanks