[PATCH mm-new 0/5] mm: mincore: reduce lock contention and misc cleanups

[PATCH mm-new 0/5] mm: mincore: reduce lock contention and misc cleanups

[Kefeng Wang]

Kefeng Wang (5):
  mm: mincore: attempt per-vma lock during page table walk
  mm: mincore: remove special handing for VM_PFNMAP
  mm: mincore: replace __get_free_page() with kmalloc()
  mm: mincore: remove xa_is_value() in mincore_swap()
  mm: mincore: improve mincore_hugetlb()

 mm/mincore.c | 96 +++++++++++++++++++++++++---------------------------
 1 file changed, 47 insertions(+), 49 deletions(-)

-- 
2.27.0

[PATCH mm-new 1/5] mm: mincore: attempt per-vma lock during page table walk

[Kefeng Wang]

The do_mincore() is a read-only query of page residency from a single
VMA, making it an ideal candidate for per-vma locking. Try per-vma lock
first and fallback to mmap_lock, reducing global contention with page
faults and other mmap_lock holders in multi-threaded applications.                                        

Signed-off-by: Kefeng Wang <wangkefeng.wang@huawei.com>
---
 mm/mincore.c | 53 +++++++++++++++++++++++++++++++++++-----------------
 1 file changed, 36 insertions(+), 17 deletions(-)

diff --git a/mm/mincore.c b/mm/mincore.c
index 866b26cec540..f561d6e01052 100644
--- a/mm/mincore.c
+++ b/mm/mincore.c
@@ -12,6 +12,7 @@
 #include <linux/gfp.h>
 #include <linux/pagewalk.h>
 #include <linux/mman.h>
+#include <linux/mmap_lock.h>
 #include <linux/syscalls.h>
 #include <linux/swap.h>
 #include <linux/leafops.h>
@@ -232,27 +233,24 @@ static inline bool can_do_mincore(struct vm_area_struct *vma)
 	       file_permission(vma->vm_file, MAY_WRITE) == 0;
 }
 
-static const struct mm_walk_ops mincore_walk_ops = {
-	.pmd_entry		= mincore_pte_range,
-	.pte_hole		= mincore_unmapped_range,
-	.hugetlb_entry		= mincore_hugetlb,
-	.walk_lock		= PGWALK_RDLOCK,
-};
-
 /*
  * Do a chunk of "sys_mincore()". We've already checked
- * all the arguments, we hold the mmap semaphore: we should
- * just return the amount of info we're asked for.
+ * all the arguments, we should just return the amount of
+ * info we're asked for.
  */
-static long do_mincore(unsigned long addr, unsigned long pages, unsigned char *vec)
+static long do_mincore(struct vm_area_struct *vma, unsigned long addr,
+		unsigned long pages, unsigned char *vec,
+		enum page_walk_lock walk_lock)
 {
-	struct vm_area_struct *vma;
 	unsigned long end;
 	int err;
+	struct mm_walk_ops mincore_walk_ops = {
+		.pmd_entry		= mincore_pte_range,
+		.pte_hole		= mincore_unmapped_range,
+		.hugetlb_entry		= mincore_hugetlb,
+		.walk_lock		= walk_lock,
+	};
 
-	vma = vma_lookup(current->mm, addr);
-	if (!vma)
-		return -ENOMEM;
 	end = min(vma->vm_end, addr + (pages << PAGE_SHIFT));
 	if (!can_do_mincore(vma)) {
 		unsigned long pages = DIV_ROUND_UP(end - addr, PAGE_SIZE);
@@ -328,13 +326,34 @@ SYSCALL_DEFINE3(mincore, unsigned long, start, size_t, len,
 
 	retval = 0;
 	while (pages) {
+		struct mm_struct *mm = current->mm;
+		enum page_walk_lock walk_lock;
+		struct vm_area_struct *vma;
+
+		vma = lock_vma_under_rcu(mm, start);
+		if (vma) {
+			walk_lock = PGWALK_VMA_RDLOCK_VERIFY;
+		} else {
+			mmap_read_lock(mm);
+			vma = vma_lookup(mm, start);
+			if (!vma) {
+				mmap_read_unlock(mm);
+				retval = -ENOMEM;
+				break;
+			}
+			walk_lock = PGWALK_RDLOCK;
+		}
+
 		/*
 		 * Do at most PAGE_SIZE entries per iteration, due to
 		 * the temporary buffer size.
 		 */
-		mmap_read_lock(current->mm);
-		retval = do_mincore(start, min(pages, PAGE_SIZE), tmp);
-		mmap_read_unlock(current->mm);
+		retval = do_mincore(vma, start, min(pages, PAGE_SIZE), tmp, walk_lock);
+
+		if (walk_lock == PGWALK_VMA_RDLOCK_VERIFY)
+			vma_end_read(vma);
+		else
+			mmap_read_unlock(mm);
 
 		if (retval <= 0)
 			break;
-- 
2.27.0

[PATCH mm-new 2/5] mm: mincore: remove special handing for VM_PFNMAP

[Kefeng Wang]

As David pointed out, "it’s hard to believe that someone depends
on pages in VM_PFNMAP to *not* be present", so remove the historical
behavior that always reports VM_PFNMAP pages as non-resident,
simplifying the code.

Link: https://lore.kernel.org/linux-mm/0e619d71-1c3d-4534-8376-2982c7348c31@kernel.org/
Suggested-by: David Hildenbrand (Arm) <david@kernel.org>
Signed-off-by: Kefeng Wang <wangkefeng.wang@huawei.com>
---
 mm/mincore.c | 8 --------
 1 file changed, 8 deletions(-)

diff --git a/mm/mincore.c b/mm/mincore.c
index f561d6e01052..6374819dd5c1 100644
--- a/mm/mincore.c
+++ b/mm/mincore.c
@@ -258,14 +258,6 @@ static long do_mincore(struct vm_area_struct *vma, unsigned long addr,
 		return pages;
 	}
 
-	/*
-	 * mincore historically reports PFNMAP mappings as non-resident.
-	 */
-	if (vma->vm_flags & VM_PFNMAP) {
-		__mincore_unmapped_range(addr, end, vma, vec);
-		return (end - addr) >> PAGE_SHIFT;
-	}
-
 	err = walk_page_range_vma(vma, addr, end, &mincore_walk_ops, vec);
 	if (err < 0)
 		return err;
-- 
2.27.0

[PATCH mm-new 3/5] mm: mincore: replace __get_free_page() with kmalloc()

[Kefeng Wang]

Remove ugly casts by using the more natural kmalloc/kfree allocation.

Signed-off-by: Kefeng Wang <wangkefeng.wang@huawei.com>
---
 mm/mincore.c | 5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)

diff --git a/mm/mincore.c b/mm/mincore.c
index 6374819dd5c1..d561f1e5c461 100644
--- a/mm/mincore.c
+++ b/mm/mincore.c
@@ -13,6 +13,7 @@
 #include <linux/pagewalk.h>
 #include <linux/mman.h>
 #include <linux/mmap_lock.h>
+#include <linux/slab.h>
 #include <linux/syscalls.h>
 #include <linux/swap.h>
 #include <linux/leafops.h>
@@ -312,7 +313,7 @@ SYSCALL_DEFINE3(mincore, unsigned long, start, size_t, len,
 	if (!access_ok(vec, pages))
 		return -EFAULT;
 
-	tmp = (void *) __get_free_page(GFP_USER);
+	tmp = kmalloc(PAGE_SIZE, GFP_USER);
 	if (!tmp)
 		return -EAGAIN;
 
@@ -358,6 +359,6 @@ SYSCALL_DEFINE3(mincore, unsigned long, start, size_t, len,
 		start += retval << PAGE_SHIFT;
 		retval = 0;
 	}
-	free_page((unsigned long) tmp);
+	kfree(tmp);
 	return retval;
 }
-- 
2.27.0

[PATCH mm-new 4/5] mm: mincore: remove xa_is_value() in mincore_swap()

[Kefeng Wang]

The swap_cache_get_folio() no longer returns shadow entries, so the 
xa_is_value() check is unnecessary.

Signed-off-by: Kefeng Wang <wangkefeng.wang@huawei.com>
---
 mm/mincore.c | 3 +--
 1 file changed, 1 insertion(+), 2 deletions(-)

diff --git a/mm/mincore.c b/mm/mincore.c
index d561f1e5c461..1c72cc3ac87b 100644
--- a/mm/mincore.c
+++ b/mm/mincore.c
@@ -92,8 +92,7 @@ static unsigned char mincore_swap(swp_entry_t entry, bool shmem)
 	folio = swap_cache_get_folio(entry);
 	if (shmem)
 		put_swap_device(si);
-	/* The swap cache space contains either folio, shadow or NULL */
-	if (folio && !xa_is_value(folio)) {
+	if (folio) {
 		present = folio_test_uptodate(folio);
 		folio_put(folio);
 	}
-- 
2.27.0

[PATCH mm-new 5/5] mm: mincore: improve mincore_hugetlb()

[Kefeng Wang]

The walk_hugetlb_range() always passes a non-NULL pte, so remove the   
dead NULL check. Replace the per-page iteration loop with memset() 
for better performance. 

Signed-off-by: Kefeng Wang <wangkefeng.wang@huawei.com>
---
 mm/mincore.c | 27 +++++++--------------------
 1 file changed, 7 insertions(+), 20 deletions(-)

diff --git a/mm/mincore.c b/mm/mincore.c
index 1c72cc3ac87b..1a2568734e34 100644
--- a/mm/mincore.c
+++ b/mm/mincore.c
@@ -29,30 +29,17 @@ static int mincore_hugetlb(pte_t *pte, unsigned long hmask, unsigned long addr,
 			unsigned long end, struct mm_walk *walk)
 {
 #ifdef CONFIG_HUGETLB_PAGE
-	unsigned char present;
+	unsigned long nr = (end - addr) >> PAGE_SHIFT;
 	unsigned char *vec = walk->private;
+	unsigned char present;
 	spinlock_t *ptl;
+	pte_t ptep;
 
 	ptl = huge_pte_lock(hstate_vma(walk->vma), walk->mm, pte);
-
-	/*
-	 * Hugepages under user process are always in RAM and never
-	 * swapped out, but theoretically it needs to be checked.
-	 */
-	if (!pte) {
-		present = 0;
-	} else {
-		const pte_t ptep = huge_ptep_get(walk->mm, addr, pte);
-
-		if (huge_pte_none(ptep) || pte_is_marker(ptep))
-			present = 0;
-		else
-			present = 1;
-	}
-
-	for (; addr != end; vec++, addr += PAGE_SIZE)
-		*vec = present;
-	walk->private = vec;
+	ptep = huge_ptep_get(walk->mm, addr, pte);
+	present = !(huge_pte_none(ptep) || pte_is_marker(ptep));
+	memset(vec, present, nr);
+	walk->private += nr;
 	spin_unlock(ptl);
 #else
 	BUG();
-- 
2.27.0

Re: [PATCH mm-new 0/5] mm: mincore: reduce lock contention and misc cleanups

[Andrew Morton]

On Wed, 1 Jul 2026 22:40:42 +0800 Kefeng Wang <wangkefeng.wang@huawei.com> wrote:

> Kefeng Wang (5):
>   mm: mincore: attempt per-vma lock during page table walk
>   mm: mincore: remove special handing for VM_PFNMAP
>   mm: mincore: replace __get_free_page() with kmalloc()
>   mm: mincore: remove xa_is_value() in mincore_swap()
>   mm: mincore: improve mincore_hugetlb()
> 
>  mm/mincore.c | 96 +++++++++++++++++++++++++---------------------------
>  1 file changed, 47 insertions(+), 49 deletions(-)

That was brief ;)

Thanks, I'll await reviewer input on this series.  AI review said a
couple of things:
	https://sashiko.dev/#/patchset/20260701144047.3786939-1-wangkefeng.wang@huawei.com

Re: [PATCH mm-new 3/5] mm: mincore: replace __get_free_page() with kmalloc()

[Andrew Morton]

On Wed, 1 Jul 2026 22:40:45 +0800 Kefeng Wang <wangkefeng.wang@huawei.com> wrote:

> Remove ugly casts by using the more natural kmalloc/kfree allocation.
> 
> --- a/mm/mincore.c
> +++ b/mm/mincore.c
> @@ -13,6 +13,7 @@
>  #include <linux/pagewalk.h>
>  #include <linux/mman.h>
>  #include <linux/mmap_lock.h>
> +#include <linux/slab.h>
>  #include <linux/syscalls.h>
>  #include <linux/swap.h>
>  #include <linux/leafops.h>
> @@ -312,7 +313,7 @@ SYSCALL_DEFINE3(mincore, unsigned long, start, size_t, len,
>  	if (!access_ok(vec, pages))
>  		return -EFAULT;
>  
> -	tmp = (void *) __get_free_page(GFP_USER);
> +	tmp = kmalloc(PAGE_SIZE, GFP_USER);

kmalloc(GFP_USER) is a weird combination.  I don't think GFP_USER ever
made sense in here.

>  	if (!tmp)
>  		return -EAGAIN;
>  
> @@ -358,6 +359,6 @@ SYSCALL_DEFINE3(mincore, unsigned long, start, size_t, len,
>  		start += retval << PAGE_SHIFT;
>  		retval = 0;
>  	}
> -	free_page((unsigned long) tmp);
> +	kfree(tmp);
>  	return retval;
>  }