[PATCH v4 0/5] mm: Fix vmemmap optimization accounting and initialization

[PATCH v4 0/5] mm: Fix vmemmap optimization accounting and initialization

[Muchun Song]

The series fixes several bugs in vmemmap optimization, mainly around
incorrect page accounting and memmap initialization in DAX and memory
hotplug paths. It also fixes pageblock migratetype initialization and
struct page initialization for ZONE_DEVICE compound pages.

The first three patches fix vmemmap accounting issues. The first patch
fixes an accounting underflow in the section activation failure path.
The second patch passes pgmap through memory deactivation paths so the
teardown side can determine whether vmemmap optimization was in effect.
The third patch uses that information to account the optimized DAX
vmemmap size correctly.

The last two patches fix initialization issues in mm/mm_init. One makes
sure all pageblocks in ZONE_DEVICE compound pages get their migratetype
initialized. The other fixes a case where DAX memory hotplug reuses an
unoptimized early-section memmap while compound_nr_pages() still assumes
vmemmap optimization, leaving tail struct pages uninitialized.

v3 -> v4:
- Collect Acked-by and Reviewed-by tags from Oscar Salvador.
- Fix the build issue reported by Sashiko.
- Add a fix for uninitialized struct pages for ZONE_DEVICE, based on
  analysis from Sashiko and verified locally.

Muchun Song (5):
  mm/sparse-vmemmap: Fix vmemmap accounting underflow
  mm/sparse-vmemmap: Pass @pgmap argument to memory deactivation paths
  mm/sparse-vmemmap: Fix DAX vmemmap accounting with optimization
  mm/mm_init: Fix pageblock migratetype for ZONE_DEVICE compound pages
  mm/mm_init: Fix uninitialized struct pages for ZONE_DEVICE

 arch/arm64/mm/mmu.c            |  5 +--
 arch/loongarch/mm/init.c       |  5 +--
 arch/powerpc/mm/mem.c          |  5 +--
 arch/riscv/mm/init.c           |  5 +--
 arch/s390/mm/init.c            |  5 +--
 arch/x86/mm/init_64.c          |  5 +--
 include/linux/memory_hotplug.h |  8 +++--
 mm/memory_hotplug.c            | 13 ++++----
 mm/memremap.c                  |  4 +--
 mm/mm_init.c                   | 58 +++++++++++++++++++++++-----------
 mm/sparse-vmemmap.c            | 56 ++++++++++++++++++++++++--------
 11 files changed, 114 insertions(+), 55 deletions(-)

-- 
2.20.1

[PATCH v4 1/5] mm/sparse-vmemmap: Fix vmemmap accounting underflow

[Muchun Song]

In section_activate(), if populate_section_memmap() fails, the error
handling path calls section_deactivate() to roll back the state. This
causes a vmemmap accounting imbalance.

Since commit c3576889d87b ("mm: fix accounting of memmap pages"),
memmap pages are accounted for only after populate_section_memmap()
succeeds. However, the failure path unconditionally calls
section_deactivate(), which decreases the vmemmap count. Consequently,
a failure in populate_section_memmap() leads to an accounting underflow,
incorrectly reducing the system's tracked vmemmap usage.

Fix this more thoroughly by moving all accounting calls into the lower
level functions that actually perform the vmemmap allocation and freeing:

  - populate_section_memmap() accounts for newly allocated vmemmap pages
  - depopulate_section_memmap() unaccounts when vmemmap is freed

This ensures proper accounting in all code paths, including error
handling and early section cases.

Fixes: c3576889d87b ("mm: fix accounting of memmap pages")
Signed-off-by: Muchun Song <songmuchun@bytedance.com>
Acked-by: Mike Rapoport (Microsoft) <rppt@kernel.org>
Acked-by: Oscar Salvador <osalvador@suse.de>
---
 mm/sparse-vmemmap.c | 20 ++++++++++++--------
 1 file changed, 12 insertions(+), 8 deletions(-)

diff --git a/mm/sparse-vmemmap.c b/mm/sparse-vmemmap.c
index 6eadb9d116e4..a7b11248b989 100644
--- a/mm/sparse-vmemmap.c
+++ b/mm/sparse-vmemmap.c
@@ -656,7 +656,12 @@ static struct page * __meminit populate_section_memmap(unsigned long pfn,
 		unsigned long nr_pages, int nid, struct vmem_altmap *altmap,
 		struct dev_pagemap *pgmap)
 {
-	return __populate_section_memmap(pfn, nr_pages, nid, altmap, pgmap);
+	struct page *page = __populate_section_memmap(pfn, nr_pages, nid, altmap,
+						      pgmap);
+
+	memmap_pages_add(DIV_ROUND_UP(nr_pages * sizeof(struct page), PAGE_SIZE));
+
+	return page;
 }
 
 static void depopulate_section_memmap(unsigned long pfn, unsigned long nr_pages,
@@ -665,13 +670,17 @@ static void depopulate_section_memmap(unsigned long pfn, unsigned long nr_pages,
 	unsigned long start = (unsigned long) pfn_to_page(pfn);
 	unsigned long end = start + nr_pages * sizeof(struct page);
 
+	memmap_pages_add(-1L * (DIV_ROUND_UP(nr_pages * sizeof(struct page), PAGE_SIZE)));
 	vmemmap_free(start, end, altmap);
 }
+
 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(PAGES_PER_SECTION * sizeof(struct page),
+						  PAGE_SIZE)));
 	vmemmap_free(start, end, NULL);
 }
 
@@ -774,14 +783,10 @@ static void section_deactivate(unsigned long pfn, unsigned long nr_pages,
 	 * The memmap of early sections is always fully populated. See
 	 * section_activate() and pfn_valid() .
 	 */
-	if (!section_is_early) {
-		memmap_pages_add(-1L * (DIV_ROUND_UP(nr_pages * sizeof(struct page), PAGE_SIZE)));
+	if (!section_is_early)
 		depopulate_section_memmap(pfn, nr_pages, altmap);
-	} else if (memmap) {
-		memmap_boot_pages_add(-1L * (DIV_ROUND_UP(nr_pages * sizeof(struct page),
-							  PAGE_SIZE)));
+	else if (memmap)
 		free_map_bootmem(memmap);
-	}
 
 	if (empty)
 		ms->section_mem_map = (unsigned long)NULL;
@@ -826,7 +831,6 @@ static struct page * __meminit section_activate(int nid, unsigned long pfn,
 		section_deactivate(pfn, nr_pages, altmap);
 		return ERR_PTR(-ENOMEM);
 	}
-	memmap_pages_add(DIV_ROUND_UP(nr_pages * sizeof(struct page), PAGE_SIZE));
 
 	return memmap;
 }
-- 
2.20.1

[PATCH v4 2/5] mm/sparse-vmemmap: Pass @pgmap argument to memory deactivation paths

[Muchun Song]

Currently, the memory hot-remove call chain -- arch_remove_memory(),
__remove_pages(), sparse_remove_section() and section_deactivate() --
does not carry the struct dev_pagemap pointer. This prevents the lower
levels from knowing whether the section was originally populated with
vmemmap optimizations (e.g., DAX with vmemmap optimization enabled).

Without this information, we cannot call vmemmap_can_optimize() to
determine if the vmemmap pages were optimized. As a result, the vmemmap
page accounting during teardown will mistakenly assume a non-optimized
allocation, leading to incorrect memmap statistics.

To lay the groundwork for fixing the vmemmap page accounting, we need
to pass the @pgmap pointer down to the deactivation location. Plumb the
@pgmap argument through the APIs of arch_remove_memory(), __remove_pages()
and sparse_remove_section(), mirroring the corresponding *_activate()
paths.

Signed-off-by: Muchun Song <songmuchun@bytedance.com>
Acked-by: Mike Rapoport (Microsoft) <rppt@kernel.org>
Reviewed-by: Oscar Salvador <osalvador@suse.de>
---
 arch/arm64/mm/mmu.c            |  5 +++--
 arch/loongarch/mm/init.c       |  5 +++--
 arch/powerpc/mm/mem.c          |  5 +++--
 arch/riscv/mm/init.c           |  5 +++--
 arch/s390/mm/init.c            |  5 +++--
 arch/x86/mm/init_64.c          |  5 +++--
 include/linux/memory_hotplug.h |  8 +++++---
 mm/memory_hotplug.c            | 13 +++++++------
 mm/memremap.c                  |  4 ++--
 mm/sparse-vmemmap.c            | 12 ++++++------
 10 files changed, 38 insertions(+), 29 deletions(-)

diff --git a/arch/arm64/mm/mmu.c b/arch/arm64/mm/mmu.c
index dd85e093ffdb..e5a42b7a0160 100644
--- a/arch/arm64/mm/mmu.c
+++ b/arch/arm64/mm/mmu.c
@@ -2024,12 +2024,13 @@ int arch_add_memory(int nid, u64 start, u64 size,
 	return ret;
 }
 
-void arch_remove_memory(u64 start, u64 size, struct vmem_altmap *altmap)
+void arch_remove_memory(u64 start, u64 size, struct vmem_altmap *altmap,
+			struct dev_pagemap *pgmap)
 {
 	unsigned long start_pfn = start >> PAGE_SHIFT;
 	unsigned long nr_pages = size >> PAGE_SHIFT;
 
-	__remove_pages(start_pfn, nr_pages, altmap);
+	__remove_pages(start_pfn, nr_pages, altmap, pgmap);
 	__remove_pgd_mapping(swapper_pg_dir, __phys_to_virt(start), size);
 }
 
diff --git a/arch/loongarch/mm/init.c b/arch/loongarch/mm/init.c
index 00f3822b6e47..c9c57f08fa2c 100644
--- a/arch/loongarch/mm/init.c
+++ b/arch/loongarch/mm/init.c
@@ -86,7 +86,8 @@ int arch_add_memory(int nid, u64 start, u64 size, struct mhp_params *params)
 	return ret;
 }
 
-void arch_remove_memory(u64 start, u64 size, struct vmem_altmap *altmap)
+void arch_remove_memory(u64 start, u64 size, struct vmem_altmap *altmap,
+			struct dev_pagemap *pgmap)
 {
 	unsigned long start_pfn = start >> PAGE_SHIFT;
 	unsigned long nr_pages = size >> PAGE_SHIFT;
@@ -95,7 +96,7 @@ void arch_remove_memory(u64 start, u64 size, struct vmem_altmap *altmap)
 	/* With altmap the first mapped page is offset from @start */
 	if (altmap)
 		page += vmem_altmap_offset(altmap);
-	__remove_pages(start_pfn, nr_pages, altmap);
+	__remove_pages(start_pfn, nr_pages, altmap, pgmap);
 }
 #endif
 
diff --git a/arch/powerpc/mm/mem.c b/arch/powerpc/mm/mem.c
index 648d0c5602ec..4c1afab91996 100644
--- a/arch/powerpc/mm/mem.c
+++ b/arch/powerpc/mm/mem.c
@@ -158,12 +158,13 @@ int __ref arch_add_memory(int nid, u64 start, u64 size,
 	return rc;
 }
 
-void __ref arch_remove_memory(u64 start, u64 size, struct vmem_altmap *altmap)
+void __ref arch_remove_memory(u64 start, u64 size, struct vmem_altmap *altmap,
+			      struct dev_pagemap *pgmap)
 {
 	unsigned long start_pfn = start >> PAGE_SHIFT;
 	unsigned long nr_pages = size >> PAGE_SHIFT;
 
-	__remove_pages(start_pfn, nr_pages, altmap);
+	__remove_pages(start_pfn, nr_pages, altmap, pgmap);
 	arch_remove_linear_mapping(start, size);
 }
 #endif
diff --git a/arch/riscv/mm/init.c b/arch/riscv/mm/init.c
index decd7df40fa4..b0092fb842a3 100644
--- a/arch/riscv/mm/init.c
+++ b/arch/riscv/mm/init.c
@@ -1717,9 +1717,10 @@ int __ref arch_add_memory(int nid, u64 start, u64 size, struct mhp_params *param
 	return ret;
 }
 
-void __ref arch_remove_memory(u64 start, u64 size, struct vmem_altmap *altmap)
+void __ref arch_remove_memory(u64 start, u64 size, struct vmem_altmap *altmap,
+			      struct dev_pagemap *pgmap)
 {
-	__remove_pages(start >> PAGE_SHIFT, size >> PAGE_SHIFT, altmap);
+	__remove_pages(start >> PAGE_SHIFT, size >> PAGE_SHIFT, altmap, pgmap);
 	remove_linear_mapping(start, size);
 	flush_tlb_all();
 }
diff --git a/arch/s390/mm/init.c b/arch/s390/mm/init.c
index 1f72efc2a579..11a689423440 100644
--- a/arch/s390/mm/init.c
+++ b/arch/s390/mm/init.c
@@ -276,12 +276,13 @@ int arch_add_memory(int nid, u64 start, u64 size,
 	return rc;
 }
 
-void arch_remove_memory(u64 start, u64 size, struct vmem_altmap *altmap)
+void arch_remove_memory(u64 start, u64 size, struct vmem_altmap *altmap,
+			struct dev_pagemap *pgmap)
 {
 	unsigned long start_pfn = start >> PAGE_SHIFT;
 	unsigned long nr_pages = size >> PAGE_SHIFT;
 
-	__remove_pages(start_pfn, nr_pages, altmap);
+	__remove_pages(start_pfn, nr_pages, altmap, pgmap);
 	vmem_remove_mapping(start, size);
 }
 #endif /* CONFIG_MEMORY_HOTPLUG */
diff --git a/arch/x86/mm/init_64.c b/arch/x86/mm/init_64.c
index df2261fa4f98..77b889b71cf3 100644
--- a/arch/x86/mm/init_64.c
+++ b/arch/x86/mm/init_64.c
@@ -1288,12 +1288,13 @@ kernel_physical_mapping_remove(unsigned long start, unsigned long end)
 	remove_pagetable(start, end, true, NULL);
 }
 
-void __ref arch_remove_memory(u64 start, u64 size, struct vmem_altmap *altmap)
+void __ref arch_remove_memory(u64 start, u64 size, struct vmem_altmap *altmap,
+			      struct dev_pagemap *pgmap)
 {
 	unsigned long start_pfn = start >> PAGE_SHIFT;
 	unsigned long nr_pages = size >> PAGE_SHIFT;
 
-	__remove_pages(start_pfn, nr_pages, altmap);
+	__remove_pages(start_pfn, nr_pages, altmap, pgmap);
 	kernel_physical_mapping_remove(start, start + size);
 }
 #endif /* CONFIG_MEMORY_HOTPLUG */
diff --git a/include/linux/memory_hotplug.h b/include/linux/memory_hotplug.h
index 815e908c4135..7c9d66729c60 100644
--- a/include/linux/memory_hotplug.h
+++ b/include/linux/memory_hotplug.h
@@ -135,9 +135,10 @@ static inline bool movable_node_is_enabled(void)
 	return movable_node_enabled;
 }
 
-extern void arch_remove_memory(u64 start, u64 size, struct vmem_altmap *altmap);
+extern void arch_remove_memory(u64 start, u64 size, struct vmem_altmap *altmap,
+			       struct dev_pagemap *pgmap);
 extern void __remove_pages(unsigned long start_pfn, unsigned long nr_pages,
-			   struct vmem_altmap *altmap);
+			   struct vmem_altmap *altmap, struct dev_pagemap *pgmap);
 
 /* reasonably generic interface to expand the physical pages */
 extern int __add_pages(int nid, unsigned long start_pfn, unsigned long nr_pages,
@@ -307,7 +308,8 @@ extern int sparse_add_section(int nid, unsigned long pfn,
 		unsigned long nr_pages, struct vmem_altmap *altmap,
 		struct dev_pagemap *pgmap);
 extern void sparse_remove_section(unsigned long pfn, unsigned long nr_pages,
-				  struct vmem_altmap *altmap);
+				  struct vmem_altmap *altmap,
+				  struct dev_pagemap *pgmap);
 extern struct zone *zone_for_pfn_range(enum mmop online_type,
 		int nid, struct memory_group *group, unsigned long start_pfn,
 		unsigned long nr_pages);
diff --git a/mm/memory_hotplug.c b/mm/memory_hotplug.c
index 2a943ec57c85..ec1a86b12477 100644
--- a/mm/memory_hotplug.c
+++ b/mm/memory_hotplug.c
@@ -576,6 +576,7 @@ void remove_pfn_range_from_zone(struct zone *zone,
  * @pfn: starting pageframe (must be aligned to start of a section)
  * @nr_pages: number of pages to remove (must be multiple of section size)
  * @altmap: alternative device page map or %NULL if default memmap is used
+ * @pgmap: device page map or %NULL if not ZONE_DEVICE
  *
  * Generic helper function to remove section mappings and sysfs entries
  * for the section of the memory we are removing. Caller needs to make
@@ -583,7 +584,7 @@ void remove_pfn_range_from_zone(struct zone *zone,
  * calling offline_pages().
  */
 void __remove_pages(unsigned long pfn, unsigned long nr_pages,
-		    struct vmem_altmap *altmap)
+		    struct vmem_altmap *altmap, struct dev_pagemap *pgmap)
 {
 	const unsigned long end_pfn = pfn + nr_pages;
 	unsigned long cur_nr_pages;
@@ -598,7 +599,7 @@ void __remove_pages(unsigned long pfn, unsigned long nr_pages,
 		/* Select all remaining pages up to the next section boundary */
 		cur_nr_pages = min(end_pfn - pfn,
 				   SECTION_ALIGN_UP(pfn + 1) - pfn);
-		sparse_remove_section(pfn, cur_nr_pages, altmap);
+		sparse_remove_section(pfn, cur_nr_pages, altmap, pgmap);
 	}
 }
 
@@ -1425,7 +1426,7 @@ static void remove_memory_blocks_and_altmaps(u64 start, u64 size)
 
 		remove_memory_block_devices(cur_start, memblock_size);
 
-		arch_remove_memory(cur_start, memblock_size, altmap);
+		arch_remove_memory(cur_start, memblock_size, altmap, NULL);
 
 		/* Verify that all vmemmap pages have actually been freed. */
 		WARN(altmap->alloc, "Altmap not fully unmapped");
@@ -1468,7 +1469,7 @@ static int create_altmaps_and_memory_blocks(int nid, struct memory_group *group,
 		ret = create_memory_block_devices(cur_start, memblock_size, nid,
 						  params.altmap, group);
 		if (ret) {
-			arch_remove_memory(cur_start, memblock_size, NULL);
+			arch_remove_memory(cur_start, memblock_size, NULL, NULL);
 			kfree(params.altmap);
 			goto out;
 		}
@@ -1554,7 +1555,7 @@ int add_memory_resource(int nid, struct resource *res, mhp_t mhp_flags)
 		/* create memory block devices after memory was added */
 		ret = create_memory_block_devices(start, size, nid, NULL, group);
 		if (ret) {
-			arch_remove_memory(start, size, params.altmap);
+			arch_remove_memory(start, size, params.altmap, NULL);
 			goto error;
 		}
 	}
@@ -2266,7 +2267,7 @@ static int try_remove_memory(u64 start, u64 size)
 		 * No altmaps present, do the removal directly
 		 */
 		remove_memory_block_devices(start, size);
-		arch_remove_memory(start, size, NULL);
+		arch_remove_memory(start, size, NULL, NULL);
 	} else {
 		/* all memblocks in the range have altmaps */
 		remove_memory_blocks_and_altmaps(start, size);
diff --git a/mm/memremap.c b/mm/memremap.c
index 053842d45cb1..81766d822400 100644
--- a/mm/memremap.c
+++ b/mm/memremap.c
@@ -97,10 +97,10 @@ static void pageunmap_range(struct dev_pagemap *pgmap, int range_id)
 				   PHYS_PFN(range_len(range)));
 	if (pgmap->type == MEMORY_DEVICE_PRIVATE) {
 		__remove_pages(PHYS_PFN(range->start),
-			       PHYS_PFN(range_len(range)), NULL);
+			       PHYS_PFN(range_len(range)), NULL, pgmap);
 	} else {
 		arch_remove_memory(range->start, range_len(range),
-				pgmap_altmap(pgmap));
+				pgmap_altmap(pgmap), pgmap);
 		kasan_remove_zero_shadow(__va(range->start), range_len(range));
 	}
 	mem_hotplug_done();
diff --git a/mm/sparse-vmemmap.c b/mm/sparse-vmemmap.c
index a7b11248b989..c208187a4b00 100644
--- a/mm/sparse-vmemmap.c
+++ b/mm/sparse-vmemmap.c
@@ -665,7 +665,7 @@ static struct page * __meminit populate_section_memmap(unsigned long pfn,
 }
 
 static void depopulate_section_memmap(unsigned long pfn, unsigned long nr_pages,
-		struct vmem_altmap *altmap)
+		struct vmem_altmap *altmap, struct dev_pagemap *pgmap)
 {
 	unsigned long start = (unsigned long) pfn_to_page(pfn);
 	unsigned long end = start + nr_pages * sizeof(struct page);
@@ -746,7 +746,7 @@ static int fill_subsection_map(unsigned long pfn, unsigned long nr_pages)
  * usage map, but still need to free the vmemmap range.
  */
 static void section_deactivate(unsigned long pfn, unsigned long nr_pages,
-		struct vmem_altmap *altmap)
+		struct vmem_altmap *altmap, struct dev_pagemap *pgmap)
 {
 	struct mem_section *ms = __pfn_to_section(pfn);
 	bool section_is_early = early_section(ms);
@@ -784,7 +784,7 @@ static void section_deactivate(unsigned long pfn, unsigned long nr_pages,
 	 * section_activate() and pfn_valid() .
 	 */
 	if (!section_is_early)
-		depopulate_section_memmap(pfn, nr_pages, altmap);
+		depopulate_section_memmap(pfn, nr_pages, altmap, pgmap);
 	else if (memmap)
 		free_map_bootmem(memmap);
 
@@ -828,7 +828,7 @@ static struct page * __meminit section_activate(int nid, unsigned long pfn,
 
 	memmap = populate_section_memmap(pfn, nr_pages, nid, altmap, pgmap);
 	if (!memmap) {
-		section_deactivate(pfn, nr_pages, altmap);
+		section_deactivate(pfn, nr_pages, altmap, pgmap);
 		return ERR_PTR(-ENOMEM);
 	}
 
@@ -889,13 +889,13 @@ int __meminit sparse_add_section(int nid, unsigned long start_pfn,
 }
 
 void sparse_remove_section(unsigned long pfn, unsigned long nr_pages,
-			   struct vmem_altmap *altmap)
+			   struct vmem_altmap *altmap, struct dev_pagemap *pgmap)
 {
 	struct mem_section *ms = __pfn_to_section(pfn);
 
 	if (WARN_ON_ONCE(!valid_section(ms)))
 		return;
 
-	section_deactivate(pfn, nr_pages, altmap);
+	section_deactivate(pfn, nr_pages, altmap, pgmap);
 }
 #endif /* CONFIG_MEMORY_HOTPLUG */
-- 
2.20.1

[PATCH v4 3/5] mm/sparse-vmemmap: Fix DAX vmemmap accounting with optimization

[Muchun Song]

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_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")
Signed-off-by: Muchun Song <songmuchun@bytedance.com>
Acked-by: Mike Rapoport (Microsoft) <rppt@kernel.org>
Acked-by: Oscar Salvador <osalvador@suse.de>
---
 mm/sparse-vmemmap.c | 32 ++++++++++++++++++++++++++++----
 1 file changed, 28 insertions(+), 4 deletions(-)

diff --git a/mm/sparse-vmemmap.c b/mm/sparse-vmemmap.c
index c208187a4b00..fcc5e0eda9e7 100644
--- a/mm/sparse-vmemmap.c
+++ b/mm/sparse-vmemmap.c
@@ -652,6 +652,29 @@ void offline_mem_sections(unsigned long start_pfn, unsigned long end_pfn)
 	}
 }
 
+static int __meminit section_vmemmap_pages(unsigned long pfn, unsigned long nr_pages,
+					   struct vmem_altmap *altmap,
+					   struct dev_pagemap *pgmap)
+{
+	unsigned int order = pgmap ? pgmap->vmemmap_shift : 0;
+	unsigned long pages_per_compound = 1L << order;
+
+	VM_WARN_ON_ONCE(!IS_ALIGNED(pfn | nr_pages, min(pages_per_compound,
+							PAGES_PER_SECTION)));
+	VM_WARN_ON_ONCE(pfn_to_section_nr(pfn) != pfn_to_section_nr(pfn + nr_pages - 1));
+
+	if (!vmemmap_can_optimize(altmap, pgmap))
+		return DIV_ROUND_UP(nr_pages * sizeof(struct page), PAGE_SIZE);
+
+	if (order < PFN_SECTION_SHIFT)
+		return VMEMMAP_RESERVE_NR * nr_pages / pages_per_compound;
+
+	if (IS_ALIGNED(pfn, pages_per_compound))
+		return VMEMMAP_RESERVE_NR;
+
+	return 0;
+}
+
 static struct page * __meminit populate_section_memmap(unsigned long pfn,
 		unsigned long nr_pages, int nid, struct vmem_altmap *altmap,
 		struct dev_pagemap *pgmap)
@@ -659,7 +682,7 @@ static struct page * __meminit populate_section_memmap(unsigned long pfn,
 	struct page *page = __populate_section_memmap(pfn, nr_pages, nid, altmap,
 						      pgmap);
 
-	memmap_pages_add(DIV_ROUND_UP(nr_pages * sizeof(struct page), PAGE_SIZE));
+	memmap_pages_add(section_vmemmap_pages(pfn, nr_pages, altmap, pgmap));
 
 	return page;
 }
@@ -670,7 +693,7 @@ static void depopulate_section_memmap(unsigned long pfn, unsigned long nr_pages,
 	unsigned long start = (unsigned long) pfn_to_page(pfn);
 	unsigned long end = start + nr_pages * sizeof(struct page);
 
-	memmap_pages_add(-1L * (DIV_ROUND_UP(nr_pages * sizeof(struct page), PAGE_SIZE)));
+	memmap_pages_add(-section_vmemmap_pages(pfn, nr_pages, altmap, pgmap));
 	vmemmap_free(start, end, altmap);
 }
 
@@ -678,9 +701,10 @@ static void free_map_bootmem(struct page *memmap)
 {
 	unsigned long start = (unsigned long)memmap;
 	unsigned long end = (unsigned long)(memmap + PAGES_PER_SECTION);
+	unsigned long pfn = page_to_pfn(memmap);
 
-	memmap_boot_pages_add(-1L * (DIV_ROUND_UP(PAGES_PER_SECTION * sizeof(struct page),
-						  PAGE_SIZE)));
+	memmap_boot_pages_add(-section_vmemmap_pages(pfn, PAGES_PER_SECTION,
+						     NULL, NULL));
 	vmemmap_free(start, end, NULL);
 }
 
-- 
2.20.1

[PATCH v4 4/5] mm/mm_init: Fix pageblock migratetype for ZONE_DEVICE compound pages

[Muchun Song]

The memmap_init_zone_device() function only initializes the migratetype
of the first pageblock of a compound page. If the compound page size
exceeds pageblock_nr_pages (e.g., 1GB hugepages with 2MB pageblocks),
subsequent pageblocks in the compound page remain uninitialized.

Move the migratetype initialization out of __init_zone_device_page()
and into a separate pageblock_migratetype_init_range() function. This
iterates over the entire PFN range of the memory, ensuring that all
pageblocks are correctly initialized.

Fixes: c4386bd8ee3a ("mm/memremap: add ZONE_DEVICE support for compound pages")
Signed-off-by: Muchun Song <songmuchun@bytedance.com>
Reviewed-by: Mike Rapoport (Microsoft) <rppt@kernel.org>
Reviewed-by: Oscar Salvador <osalvador@suse.de>
---
 mm/mm_init.c | 45 ++++++++++++++++++++++++++++++---------------
 1 file changed, 30 insertions(+), 15 deletions(-)

diff --git a/mm/mm_init.c b/mm/mm_init.c
index f9f8e1af921c..9d0fe79a94de 100644
--- a/mm/mm_init.c
+++ b/mm/mm_init.c
@@ -674,6 +674,21 @@ static inline void fixup_hashdist(void)
 static inline void fixup_hashdist(void) {}
 #endif /* CONFIG_NUMA */
 
+#ifdef CONFIG_ZONE_DEVICE
+static __meminit void pageblock_migratetype_init_range(unsigned long pfn,
+						       unsigned long nr_pages,
+						       int migratetype)
+{
+	unsigned long end = pfn + nr_pages;
+
+	for (pfn = pageblock_align(pfn); pfn < end; pfn += pageblock_nr_pages) {
+		init_pageblock_migratetype(pfn_to_page(pfn), migratetype, false);
+		if (IS_ALIGNED(pfn, PAGES_PER_SECTION))
+			cond_resched();
+	}
+}
+#endif
+
 /*
  * Initialize a reserved page unconditionally, finding its zone first.
  */
@@ -1011,21 +1026,6 @@ static void __ref __init_zone_device_page(struct page *page, unsigned long pfn,
 	page_folio(page)->pgmap = pgmap;
 	page->zone_device_data = NULL;
 
-	/*
-	 * Mark the block movable so that blocks are reserved for
-	 * movable at startup. This will force kernel allocations
-	 * to reserve their blocks rather than leaking throughout
-	 * the address space during boot when many long-lived
-	 * kernel allocations are made.
-	 *
-	 * Please note that MEMINIT_HOTPLUG path doesn't clear memmap
-	 * because this is done early in section_activate()
-	 */
-	if (pageblock_aligned(pfn)) {
-		init_pageblock_migratetype(page, MIGRATE_MOVABLE, false);
-		cond_resched();
-	}
-
 	/*
 	 * ZONE_DEVICE pages other than MEMORY_TYPE_GENERIC are released
 	 * directly to the driver page allocator which will set the page count
@@ -1122,6 +1122,9 @@ void __ref memmap_init_zone_device(struct zone *zone,
 
 		__init_zone_device_page(page, pfn, zone_idx, nid, pgmap);
 
+		if (IS_ALIGNED(pfn, PAGES_PER_SECTION))
+			cond_resched();
+
 		if (pfns_per_compound == 1)
 			continue;
 
@@ -1129,6 +1132,18 @@ void __ref memmap_init_zone_device(struct zone *zone,
 				     compound_nr_pages(altmap, pgmap));
 	}
 
+	/*
+	 * Mark the block movable so that blocks are reserved for
+	 * movable at startup. This will force kernel allocations
+	 * to reserve their blocks rather than leaking throughout
+	 * the address space during boot when many long-lived
+	 * kernel allocations are made.
+	 *
+	 * Please note that MEMINIT_HOTPLUG path doesn't clear memmap
+	 * because this is done early in section_activate()
+	 */
+	pageblock_migratetype_init_range(start_pfn, nr_pages, MIGRATE_MOVABLE);
+
 	pr_debug("%s initialised %lu pages in %ums\n", __func__,
 		nr_pages, jiffies_to_msecs(jiffies - start));
 }
-- 
2.20.1

[PATCH v4 5/5] mm/mm_init: Fix uninitialized struct pages for ZONE_DEVICE

[Muchun Song]

If DAX memory is hotplugged into an unoccupied subsection of an early
section, section_activate() reuses the unoptimized boot memmap.
However, compound_nr_pages() still assumes that vmemmap optimization is
in effect and initializes only the reduced number of struct pages. As a
result, the remaining tail struct pages are left uninitialized, which
can later lead to unexpected behavior or crashes.

Fix this by treating early sections as unoptimized when calculating how
many struct pages to initialize.

Fixes: 6fd3620b3428 ("mm/page_alloc: reuse tail struct pages for compound devmaps")
Signed-off-by: Muchun Song <songmuchun@bytedance.com>
---
 mm/mm_init.c | 13 ++++++++++---
 1 file changed, 10 insertions(+), 3 deletions(-)

diff --git a/mm/mm_init.c b/mm/mm_init.c
index 9d0fe79a94de..3d5af40d0943 100644
--- a/mm/mm_init.c
+++ b/mm/mm_init.c
@@ -1056,10 +1056,17 @@ static void __ref __init_zone_device_page(struct page *page, unsigned long pfn,
  * of how the sparse_vmemmap internals handle compound pages in the lack
  * of an altmap. See vmemmap_populate_compound_pages().
  */
-static inline unsigned long compound_nr_pages(struct vmem_altmap *altmap,
+static inline unsigned long compound_nr_pages(unsigned long pfn,
+					      struct vmem_altmap *altmap,
 					      struct dev_pagemap *pgmap)
 {
-	if (!vmemmap_can_optimize(altmap, pgmap))
+	/*
+	 * If DAX memory is hot-plugged into an unoccupied subsection
+	 * of an early section, the unoptimized boot memmap is reused.
+	 * See section_activate().
+	 */
+	if (early_section(__pfn_to_section(pfn)) ||
+	    !vmemmap_can_optimize(altmap, pgmap))
 		return pgmap_vmemmap_nr(pgmap);
 
 	return VMEMMAP_RESERVE_NR * (PAGE_SIZE / sizeof(struct page));
@@ -1129,7 +1136,7 @@ void __ref memmap_init_zone_device(struct zone *zone,
 			continue;
 
 		memmap_init_compound(page, pfn, zone_idx, nid, pgmap,
-				     compound_nr_pages(altmap, pgmap));
+				     compound_nr_pages(pfn, altmap, pgmap));
 	}
 
 	/*
-- 
2.20.1

Re: [PATCH v4 1/5] mm/sparse-vmemmap: Fix vmemmap accounting underflow

[David Hildenbrand (Arm)]

On 4/22/26 10:14, Muchun Song wrote:
> In section_activate(), if populate_section_memmap() fails, the error
> handling path calls section_deactivate() to roll back the state. This
> causes a vmemmap accounting imbalance.
> 
> Since commit c3576889d87b ("mm: fix accounting of memmap pages"),
> memmap pages are accounted for only after populate_section_memmap()
> succeeds. However, the failure path unconditionally calls
> section_deactivate(), which decreases the vmemmap count. Consequently,
> a failure in populate_section_memmap() leads to an accounting underflow,
> incorrectly reducing the system's tracked vmemmap usage.
> 
> Fix this more thoroughly by moving all accounting calls into the lower
> level functions that actually perform the vmemmap allocation and freeing:
> 
>   - populate_section_memmap() accounts for newly allocated vmemmap pages
>   - depopulate_section_memmap() unaccounts when vmemmap is freed
> 
> This ensures proper accounting in all code paths, including error
> handling and early section cases.
> 
> Fixes: c3576889d87b ("mm: fix accounting of memmap pages")
> Signed-off-by: Muchun Song <songmuchun@bytedance.com>
> Acked-by: Mike Rapoport (Microsoft) <rppt@kernel.org>
> Acked-by: Oscar Salvador <osalvador@suse.de>
> ---

Much clearer, thanks!

Acked-by: David Hildenbrand (Arm) <david@kernel.org>

-- 
Cheers,

David

Re: [PATCH v4 2/5] mm/sparse-vmemmap: Pass @pgmap argument to memory deactivation paths

[David Hildenbrand (Arm)]

On 4/22/26 10:14, Muchun Song wrote:
> Currently, the memory hot-remove call chain -- arch_remove_memory(),
> __remove_pages(), sparse_remove_section() and section_deactivate() --
> does not carry the struct dev_pagemap pointer. This prevents the lower
> levels from knowing whether the section was originally populated with
> vmemmap optimizations (e.g., DAX with vmemmap optimization enabled).
> 
> Without this information, we cannot call vmemmap_can_optimize() to
> determine if the vmemmap pages were optimized. As a result, the vmemmap
> page accounting during teardown will mistakenly assume a non-optimized
> allocation, leading to incorrect memmap statistics.
> 
> To lay the groundwork for fixing the vmemmap page accounting, we need
> to pass the @pgmap pointer down to the deactivation location. Plumb the
> @pgmap argument through the APIs of arch_remove_memory(), __remove_pages()
> and sparse_remove_section(), mirroring the corresponding *_activate()
> paths.
> 
> Signed-off-by: Muchun Song <songmuchun@bytedance.com>
> Acked-by: Mike Rapoport (Microsoft) <rppt@kernel.org>
> Reviewed-by: Oscar Salvador <osalvador@suse.de>


[...]

>  static void depopulate_section_memmap(unsigned long pfn, unsigned long nr_pages,
> -		struct vmem_altmap *altmap)
> +		struct vmem_altmap *altmap, struct dev_pagemap *pgmap)
>  {
>  	unsigned long start = (unsigned long) pfn_to_page(pfn);
>  	unsigned long end = start + nr_pages * sizeof(struct page);
> @@ -746,7 +746,7 @@ static int fill_subsection_map(unsigned long pfn, unsigned long nr_pages)
>   * usage map, but still need to free the vmemmap range.
>   */
>  static void section_deactivate(unsigned long pfn, unsigned long nr_pages,
> -		struct vmem_altmap *altmap)
> +		struct vmem_altmap *altmap, struct dev_pagemap *pgmap)
>  {
>  	struct mem_section *ms = __pfn_to_section(pfn);
>  	bool section_is_early = early_section(ms);
> @@ -784,7 +784,7 @@ static void section_deactivate(unsigned long pfn, unsigned long nr_pages,
>  	 * section_activate() and pfn_valid() .
>  	 */
>  	if (!section_is_early)
> -		depopulate_section_memmap(pfn, nr_pages, altmap);
> +		depopulate_section_memmap(pfn, nr_pages, altmap, pgmap);
>  	else if (memmap)
>  		free_map_bootmem(memmap);
>  
> @@ -828,7 +828,7 @@ static struct page * __meminit section_activate(int nid, unsigned long pfn,
>  
>  	memmap = populate_section_memmap(pfn, nr_pages, nid, altmap, pgmap);
>  	if (!memmap) {
> -		section_deactivate(pfn, nr_pages, altmap);
> +		section_deactivate(pfn, nr_pages, altmap, pgmap);
>  		return ERR_PTR(-ENOMEM);
>  	}
>  
> @@ -889,13 +889,13 @@ int __meminit sparse_add_section(int nid, unsigned long start_pfn,
>  }
>  
>  void sparse_remove_section(unsigned long pfn, unsigned long nr_pages,
> -			   struct vmem_altmap *altmap)
> +			   struct vmem_altmap *altmap, struct dev_pagemap *pgmap)

While at it, could switch to two-tab indent here as well.

Acked-by: David Hildenbrand (Arm) <david@kernel.org>

-- 
Cheers,

David

Re: [PATCH v4 3/5] mm/sparse-vmemmap: Fix DAX vmemmap accounting with optimization

[David Hildenbrand (Arm)]

On 4/22/26 10:14, 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_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")
> Signed-off-by: Muchun Song <songmuchun@bytedance.com>
> Acked-by: Mike Rapoport (Microsoft) <rppt@kernel.org>
> Acked-by: Oscar Salvador <osalvador@suse.de>
> ---
>  mm/sparse-vmemmap.c | 32 ++++++++++++++++++++++++++++----
>  1 file changed, 28 insertions(+), 4 deletions(-)
> 
> diff --git a/mm/sparse-vmemmap.c b/mm/sparse-vmemmap.c
> index c208187a4b00..fcc5e0eda9e7 100644
> --- a/mm/sparse-vmemmap.c
> +++ b/mm/sparse-vmemmap.c
> @@ -652,6 +652,29 @@ void offline_mem_sections(unsigned long start_pfn, unsigned long end_pfn)
>  	}
>  }
>  
> +static int __meminit section_vmemmap_pages(unsigned long pfn, unsigned long nr_pages,

I'd have called this "section_nr_vmemmap_pages"

> +					   struct vmem_altmap *altmap,
> +					   struct dev_pagemap *pgmap)

Two-tab indent.

> +{
> +	unsigned int order = pgmap ? pgmap->vmemmap_shift : 0;
> +	unsigned long pages_per_compound = 1L << order;

1UL

Both can be const.

> +
> +	VM_WARN_ON_ONCE(!IS_ALIGNED(pfn | nr_pages, min(pages_per_compound,
> +							PAGES_PER_SECTION)));

Maybe simply

	VM_WARN_ON_ONCE(!IS_ALIGNED(pfn | nr_pages, pages_per_compound));
	VM_WARN_ON_ONCE(!IS_ALIGNED(pfn | nr_pages, PAGES_PER_SECTION));

Which is more readable?


> +	VM_WARN_ON_ONCE(pfn_to_section_nr(pfn) != pfn_to_section_nr(pfn + nr_pages - 1));
> +
> +	if (!vmemmap_can_optimize(altmap, pgmap))
> +		return DIV_ROUND_UP(nr_pages * sizeof(struct page), PAGE_SIZE);
> +
> +	if (order < PFN_SECTION_SHIFT)
> +		return VMEMMAP_RESERVE_NR * nr_pages / pages_per_compound;
> +
> +	if (IS_ALIGNED(pfn, pages_per_compound))
> +		return VMEMMAP_RESERVE_NR;
> +

I'll have to trust you on these ones :)

> +	return 0;
> +}

-- 
Cheers,

David

Re: [PATCH v4 4/5] mm/mm_init: Fix pageblock migratetype for ZONE_DEVICE compound pages

[David Hildenbrand (Arm)]

On 4/22/26 10:14, Muchun Song wrote:
> The memmap_init_zone_device() function only initializes the migratetype
> of the first pageblock of a compound page. If the compound page size
> exceeds pageblock_nr_pages (e.g., 1GB hugepages with 2MB pageblocks),
> subsequent pageblocks in the compound page remain uninitialized.
> 
> Move the migratetype initialization out of __init_zone_device_page()
> and into a separate pageblock_migratetype_init_range() function. This
> iterates over the entire PFN range of the memory, ensuring that all
> pageblocks are correctly initialized.
> 
> Fixes: c4386bd8ee3a ("mm/memremap: add ZONE_DEVICE support for compound pages")
> Signed-off-by: Muchun Song <songmuchun@bytedance.com>
> Reviewed-by: Mike Rapoport (Microsoft) <rppt@kernel.org>
> Reviewed-by: Oscar Salvador <osalvador@suse.de>
> ---
>  mm/mm_init.c | 45 ++++++++++++++++++++++++++++++---------------
>  1 file changed, 30 insertions(+), 15 deletions(-)
> 
> diff --git a/mm/mm_init.c b/mm/mm_init.c
> index f9f8e1af921c..9d0fe79a94de 100644
> --- a/mm/mm_init.c
> +++ b/mm/mm_init.c
> @@ -674,6 +674,21 @@ static inline void fixup_hashdist(void)
>  static inline void fixup_hashdist(void) {}
>  #endif /* CONFIG_NUMA */
>  
> +#ifdef CONFIG_ZONE_DEVICE
> +static __meminit void pageblock_migratetype_init_range(unsigned long pfn,
> +						       unsigned long nr_pages,
> +						       int migratetype)

Two-tab indent says hi :)

> +{
> +	unsigned long end = pfn + nr_pages;

Can be const.

> +
> +	for (pfn = pageblock_align(pfn); pfn < end; pfn += pageblock_nr_pages) {
> +		init_pageblock_migratetype(pfn_to_page(pfn), migratetype, false);
> +		if (IS_ALIGNED(pfn, PAGES_PER_SECTION))
> +			cond_resched();
> +	}
> +}
> +#endif
> +
>  /*
>   * Initialize a reserved page unconditionally, finding its zone first.
>   */
> @@ -1011,21 +1026,6 @@ static void __ref __init_zone_device_page(struct page *page, unsigned long pfn,
>  	page_folio(page)->pgmap = pgmap;
>  	page->zone_device_data = NULL;
>  
> -	/*
> -	 * Mark the block movable so that blocks are reserved for
> -	 * movable at startup. This will force kernel allocations
> -	 * to reserve their blocks rather than leaking throughout
> -	 * the address space during boot when many long-lived
> -	 * kernel allocations are made.
> -	 *
> -	 * Please note that MEMINIT_HOTPLUG path doesn't clear memmap
> -	 * because this is done early in section_activate()
> -	 */
> -	if (pageblock_aligned(pfn)) {
> -		init_pageblock_migratetype(page, MIGRATE_MOVABLE, false);
> -		cond_resched();
> -	}
> -
>  	/*
>  	 * ZONE_DEVICE pages other than MEMORY_TYPE_GENERIC are released
>  	 * directly to the driver page allocator which will set the page count
> @@ -1122,6 +1122,9 @@ void __ref memmap_init_zone_device(struct zone *zone,
>  
>  		__init_zone_device_page(page, pfn, zone_idx, nid, pgmap);
>  
> +		if (IS_ALIGNED(pfn, PAGES_PER_SECTION))
> +			cond_resched();
> +
>  		if (pfns_per_compound == 1)
>  			continue;
>  
> @@ -1129,6 +1132,18 @@ void __ref memmap_init_zone_device(struct zone *zone,
>  				     compound_nr_pages(altmap, pgmap));
>  	}
>  
> +	/*
> +	 * Mark the block movable so that blocks are reserved for
> +	 * movable at startup. This will force kernel allocations
> +	 * to reserve their blocks rather than leaking throughout
> +	 * the address space during boot when many long-lived
> +	 * kernel allocations are made.
> +	 *
> +	 * Please note that MEMINIT_HOTPLUG path doesn't clear memmap
> +	 * because this is done early in section_activate()
> +	 */

I am deeply confused about the "memmap" comment here.

There is no MEMINIT_HOTPLUG special-casing. Nobody "clears memmap".


That comment was introduced in

commit 966cf44f637e6aeea7e3d01ba004bf8b5beac78f
Author: Alexander Duyck <alexander.h.duyck@linux.intel.com>
Date:   Fri Oct 26 15:07:52 2018 -0700

    mm: defer ZONE_DEVICE page initialization to the point where we init pgmap

Where we added memmap_init_zone_device().



Is that commit still valid, and this it actually belong here, above the
migratetype setting?


> +	pageblock_migratetype_init_range(start_pfn, nr_pages, MIGRATE_MOVABLE);
> +
>  	pr_debug("%s initialised %lu pages in %ums\n", __func__,
>  		nr_pages, jiffies_to_msecs(jiffies - start));
>  }


-- 
Cheers,

David

Re: [PATCH v4 5/5] mm/mm_init: Fix uninitialized struct pages for ZONE_DEVICE

[David Hildenbrand (Arm)]

On 4/22/26 10:14, Muchun Song wrote:
> If DAX memory is hotplugged into an unoccupied subsection of an early
> section, section_activate() reuses the unoptimized boot memmap.
> However, compound_nr_pages() still assumes that vmemmap optimization is
> in effect and initializes only the reduced number of struct pages. As a
> result, the remaining tail struct pages are left uninitialized, which
> can later lead to unexpected behavior or crashes.
> 
> Fix this by treating early sections as unoptimized when calculating how
> many struct pages to initialize.
> 
> Fixes: 6fd3620b3428 ("mm/page_alloc: reuse tail struct pages for compound devmaps")
> Signed-off-by: Muchun Song <songmuchun@bytedance.com>
> ---
>  mm/mm_init.c | 13 ++++++++++---
>  1 file changed, 10 insertions(+), 3 deletions(-)
> 
> diff --git a/mm/mm_init.c b/mm/mm_init.c
> index 9d0fe79a94de..3d5af40d0943 100644
> --- a/mm/mm_init.c
> +++ b/mm/mm_init.c
> @@ -1056,10 +1056,17 @@ static void __ref __init_zone_device_page(struct page *page, unsigned long pfn,
>   * of how the sparse_vmemmap internals handle compound pages in the lack
>   * of an altmap. See vmemmap_populate_compound_pages().
>   */
> -static inline unsigned long compound_nr_pages(struct vmem_altmap *altmap,
> +static inline unsigned long compound_nr_pages(unsigned long pfn,
> +					      struct vmem_altmap *altmap,
>  					      struct dev_pagemap *pgmap)
>  {
> -	if (!vmemmap_can_optimize(altmap, pgmap))
> +	/*
> +	 * If DAX memory is hot-plugged into an unoccupied subsection
> +	 * of an early section, the unoptimized boot memmap is reused.
> +	 * See section_activate().
> +	 */
> +	if (early_section(__pfn_to_section(pfn)) ||
> +	    !vmemmap_can_optimize(altmap, pgmap))
>  		return pgmap_vmemmap_nr(pgmap);
>  
>  	return VMEMMAP_RESERVE_NR * (PAGE_SIZE / sizeof(struct page));
> @@ -1129,7 +1136,7 @@ void __ref memmap_init_zone_device(struct zone *zone,
>  			continue;
>  
>  		memmap_init_compound(page, pfn, zone_idx, nid, pgmap,
> -				     compound_nr_pages(altmap, pgmap));
> +				     compound_nr_pages(pfn, altmap, pgmap));
>  	}
>  
>  	/*

Nasty, but yes, we cannot really optimize in that case.

Acked-by: David Hildenbrand (Arm) <david@kernel.org>

-- 
Cheers,

David

Re: [PATCH v4 2/5] mm/sparse-vmemmap: Pass @pgmap argument to memory deactivation paths

[Muchun Song]

> On Apr 23, 2026, at 02:50, David Hildenbrand (Arm) <david@kernel.org> wrote:
> 
> On 4/22/26 10:14, Muchun Song wrote:
>> Currently, the memory hot-remove call chain -- arch_remove_memory(),
>> __remove_pages(), sparse_remove_section() and section_deactivate() --
>> does not carry the struct dev_pagemap pointer. This prevents the lower
>> levels from knowing whether the section was originally populated with
>> vmemmap optimizations (e.g., DAX with vmemmap optimization enabled).
>> 
>> Without this information, we cannot call vmemmap_can_optimize() to
>> determine if the vmemmap pages were optimized. As a result, the vmemmap
>> page accounting during teardown will mistakenly assume a non-optimized
>> allocation, leading to incorrect memmap statistics.
>> 
>> To lay the groundwork for fixing the vmemmap page accounting, we need
>> to pass the @pgmap pointer down to the deactivation location. Plumb the
>> @pgmap argument through the APIs of arch_remove_memory(), __remove_pages()
>> and sparse_remove_section(), mirroring the corresponding *_activate()
>> paths.
>> 
>> Signed-off-by: Muchun Song <songmuchun@bytedance.com>
>> Acked-by: Mike Rapoport (Microsoft) <rppt@kernel.org>
>> Reviewed-by: Oscar Salvador <osalvador@suse.de>
> 
> 
> [...]
> 
>> static void depopulate_section_memmap(unsigned long pfn, unsigned long nr_pages,
>> - 		struct vmem_altmap *altmap)
>> + 		struct vmem_altmap *altmap, struct dev_pagemap *pgmap)
>> {
>> 	unsigned long start = (unsigned long) pfn_to_page(pfn);
>> 	unsigned long end = start + nr_pages * sizeof(struct page);
>> @@ -746,7 +746,7 @@ static int fill_subsection_map(unsigned long pfn, unsigned long nr_pages)
>>  * usage map, but still need to free the vmemmap range.
>>  */
>> static void section_deactivate(unsigned long pfn, unsigned long nr_pages,
>> - 		struct vmem_altmap *altmap)
>> + 		struct vmem_altmap *altmap, struct dev_pagemap *pgmap)
>> {
>> 	struct mem_section *ms = __pfn_to_section(pfn);
>> 	bool section_is_early = early_section(ms);
>> @@ -784,7 +784,7 @@ static void section_deactivate(unsigned long pfn, unsigned long nr_pages,
>>  * section_activate() and pfn_valid() .
>>  */
>> 	if (!section_is_early)
>> - 		depopulate_section_memmap(pfn, nr_pages, altmap);
>> + 		depopulate_section_memmap(pfn, nr_pages, altmap, pgmap);
>> 	else if (memmap)
>> 		free_map_bootmem(memmap);
>> 
>> @@ -828,7 +828,7 @@ static struct page * __meminit section_activate(int nid, unsigned long pfn,
>> 
>> 	memmap = populate_section_memmap(pfn, nr_pages, nid, altmap, pgmap);
>> 	if (!memmap) {
>> - 		section_deactivate(pfn, nr_pages, altmap);
>> + 		section_deactivate(pfn, nr_pages, altmap, pgmap);
>> 		return ERR_PTR(-ENOMEM);
>> 	}
>> 
>> @@ -889,13 +889,13 @@ int __meminit sparse_add_section(int nid, unsigned long start_pfn,
>> }
>> 
>> void sparse_remove_section(unsigned long pfn, unsigned long nr_pages,
>> -    			   struct vmem_altmap *altmap)
>> +    			   struct vmem_altmap *altmap, struct dev_pagemap *pgmap)
> 
> While at it, could switch to two-tab indent here as well.

OK.

> 
> Acked-by: David Hildenbrand (Arm) <david@kernel.org>

Thanks.

Muchun.

> 
> -- 
> Cheers,
> 
> David

Re: [PATCH v4 3/5] mm/sparse-vmemmap: Fix DAX vmemmap accounting with optimization

[Muchun Song]

> On Apr 23, 2026, at 02:53, David Hildenbrand (Arm) <david@kernel.org> wrote:
> 
> On 4/22/26 10:14, 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_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")
>> Signed-off-by: Muchun Song <songmuchun@bytedance.com>
>> Acked-by: Mike Rapoport (Microsoft) <rppt@kernel.org>
>> Acked-by: Oscar Salvador <osalvador@suse.de>
>> ---
>> mm/sparse-vmemmap.c | 32 ++++++++++++++++++++++++++++----
>> 1 file changed, 28 insertions(+), 4 deletions(-)
>> 
>> diff --git a/mm/sparse-vmemmap.c b/mm/sparse-vmemmap.c
>> index c208187a4b00..fcc5e0eda9e7 100644
>> --- a/mm/sparse-vmemmap.c
>> +++ b/mm/sparse-vmemmap.c
>> @@ -652,6 +652,29 @@ void offline_mem_sections(unsigned long start_pfn, unsigned long end_pfn)
>> }
>> }
>> 
>> +static int __meminit section_vmemmap_pages(unsigned long pfn, unsigned long nr_pages,
> 
> I'd have called this "section_nr_vmemmap_pages"

No problem.

> 
>> +    struct vmem_altmap *altmap,
>> +    struct dev_pagemap *pgmap)
> 
> Two-tab indent.

OK.

> 
>> +{
>> + unsigned int order = pgmap ? pgmap->vmemmap_shift : 0;
>> + unsigned long pages_per_compound = 1L << order;
> 
> 1UL
> 
> Both can be const.

Right.

> 
>> +
>> + VM_WARN_ON_ONCE(!IS_ALIGNED(pfn | nr_pages, min(pages_per_compound,
>> + PAGES_PER_SECTION)));
> 
> Maybe simply
> 
> VM_WARN_ON_ONCE(!IS_ALIGNED(pfn | nr_pages, pages_per_compound));
> VM_WARN_ON_ONCE(!IS_ALIGNED(pfn | nr_pages, PAGES_PER_SECTION));
> 
> Which is more readable?

That's also quite good.

> 
> 
>> + 	VM_WARN_ON_ONCE(pfn_to_section_nr(pfn) != pfn_to_section_nr(pfn + nr_pages - 1));
>> +
>> + 	if (!vmemmap_can_optimize(altmap, pgmap))
>> + 		return DIV_ROUND_UP(nr_pages * sizeof(struct page), PAGE_SIZE);
>> +
>> + 	if (order < PFN_SECTION_SHIFT)
>> + 		return VMEMMAP_RESERVE_NR * nr_pages / pages_per_compound;
>> +
>> + 	if (IS_ALIGNED(pfn, pages_per_compound))
>> + 		return VMEMMAP_RESERVE_NR;
>> +
> 
> I'll have to trust you on these ones :)

Thanks for your trust.

Thanks,
Muchun.

> 
>> + return 0;
>> +}
> 
> -- 
> Cheers,
> 
> David

Re: [PATCH v4 4/5] mm/mm_init: Fix pageblock migratetype for ZONE_DEVICE compound pages

[Muchun Song]

> On Apr 23, 2026, at 03:03, David Hildenbrand (Arm) <david@kernel.org> wrote:
> 
> On 4/22/26 10:14, Muchun Song wrote:
>> The memmap_init_zone_device() function only initializes the migratetype
>> of the first pageblock of a compound page. If the compound page size
>> exceeds pageblock_nr_pages (e.g., 1GB hugepages with 2MB pageblocks),
>> subsequent pageblocks in the compound page remain uninitialized.
>> 
>> Move the migratetype initialization out of __init_zone_device_page()
>> and into a separate pageblock_migratetype_init_range() function. This
>> iterates over the entire PFN range of the memory, ensuring that all
>> pageblocks are correctly initialized.
>> 
>> Fixes: c4386bd8ee3a ("mm/memremap: add ZONE_DEVICE support for compound pages")
>> Signed-off-by: Muchun Song <songmuchun@bytedance.com>
>> Reviewed-by: Mike Rapoport (Microsoft) <rppt@kernel.org>
>> Reviewed-by: Oscar Salvador <osalvador@suse.de>
>> ---
>> mm/mm_init.c | 45 ++++++++++++++++++++++++++++++---------------
>> 1 file changed, 30 insertions(+), 15 deletions(-)
>> 
>> diff --git a/mm/mm_init.c b/mm/mm_init.c
>> index f9f8e1af921c..9d0fe79a94de 100644
>> --- a/mm/mm_init.c
>> +++ b/mm/mm_init.c
>> @@ -674,6 +674,21 @@ static inline void fixup_hashdist(void)
>> static inline void fixup_hashdist(void) {}
>> #endif /* CONFIG_NUMA */
>> 
>> +#ifdef CONFIG_ZONE_DEVICE
>> +static __meminit void pageblock_migratetype_init_range(unsigned long pfn,
>> +        						unsigned long nr_pages,
>> +        						int migratetype)
> 
> Two-tab indent says hi :)

No problem.

> 
>> +{
>> + 	unsigned long end = pfn + nr_pages;
> 
> Can be const.

Ah, yes.

> 
>> +
>> + 	for (pfn = pageblock_align(pfn); pfn < end; pfn += pageblock_nr_pages) {
>> + 		init_pageblock_migratetype(pfn_to_page(pfn), migratetype, false);
>> + 		if (IS_ALIGNED(pfn, PAGES_PER_SECTION))
>> + 			cond_resched();
>> + 	}
>> +}
>> +#endif
>> +
>> /*
>>  * Initialize a reserved page unconditionally, finding its zone first.
>>  */
>> @@ -1011,21 +1026,6 @@ static void __ref __init_zone_device_page(struct page *page, unsigned long pfn,
>> 	page_folio(page)->pgmap = pgmap;
>> 	page->zone_device_data = NULL;
>> 
>> - 	/*
>> - 	 * Mark the block movable so that blocks are reserved for
>> - 	 * movable at startup. This will force kernel allocations
>> - 	 * to reserve their blocks rather than leaking throughout
>> - 	 * the address space during boot when many long-lived
>> - 	 * kernel allocations are made.
>> - 	 *
>> - 	 * Please note that MEMINIT_HOTPLUG path doesn't clear memmap
>> - 	 * because this is done early in section_activate()
>> - 	 */
>> - 	if (pageblock_aligned(pfn)) {
>> - 		init_pageblock_migratetype(page, MIGRATE_MOVABLE, false);
>> - 		cond_resched();
>> - 	}
>> -
>> /*
>>  * ZONE_DEVICE pages other than MEMORY_TYPE_GENERIC are released
>>  * directly to the driver page allocator which will set the page count
>> @@ -1122,6 +1122,9 @@ void __ref memmap_init_zone_device(struct zone *zone,
>> 
>> 	__init_zone_device_page(page, pfn, zone_idx, nid, pgmap);
>> 
>> + 	if (IS_ALIGNED(pfn, PAGES_PER_SECTION))
>> + 		cond_resched();
>> +
>> 	if (pfns_per_compound == 1)
>> 		continue;
>> 
>> @@ -1129,6 +1132,18 @@ void __ref memmap_init_zone_device(struct zone *zone,
>>      compound_nr_pages(altmap, pgmap));
>> }
>> 
>> + /*
>> +  * Mark the block movable so that blocks are reserved for
>> +  * movable at startup. This will force kernel allocations
>> +  * to reserve their blocks rather than leaking throughout
>> +  * the address space during boot when many long-lived
>> +  * kernel allocations are made.
>> +  *
>> +  * Please note that MEMINIT_HOTPLUG path doesn't clear memmap
>> +  * because this is done early in section_activate()
>> +  */
> 
> I am deeply confused about the "memmap" comment here.
> 
> There is no MEMINIT_HOTPLUG special-casing. Nobody "clears memmap".

Absolutely right. This comment was originally copied from the regular
memory initialization path long ago (commit 966cf44f637e), and it has
become completely outdated and misleading over the years after several
MM hotplug refactorings.

> 
> 
> That comment was introduced in
> 
> commit 966cf44f637e6aeea7e3d01ba004bf8b5beac78f
> Author: Alexander Duyck <alexander.h.duyck@linux.intel.com>
> Date:   Fri Oct 26 15:07:52 2018 -0700
> 
>    mm: defer ZONE_DEVICE page initialization to the point where we init pgmap
> 
> Where we added memmap_init_zone_device().
> 
> 
> 
> Is that commit still valid, and this it actually belong here, above the
> migratetype setting?

To be honest, I just blindly moved this comment along with the migratetype
initialization code out of __init_zone_device_page() without deeply analyzing
its historical context and current validity.

After looking into it, I think this comment is an obsolete relic from commit
966cf44f637e and no longer makes sense here.

I will completely remove this confusing comment in v5. Thanks for catching this!

Thanks,
Muchun

> 
> 
>> + 	pageblock_migratetype_init_range(start_pfn, nr_pages, MIGRATE_MOVABLE);
>> +
>> 	pr_debug("%s initialised %lu pages in %ums\n", __func__,
>> 		nr_pages, jiffies_to_msecs(jiffies - start));
>> }
> 
> 
> -- 
> Cheers,
> 
> David