[PATCH v2] mm: mincore: use pte_batch_bint() to batch process large folios

[PATCH v2] mm: mincore: use pte_batch_bint() to batch process large folios

[Baolin Wang]

When I tested the mincore() syscall, I observed that it takes longer with
64K mTHP enabled on my Arm64 server. The reason is the mincore_pte_range()
still checks each PTE individually, even when the PTEs are contiguous,
which is not efficient.

Thus we can use pte_batch_hint() to get the batch number of the present
contiguous PTEs, which can improve the performance. I tested the mincore()
syscall with 1G anonymous memory populated with 64K mTHP, and observed an
obvious performance improvement:

w/o patch		w/ patch		changes
6022us			549us			+91%

Moreover, I also tested mincore() with disabling mTHP/THP, and did not
see any obvious regression for base pages.

Signed-off-by: Baolin Wang <baolin.wang@linux.alibaba.com>
---
Changes from v1:
 - Change to use pte_batch_hint() to get the batch number, per Ryan.

Note: I observed the min_t() can introduce a slight performance regression
for base pages, so I change to add a batch size check for base pages,
which can resolve the performance regression issue.
---
 mm/mincore.c | 19 ++++++++++++++-----
 1 file changed, 14 insertions(+), 5 deletions(-)

diff --git a/mm/mincore.c b/mm/mincore.c
index 832f29f46767..2e6a9123305e 100644
--- a/mm/mincore.c
+++ b/mm/mincore.c
@@ -21,6 +21,7 @@
 
 #include <linux/uaccess.h>
 #include "swap.h"
+#include "internal.h"
 
 static int mincore_hugetlb(pte_t *pte, unsigned long hmask, unsigned long addr,
 			unsigned long end, struct mm_walk *walk)
@@ -105,6 +106,7 @@ static int mincore_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
 	pte_t *ptep;
 	unsigned char *vec = walk->private;
 	int nr = (end - addr) >> PAGE_SHIFT;
+	int step, i;
 
 	ptl = pmd_trans_huge_lock(pmd, vma);
 	if (ptl) {
@@ -118,16 +120,23 @@ static int mincore_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
 		walk->action = ACTION_AGAIN;
 		return 0;
 	}
-	for (; addr != end; ptep++, addr += PAGE_SIZE) {
+	for (; addr != end; ptep += step, addr += step * PAGE_SIZE) {
 		pte_t pte = ptep_get(ptep);
 
+		step = 1;
 		/* We need to do cache lookup too for pte markers */
 		if (pte_none_mostly(pte))
 			__mincore_unmapped_range(addr, addr + PAGE_SIZE,
 						 vma, vec);
-		else if (pte_present(pte))
-			*vec = 1;
-		else { /* pte is a swap entry */
+		else if (pte_present(pte)) {
+			unsigned int batch = pte_batch_hint(ptep, pte);
+
+			if (batch > 1)
+				step = min_t(unsigned int, batch, nr);
+
+			for (i = 0; i < step; i++)
+				vec[i] = 1;
+		} else { /* pte is a swap entry */
 			swp_entry_t entry = pte_to_swp_entry(pte);
 
 			if (non_swap_entry(entry)) {
@@ -146,7 +155,7 @@ static int mincore_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
 #endif
 			}
 		}
-		vec++;
+		vec += step;
 	}
 	pte_unmap_unlock(ptep - 1, ptl);
 out:
-- 
2.43.5

Re: [PATCH v2] mm: mincore: use pte_batch_bint() to batch process large folios

[Barry Song]

On Thu, May 8, 2025 at 4:09 PM Baolin Wang
<baolin.wang@linux.alibaba.com> wrote:
>
> When I tested the mincore() syscall, I observed that it takes longer with
> 64K mTHP enabled on my Arm64 server. The reason is the mincore_pte_range()
> still checks each PTE individually, even when the PTEs are contiguous,
> which is not efficient.
>
> Thus we can use pte_batch_hint() to get the batch number of the present
> contiguous PTEs, which can improve the performance. I tested the mincore()
> syscall with 1G anonymous memory populated with 64K mTHP, and observed an
> obvious performance improvement:
>
> w/o patch               w/ patch                changes
> 6022us                  549us                   +91%
>
> Moreover, I also tested mincore() with disabling mTHP/THP, and did not
> see any obvious regression for base pages.
>
> Signed-off-by: Baolin Wang <baolin.wang@linux.alibaba.com>
> ---
> Changes from v1:
>  - Change to use pte_batch_hint() to get the batch number, per Ryan.
>
> Note: I observed the min_t() can introduce a slight performance regression
> for base pages, so I change to add a batch size check for base pages,
> which can resolve the performance regression issue.
> ---
>  mm/mincore.c | 19 ++++++++++++++-----
>  1 file changed, 14 insertions(+), 5 deletions(-)
>
> diff --git a/mm/mincore.c b/mm/mincore.c
> index 832f29f46767..2e6a9123305e 100644
> --- a/mm/mincore.c
> +++ b/mm/mincore.c
> @@ -21,6 +21,7 @@
>
>  #include <linux/uaccess.h>
>  #include "swap.h"
> +#include "internal.h"
>
>  static int mincore_hugetlb(pte_t *pte, unsigned long hmask, unsigned long addr,
>                         unsigned long end, struct mm_walk *walk)
> @@ -105,6 +106,7 @@ static int mincore_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
>         pte_t *ptep;
>         unsigned char *vec = walk->private;
>         int nr = (end - addr) >> PAGE_SHIFT;
> +       int step, i;
>
>         ptl = pmd_trans_huge_lock(pmd, vma);
>         if (ptl) {
> @@ -118,16 +120,23 @@ static int mincore_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
>                 walk->action = ACTION_AGAIN;
>                 return 0;
>         }
> -       for (; addr != end; ptep++, addr += PAGE_SIZE) {
> +       for (; addr != end; ptep += step, addr += step * PAGE_SIZE) {
>                 pte_t pte = ptep_get(ptep);
>
> +               step = 1;
>                 /* We need to do cache lookup too for pte markers */
>                 if (pte_none_mostly(pte))
>                         __mincore_unmapped_range(addr, addr + PAGE_SIZE,
>                                                  vma, vec);
> -               else if (pte_present(pte))
> -                       *vec = 1;
> -               else { /* pte is a swap entry */
> +               else if (pte_present(pte)) {
> +                       unsigned int batch = pte_batch_hint(ptep, pte);
> +
> +                       if (batch > 1)
> +                               step = min_t(unsigned int, batch, nr);

Not quite sure if nr should be (end - addr) / PAGE_SIZE as nr
is always the initial value. For example, nr = 50, and we have
scanned 48 PTEs, then we have 2 ptes left. No?

> +
> +                       for (i = 0; i < step; i++)
> +                               vec[i] = 1;
> +               } else { /* pte is a swap entry */
>                         swp_entry_t entry = pte_to_swp_entry(pte);
>
>                         if (non_swap_entry(entry)) {
> @@ -146,7 +155,7 @@ static int mincore_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
>  #endif
>                         }
>                 }
> -               vec++;
> +               vec += step;
>         }
>         pte_unmap_unlock(ptep - 1, ptl);
>  out:
> --
> 2.43.5
>

Thanks
Barry

Re: [PATCH v2] mm: mincore: use pte_batch_bint() to batch process large folios

[Baolin Wang]

On 2025/5/8 15:42, Barry Song wrote:
> On Thu, May 8, 2025 at 4:09 PM Baolin Wang
> <baolin.wang@linux.alibaba.com> wrote:
>>
>> When I tested the mincore() syscall, I observed that it takes longer with
>> 64K mTHP enabled on my Arm64 server. The reason is the mincore_pte_range()
>> still checks each PTE individually, even when the PTEs are contiguous,
>> which is not efficient.
>>
>> Thus we can use pte_batch_hint() to get the batch number of the present
>> contiguous PTEs, which can improve the performance. I tested the mincore()
>> syscall with 1G anonymous memory populated with 64K mTHP, and observed an
>> obvious performance improvement:
>>
>> w/o patch               w/ patch                changes
>> 6022us                  549us                   +91%
>>
>> Moreover, I also tested mincore() with disabling mTHP/THP, and did not
>> see any obvious regression for base pages.
>>
>> Signed-off-by: Baolin Wang <baolin.wang@linux.alibaba.com>
>> ---
>> Changes from v1:
>>   - Change to use pte_batch_hint() to get the batch number, per Ryan.
>>
>> Note: I observed the min_t() can introduce a slight performance regression
>> for base pages, so I change to add a batch size check for base pages,
>> which can resolve the performance regression issue.
>> ---
>>   mm/mincore.c | 19 ++++++++++++++-----
>>   1 file changed, 14 insertions(+), 5 deletions(-)
>>
>> diff --git a/mm/mincore.c b/mm/mincore.c
>> index 832f29f46767..2e6a9123305e 100644
>> --- a/mm/mincore.c
>> +++ b/mm/mincore.c
>> @@ -21,6 +21,7 @@
>>
>>   #include <linux/uaccess.h>
>>   #include "swap.h"
>> +#include "internal.h"
>>
>>   static int mincore_hugetlb(pte_t *pte, unsigned long hmask, unsigned long addr,
>>                          unsigned long end, struct mm_walk *walk)
>> @@ -105,6 +106,7 @@ static int mincore_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
>>          pte_t *ptep;
>>          unsigned char *vec = walk->private;
>>          int nr = (end - addr) >> PAGE_SHIFT;
>> +       int step, i;
>>
>>          ptl = pmd_trans_huge_lock(pmd, vma);
>>          if (ptl) {
>> @@ -118,16 +120,23 @@ static int mincore_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
>>                  walk->action = ACTION_AGAIN;
>>                  return 0;
>>          }
>> -       for (; addr != end; ptep++, addr += PAGE_SIZE) {
>> +       for (; addr != end; ptep += step, addr += step * PAGE_SIZE) {
>>                  pte_t pte = ptep_get(ptep);
>>
>> +               step = 1;
>>                  /* We need to do cache lookup too for pte markers */
>>                  if (pte_none_mostly(pte))
>>                          __mincore_unmapped_range(addr, addr + PAGE_SIZE,
>>                                                   vma, vec);
>> -               else if (pte_present(pte))
>> -                       *vec = 1;
>> -               else { /* pte is a swap entry */
>> +               else if (pte_present(pte)) {
>> +                       unsigned int batch = pte_batch_hint(ptep, pte);
>> +
>> +                       if (batch > 1)
>> +                               step = min_t(unsigned int, batch, nr);
> 
> Not quite sure if nr should be (end - addr) / PAGE_SIZE as nr
> is always the initial value. For example, nr = 50, and we have
> scanned 48 PTEs, then we have 2 ptes left. No?

Ah, you are right. I missed this part when I revised the original 
patch[1]. Thanks for pointing this out.

[1] 
https://lore.kernel.org/all/6a8418ba-dbd1-489f-929b-e31831bea0cf@linux.alibaba.com/