[PATCH] mm/memory: reuse the whole exclusive large folio on a write fault

[PATCH] mm/memory: reuse the whole exclusive large folio on a write fault

[Yuan-Hao Hsu]

fork() maps the anonymous pages of the parent read-only and clears
PageAnonExclusive on them.  Once the child has exec'ed or exited, a
write fault of the parent ends up in the reuse path of do_wp_page():
wp_can_reuse_anon_folio() finds that all references to the folio come
from mappings in this MM, the page is marked exclusive again and its PTE
is made writable.

For a large folio that check is about the folio, and what it finds holds
for every page of it.  Still only the page that faulted is marked
exclusive and only its PTE becomes writable, so each of the other pages
takes a write fault of its own, and each of those takes the large
mapcount lock to find out the same thing again: 16 faults for a 64K
folio, 512 for a 2M THP that is mapped by PTEs.  The same THP mapped by
a PMD is reused by one fault in do_huge_pmd_wp_page(), do_swap_page()
maps all PTEs of an exclusive large folio writable at once, and
numa_rebuild_large_mapping() upgrades all PTEs of the folio from one
hinting fault.

Commit 1da190f4d0a6 ("mm: Copy-on-Write (COW) reuse support for
PTE-mapped THP") left this for later because faulting around might
increase the COW latency.  Numbers for that are below.

When a large folio has been found exclusive, walk the part of it that
this page table maps inside the VMA.  PTEs that still map it read-only
are batched with folio_pte_batch_flags(), their pages are marked
exclusive, and where can_change_pte_writable() agrees the batch is made
writable with modify_prot_start_ptes()/modify_prot_commit_ptes(), the
way mprotect() does it.  That leaves NUMA hinting and uffd-wp PTEs
alone, keeps soft-dirty tracking exact, and on arm64 writes a contpte
block back as a whole where ptep_set_access_flags() on a single PTE has
to unfold it.  Pages that cannot be made writable are marked exclusive
all the same, so their own fault skips the folio check.  The PTE that
faulted is in one of the batches and is then completed by
wp_page_reuse() as before.  Small folios, PMD-mapped THPs, unsharing
faults and the copy path are not changed.

x86-64, i7-12700KF, 256 MiB of anonymous memory, fork(), the child
exits, then the parent stores to the memory.  Medians of 15 runs, two
boots of each kernel, taken alternately:

                          v7.3-rc3+           patched
  one byte per page, a clock_gettime() between the stores
    write faults
      4K pages            65,601              65,601
      64K mTHP            65,601               4,161
      1M mTHP             65,601                 321
      2M THP, PTE-mapped  65,601                 193
      2M THP, PMD-mapped     193                 193
    time (ms)
      4K pages            29.7 / 30.1         30.0 / 31.1
      64K mTHP            29.4 / 29.7          5.5 /  5.6
      1M mTHP             29.3 / 29.8          3.9 /  3.9
      2M THP, PTE-mapped  29.9 / 29.2          3.8 /  3.9
      2M THP, PMD-mapped   2.5 /  2.5          2.5 /  2.6
  memset() of all of it (ms)
      4K pages            57.2 / 56.5         56.9 / 57.1
      64K mTHP            56.5 / 56.3         39.0 / 38.7
      2M THP, PTE-mapped  58.6 / 56.5         36.8 / 37.7
      2M THP, PMD-mapped  37.5 / 35.9         35.9 / 36.5
  8 threads, one byte per page, random order (ms)
      64K mTHP             6.1 /  6.2          0.9 /  1.0
      2M THP, PTE-mapped   6.0 /  6.7          0.7 /  0.7

The latency of the one fault that now does the work for the folio,
measured as the time of the store that takes it, against 420 ns for a
reuse fault today (medians, ns):

                          reuse fault   fault that     COW fault that
                          (patched)     allocated it   copies 4K
      64K mTHP                   730          4,600          1,500
      1M mTHP                  5,500         63,000          1,500
      2M THP, PTE-mapped      10,100        126,000          1,500

That is 14 to 20 ns per PTE.  Builds that differ only by NOPs in front
of the new function take either 10,100 or 7,500 ns for the 2M folio,
with a period of 32 bytes: it is the loop of modify_prot_commit_ptes()
that changes speed with its address.

Capping the walk to the 16 PTEs around the fault instead was measured as
well: it takes 5.4 ms where the above takes 3.9 ms on 1M and 2M folios,
it is slower than today when only one page per 64K is written (3.1 ms
against 1.9 ms, the whole folio takes 1.5 ms), and it is only ahead when
no more than one page per 2M is ever written (0.1 ms against 1.3 ms for
the 256 MiB).

Redis 7.0.15 with 1.3 M keys of 512 bytes on 64K mTHP, BGSAVE and then
1,000,000 SETs: the faults of redis-server during the SETs go from
262,100 to 18,500, its CPU time from 1.48-1.53 s to 1.25-1.33 s, and
redis-benchmark reports 742,000 to 794,000 requests per second instead
of 652,000 to 658,000.  With THP off all three stay where they were.

arm64 was only run under QEMU, for the counters and with DEBUG_VM and
PAGE_TABLE_CHECK: the faults are the same as above, and with 64K folios
the first pass after fork() unfolds every contpte block today (512
contpte_convert() calls for 512 blocks, and nothing folds them again)
while none is unfolded with this patch.

What does not get faster on x86 are stores to pages that this CPU still
has a read-only TLB entry for.  The fault makes the PTEs writable but,
like mprotect(), does not flush, so such a page still takes a fault, a
spurious one that costs about the same as the reuse fault it replaces.
That happens to pages that were read since fork(): reading the 16 pages
of every 64K folio before storing to them takes 65,500 faults and 28 ms
before and after.  And it happens in a loop that does nothing but store
one byte to every page in ascending order, which is what the reuse-byte
mode of David's pte-mapped-folio-benchmarks does (120 ms before and
after for 1 GiB of 64K folios; 2M folios: 119 ms to 18 ms; the reuse
mode, a memset(), goes from 232 ms to 157 ms with 64K folios): while the
first store of a folio is faulting, the CPU has already run the next
stores speculatively and has filled the TLB with the read-only
translations of their pages.  Counting with kprobes, such a run enters
handle_mm_fault() 135,687 times and do_wp_page() 8,457 times; with an
LFENCE after every store the faults are 4,100 instead of 65,400 and the
loop takes 4 ms instead of 29 ms.  The PMD-mapped case, which this patch
does not touch, shows the same: 8,800 faults for 128 THPs, 133 with the
LFENCE.

With a flush_tlb_local() in the new function, as an experiment, the
ascending loop takes 4.1 ms instead of 28 ms on 64K folios, the
read-then-store loop 5.3 ms instead of 28 ms and the memset() 21 ms
instead of 38 ms, for a fault of 870 instead of 730 ns and 11% more time
for the pass with the clock_gettime().  Generic code has no way to ask
x86 for a flush that stays on this CPU, so that is left for later.

Assisted-by: LLM sparse
Signed-off-by: Yuan-Hao Hsu <aa9736195201@gmail.com>
---
 mm/memory.c | 68 +++++++++++++++++++++++++++++++++++++++++++++++++++--
 1 file changed, 66 insertions(+), 2 deletions(-)

diff --git a/mm/memory.c b/mm/memory.c
index 8b0c2c735d3d..85c883d1e558 100644
--- a/mm/memory.c
+++ b/mm/memory.c
@@ -4355,6 +4355,64 @@ static bool wp_can_reuse_anon_folio(struct folio *folio,
 	return true;
 }
 
+/*
+ * wp_can_reuse_anon_folio() found a large folio to be exclusive to this MM.
+ * That holds for all of its pages and not only for the one that faulted: mark
+ * the ones that this page table maps exclusive as well and map them writable,
+ * like mprotect() would. Each of them would otherwise take a write fault of
+ * its own that repeats the check on the very same folio.
+ *
+ * The PTE that faulted is among them; wp_page_reuse() completes it.
+ */
+static void wp_reuse_large_anon_folio(struct vm_fault *vmf,
+		struct folio *folio)
+{
+	const fpb_t flags = FPB_RESPECT_WRITE | FPB_RESPECT_SOFT_DIRTY;
+	const unsigned long idx = folio_page_idx(folio, vmf->page);
+	struct vm_area_struct *vma = vmf->vma;
+	unsigned long addr = vmf->address;
+	unsigned long pt_start = ALIGN_DOWN(addr, PMD_SIZE);
+	unsigned long nr_before, nr_after, end;
+	struct page *page;
+	unsigned int nr, i;
+	pte_t *ptep, pte;
+
+	/* Stay within the folio, the VMA and the page table. */
+	nr_before = min3(idx, (addr - pt_start) >> PAGE_SHIFT,
+			 (addr - vma->vm_start) >> PAGE_SHIFT);
+	nr_after = min3(folio_nr_pages(folio) - idx,
+			(pt_start + PMD_SIZE - addr) >> PAGE_SHIFT,
+			(vma->vm_end - addr) >> PAGE_SHIFT);
+	end = addr + (nr_after << PAGE_SHIFT);
+	addr -= nr_before << PAGE_SHIFT;
+	ptep = vmf->pte - nr_before;
+	page = vmf->page - nr_before;
+
+	for (; addr != end; addr += nr * PAGE_SIZE, ptep += nr, page += nr) {
+		pte = ptep_get(ptep);
+		nr = 1;
+
+		/* Unmapped or replaced since, or writable already. */
+		if (!pte_present(pte) || pte_pfn(pte) != page_to_pfn(page) ||
+		    pte_write(pte))
+			continue;
+
+		nr = folio_pte_batch_flags(folio, NULL, ptep, &pte,
+					   (end - addr) >> PAGE_SHIFT, flags);
+		for (i = 0; i < nr; i++)
+			if (!PageAnonExclusive(page + i))
+				SetPageAnonExclusive(page + i);
+
+		/* The PTEs of a batch agree on everything this looks at. */
+		if (!can_change_pte_writable(vma, addr, pte))
+			continue;
+
+		pte = modify_prot_start_ptes(vma, addr, ptep, nr);
+		modify_prot_commit_ptes(vma, addr, ptep, pte,
+					pte_mkwrite(pte, vma), nr);
+	}
+}
+
 /*
  * This routine handles present pages, when
  * * users try to write to a shared page (FAULT_FLAG_WRITE)
@@ -4449,8 +4507,14 @@ static vm_fault_t do_wp_page(struct vm_fault *vmf)
 	 */
 	if (folio && folio_test_anon(folio) &&
 	    (PageAnonExclusive(vmf->page) || wp_can_reuse_anon_folio(folio, vma))) {
-		if (!PageAnonExclusive(vmf->page))
-			SetPageAnonExclusive(vmf->page);
+		if (!PageAnonExclusive(vmf->page)) {
+			if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) &&
+			    folio_test_large(folio) && likely(!unshare) &&
+			    likely(vma->vm_flags & VM_WRITE))
+				wp_reuse_large_anon_folio(vmf, folio);
+			else
+				SetPageAnonExclusive(vmf->page);
+		}
 		if (unlikely(unshare)) {
 			pte_unmap_unlock(vmf->pte, vmf->ptl);
 			return 0;

base-commit: 238650ef6c7c7cca08e032527329424c9fbd70e5
-- 
2.43.0

Re: [PATCH] mm/memory: reuse the whole exclusive large folio on a write fault

[David Hildenbrand (Arm)]

On 9/18/26 08:42, Yuan-Hao Hsu wrote:
> fork() maps the anonymous pages of the parent read-only and clears
> PageAnonExclusive on them.  Once the child has exec'ed or exited, a
> write fault of the parent ends up in the reuse path of do_wp_page():
> wp_can_reuse_anon_folio() finds that all references to the folio come
> from mappings in this MM, the page is marked exclusive again and its PTE
> is made writable.
> 
> For a large folio that check is about the folio, and what it finds holds
> for every page of it.  Still only the page that faulted is marked
> exclusive and only its PTE becomes writable, so each of the other pages
> takes a write fault of its own, and each of those takes the large
> mapcount lock to find out the same thing again: 16 faults for a 64K
> folio, 512 for a 2M THP that is mapped by PTEs.  The same THP mapped by
> a PMD is reused by one fault in do_huge_pmd_wp_page(), do_swap_page()
> maps all PTEs of an exclusive large folio writable at once, and
> numa_rebuild_large_mapping() upgrades all PTEs of the folio from one
> hinting fault.
> 
> Commit 1da190f4d0a6 ("mm: Copy-on-Write (COW) reuse support for
> PTE-mapped THP") left this for later because faulting around might
> increase the COW latency.  Numbers for that are below.
> 
> When a large folio has been found exclusive, walk the part of it that
> this page table maps inside the VMA.  PTEs that still map it read-only
> are batched with folio_pte_batch_flags(), their pages are marked
> exclusive, and where can_change_pte_writable() agrees the batch is made
> writable with modify_prot_start_ptes()/modify_prot_commit_ptes(), the
> way mprotect() does it.  That leaves NUMA hinting and uffd-wp PTEs
> alone, keeps soft-dirty tracking exact, and on arm64 writes a contpte
> block back as a whole where ptep_set_access_flags() on a single PTE has
> to unfold it.  Pages that cannot be made writable are marked exclusive
> all the same, so their own fault skips the folio check.  The PTE that
> faulted is in one of the batches and is then completed by
> wp_page_reuse() as before.  Small folios, PMD-mapped THPs, unsharing
> faults and the copy path are not changed.
> 
> x86-64, i7-12700KF, 256 MiB of anonymous memory, fork(), the child
> exits, then the parent stores to the memory.  Medians of 15 runs, two
> boots of each kernel, taken alternately:
> 
>                           v7.3-rc3+           patched
>   one byte per page, a clock_gettime() between the stores
>     write faults
>       4K pages            65,601              65,601
>       64K mTHP            65,601               4,161
>       1M mTHP             65,601                 321
>       2M THP, PTE-mapped  65,601                 193
>       2M THP, PMD-mapped     193                 193
>     time (ms)
>       4K pages            29.7 / 30.1         30.0 / 31.1
>       64K mTHP            29.4 / 29.7          5.5 /  5.6
>       1M mTHP             29.3 / 29.8          3.9 /  3.9
>       2M THP, PTE-mapped  29.9 / 29.2          3.8 /  3.9
>       2M THP, PMD-mapped   2.5 /  2.5          2.5 /  2.6
>   memset() of all of it (ms)
>       4K pages            57.2 / 56.5         56.9 / 57.1
>       64K mTHP            56.5 / 56.3         39.0 / 38.7
>       2M THP, PTE-mapped  58.6 / 56.5         36.8 / 37.7
>       2M THP, PMD-mapped  37.5 / 35.9         35.9 / 36.5
>   8 threads, one byte per page, random order (ms)
>       64K mTHP             6.1 /  6.2          0.9 /  1.0
>       2M THP, PTE-mapped   6.0 /  6.7          0.7 /  0.7
> 
> The latency of the one fault that now does the work for the folio,
> measured as the time of the store that takes it, against 420 ns for a
> reuse fault today (medians, ns):
> 
>                           reuse fault   fault that     COW fault that
>                           (patched)     allocated it   copies 4K
>       64K mTHP                   730          4,600          1,500
>       1M mTHP                  5,500         63,000          1,500
>       2M THP, PTE-mapped      10,100        126,000          1,500
> 
> That is 14 to 20 ns per PTE.  Builds that differ only by NOPs in front
> of the new function take either 10,100 or 7,500 ns for the 2M folio,
> with a period of 32 bytes: it is the loop of modify_prot_commit_ptes()
> that changes speed with its address.
> 
> Capping the walk to the 16 PTEs around the fault instead was measured as
> well: it takes 5.4 ms where the above takes 3.9 ms on 1M and 2M folios,
> it is slower than today when only one page per 64K is written (3.1 ms
> against 1.9 ms, the whole folio takes 1.5 ms), and it is only ahead when
> no more than one page per 2M is ever written (0.1 ms against 1.3 ms for
> the 256 MiB).
> 
> Redis 7.0.15 with 1.3 M keys of 512 bytes on 64K mTHP, BGSAVE and then
> 1,000,000 SETs: the faults of redis-server during the SETs go from
> 262,100 to 18,500, its CPU time from 1.48-1.53 s to 1.25-1.33 s, and
> redis-benchmark reports 742,000 to 794,000 requests per second instead
> of 652,000 to 658,000.  With THP off all three stay where they were.
> 
> arm64 was only run under QEMU, for the counters and with DEBUG_VM and
> PAGE_TABLE_CHECK: the faults are the same as above, and with 64K folios
> the first pass after fork() unfolds every contpte block today (512
> contpte_convert() calls for 512 blocks, and nothing folds them again)
> while none is unfolded with this patch.
> 
> What does not get faster on x86 are stores to pages that this CPU still
> has a read-only TLB entry for.  The fault makes the PTEs writable but,
> like mprotect(), does not flush, so such a page still takes a fault, a
> spurious one that costs about the same as the reuse fault it replaces.
> That happens to pages that were read since fork(): reading the 16 pages
> of every 64K folio before storing to them takes 65,500 faults and 28 ms
> before and after.  And it happens in a loop that does nothing but store
> one byte to every page in ascending order, which is what the reuse-byte
> mode of David's pte-mapped-folio-benchmarks does (120 ms before and
> after for 1 GiB of 64K folios; 2M folios: 119 ms to 18 ms; the reuse
> mode, a memset(), goes from 232 ms to 157 ms with 64K folios): while the
> first store of a folio is faulting, the CPU has already run the next
> stores speculatively and has filled the TLB with the read-only
> translations of their pages.  Counting with kprobes, such a run enters
> handle_mm_fault() 135,687 times and do_wp_page() 8,457 times; with an
> LFENCE after every store the faults are 4,100 instead of 65,400 and the
> loop takes 4 ms instead of 29 ms.  The PMD-mapped case, which this patch
> does not touch, shows the same: 8,800 faults for 128 THPs, 133 with the
> LFENCE.
> 
> With a flush_tlb_local() in the new function, as an experiment, the
> ascending loop takes 4.1 ms instead of 28 ms on 64K folios, the
> read-then-store loop 5.3 ms instead of 28 ms and the memset() 21 ms
> instead of 38 ms, for a fault of 870 instead of 730 ns and 11% more time
> for the pass with the clock_gettime().  Generic code has no way to ask
> x86 for a flush that stays on this CPU, so that is left for later.
> 
> Assisted-by: LLM sparse


My review backlog is large enough for me to just go through this wall of text.

There were previous discussions on this, in particular around how much we should
actually try operating around the target PTE.

How did you use the LLM for coming up with this patch + description?

-- 
Cheers,

David

Re: [PATCH] mm/memory: reuse the whole exclusive large folio on a write fault

[Lorenzo Stoakes (ARM)]

This is the friendly patch-bot of Lorenzo Stoakes.

You have sent him a patch/series that has triggered this response.

He used to manually respond to these common problems, but in order to save
his sanity (he kept writing the same thing over and over, yet to different
people), I was created.

Hopefully you will not take offence and will fix the problem in your patch
and resubmit it so that it can be accepted into the Linux kernel tree.

When sending emails to mm:

For one of several possible reasons this mail has triggered an AI
detector script.

Thank you for using the Assisted-by tag to acknowledge this.

However, please do note the policy at
https://docs.kernel.org/process/generated-content.html most notably:

        If tools permit you to generate a contribution automatically, expect
        additional scrutiny in proportion to how much of it was generated.

        As with the output of any tooling, the result may be incorrect or
        inappropriate. You are expected to understand and to be able to
        defend everything you submit. If you are unable to do so, then do
        not submit the resulting changes.

        If you do so anyway, maintainers are entitled to reject your series
        without detailed review.

This change is significant, and you are a relative newcomer to mm. As a
result, I must ask that you provide some details as to how you solved this
problem and your overall understanding of these changes and mm as a whole.

In general, if you are a newcomer to mm, we expect you to do smaller work
before moving on to larger changes, so you build understanding of both the
technical aspects of mm and how we do things.

If you wish to discuss this problem further, or you have questions about
how to resolve this issue, please feel free to respond to this email and
Lorenzo will reply once he has dug out from the pending patches received
from other developers.

thanks,

Lorenzo's patch email bot

[ Idea shamelessly stolen from greg-kh ]

--
Cheers, Lorenzo

Re: [PATCH] mm/memory: reuse the whole exclusive large folio on a write fault

[Yuan-Hao Hsu]

On Fri, 18 Sep 2026 14:14:11 +0200, David Hildenbrand (Arm) wrote:

> There were previous discussions on this, in particular around how much
> we should actually try operating around the target PTE.

Yes, Barry's RFC from 2024 and your replies to it, where you were fine
with contpte granularity and not convinced about anything bigger. That
is why the description has the numbers for both: capped to 16 PTEs, the
1M and 2M cases take 5.4 ms instead of 3.9 ms, and the only pattern where
the cap wins is one store per 2M. If you would rather start with the
contpte-sized version, that is a two-line change and I can send that
version instead.

> How did you use the LLM for coming up with this patch + description?

I did the initial investigation and made the design decisions myself.
Once I had a clear idea of the approach and the overall structure, I used
an LLM to help with parts of the implementation and the test
programs. I also used it to help with drafting the commit message.

Before sending the patch, I went through it line by line and checked the
test results and measurements to make sure they matched what the code was
actually doing and that there were no obvious issues.

Lorenzo, for your bot's question: the reuse decision in
wp_can_reuse_anon_folio() is per folio. The patch applies it to the other
PTEs of the folio within the same VMA and page table using the mprotect
helpers (can_change_pte_writable() and modify_prot_*_ptes()), so
soft-dirty, uffd-wp and NUMA hinting PTEs are handled the way mprotect()
handles them, and the PTE that faulted is finished by wp_page_reuse() as
before.

The part I spent most of the time on was the measurements, including the
cases where it does not help.

Yuan-Hao Hsu

Re: [PATCH] mm/memory: reuse the whole exclusive large folio on a write fault

[Barry Song]

On Sat, Sep 19, 2026 at 2:29 AM Yuan-Hao Hsu <aa9736195201@gmail.com> wrote:
>
> On Fri, 18 Sep 2026 14:14:11 +0200, David Hildenbrand (Arm) wrote:
>
> > There were previous discussions on this, in particular around how much
> > we should actually try operating around the target PTE.
>
> Yes, Barry's RFC from 2024 and your replies to it, where you were fine
> with contpte granularity and not convinced about anything bigger. That
> is why the description has the numbers for both: capped to 16 PTEs, the
> 1M and 2M cases take 5.4 ms instead of 3.9 ms, and the only pattern where

Hi Yuan-Hao,

I'm fine with your follow-up work, but I'd appreciate it if you could
mention the previous work and include the link in your changelog, along
with an explanation of how you address David's concerns.

> the cap wins is one store per 2M. If you would rather start with the
> contpte-sized version, that is a two-line change and I can send that
> version instead.

I think a major concern is that we may spend too much time scanning
PTEs to determine whether we can batch them. If we don't support
CONT-PTE, we may simply scan fewer PTEs; if we do support CONT-PTE, we
may end up scanning more PTEs. So we need to show that the scanning
cost is controlled and explain how it is bounded.

>
> > How did you use the LLM for coming up with this patch + description?
>
> I did the initial investigation and made the design decisions myself.
> Once I had a clear idea of the approach and the overall structure, I used
> an LLM to help with parts of the implementation and the test
> programs. I also used it to help with drafting the commit message.
>
> Before sending the patch, I went through it line by line and checked the
> test results and measurements to make sure they matched what the code was
> actually doing and that there were no obvious issues.

I also personally feel that both the changelog and the code are too
long to read. Could we somehow split the changes into smaller pieces
and describe how each of those concerns is addressed at a finer
granularity?

>
> Lorenzo, for your bot's question: the reuse decision in
> wp_can_reuse_anon_folio() is per folio. The patch applies it to the other
> PTEs of the folio within the same VMA and page table using the mprotect
> helpers (can_change_pte_writable() and modify_prot_*_ptes()), so
> soft-dirty, uffd-wp and NUMA hinting PTEs are handled the way mprotect()
> handles them, and the PTE that faulted is finished by wp_page_reuse() as
> before.
>
> The part I spent most of the time on was the measurements, including the
> cases where it does not help.
>

Best Regards
Barry

Re: [PATCH] mm/memory: reuse the whole exclusive large folio on a write fault

[Yuan-Hao Hsu]

On Sat, 19 Sep 2026 07:48:00 +0800, Barry Song wrote:
> I'm fine with your follow-up work, but I'd appreciate it if you could
> mention the previous work and include the link in your changelog, along
> with an explanation of how you address David's concerns.

Will do in v2.

> I think a major concern is that we may spend too much time scanning
> PTEs to determine whether we can batch them. If we don't support
> CONT-PTE, we may simply scan fewer PTEs; if we do support CONT-PTE, we
> may end up scanning more PTEs. So we need to show that the scanning
> cost is controlled and explain how it is bounded.

The walk stays within the folio, the VMA and the page table, so it
touches at most the PTEs that map this folio in this page table: 16
for a 64K folio, PTRS_PER_PTE (512 with 4K pages) at the most.  Each
PTE is read once by folio_pte_batch_flags(), the same scan fork() and
mprotect() already run over these PTEs.

Measured, that is 14-20 ns per PTE.  For 512 PTEs the fault takes
10 us, 1 us of it the scan, the rest marking the pages exclusive and
writing the PTEs.  Those are PTEs that would otherwise each take a
420 ns fault.  The pattern where the work is wasted, one store per
folio, is the one the description gives the numbers for.

> I also personally feel that both the changelog and the code are too
> long to read. Could we somehow split the changes into smaller pieces
> and describe how each of those concerns is addressed at a finer
> granularity?

Yes, and sorry for the wall of text.  v2 is two patches: 1/2 handles
an aligned block of 16 PTEs, the contpte-sized version David was fine
with in 2024; 2/2 lifts that to the folio.  Each has its own numbers,
the earlier discussion is linked, and the changelogs are cut to what
is needed to judge the change.

[PATCH v2 0/2] mm/memory: reuse the whole exclusive large folio on a write fault

[Yuan-Hao Hsu]

After fork() and the child's exit, the parent's write faults reuse a
large anonymous folio one PTE at a time, although
wp_can_reuse_anon_folio() has already found the whole folio exclusive.

v2, after the comments on v1 [1]:
- Split in two.  1/2 handles an aligned block of 16 PTEs around the
  fault, the contpte-sized version David was fine with in the 2024
  discussion of Barry's RFC [2].  2/2 lifts that to the folio.
- The earlier discussion is linked; the changelogs say how its two
  reservations, the latency of the individual fault and how far to go
  around it, are answered.
- The bound of the walk is spelled out (Barry): folio, VMA and block or
  page table, each PTE read once by folio_pte_batch_flags().
- Changelogs cut to what is needed to judge the change.
- Same base as v1; 1/2 + 2/2 is the code of v1.

Two points from the AI review of v1, both done as mprotect() does them:
change_pte_range() does not flush_cache_range() before making PTEs
writable, and it makes clean exclusive anonymous PTEs writable without
pte_mkdirty() (the dirty rule is for shared file mappings, see
can_change_shared_pte_writable()).

Controls, unchanged: order-0 pages, PMD-mapped THPs, the COW copy path
and the order-0 and cow/fork/write-fault modes of David's
pte-mapped-folio-benchmarks.

What does not get faster on x86: stores to pages this CPU still holds
a read-only TLB entry for.  The fault makes the PTEs writable but, like
mprotect(), does not flush, so such a page takes one spurious fault,
about the cost of the reuse fault it replaces.  That is the case for
pages read since fork(), and for a loop that only stores one byte per
page in ascending order (David's reuse-byte mode): the CPU runs the
next stores speculatively while the first one faults and caches their
read-only translations.  Shown with kprobes (135,687 handle_mm_fault()
for 8,457 do_wp_page()) and an LFENCE after every store (4,100 faults
instead of 65,400); the untouched PMD-mapped case behaves the same.  A
flush_tlb_local() in the helper would fix it (that loop 28 -> 4 ms,
memset() 38 -> 21 ms, +140 ns per fault), but generic code has no way
to ask x86 for a flush that stays on this CPU, so that is for later.

Tested with DEBUG_VM, DEBUG_VM_PGTABLE, PROVE_LOCKING and
PAGE_TABLE_CHECK_ENFORCED: the mm selftests, a 12-scenario COW test
(child alive, vmsplice, PROT_READ VMA inside the folio, soft-dirty and
uffd-wp counts, mremap, holes, pageout, FOLL_FORCE), a
fork/pageout/mprotect/vmsplice stress, NUMA balancing on numa=fake=2
(protnone PTEs left alone), and arm64 under QEMU for the counters.
Cross-built for arm64 4K/16K/64K, i386 with and without PAE, x86
without THP, arm, arm nommu, riscv64, powerpc64le and s390x.

[1] https://lore.kernel.org/r/20260918064238.868-1-aa9736195201@gmail.com
[2] https://lore.kernel.org/r/20240831092339.66085-1-21cnbao@gmail.com

Yuan-Hao Hsu (2):
  mm/memory: reuse 16 PTEs of an exclusive large folio on a write fault
  mm/memory: reuse the whole exclusive large folio on a write fault

 mm/memory.c | 68 +++++++++++++++++++++++++++++++++++++++++++++++++++--
 1 file changed, 66 insertions(+), 2 deletions(-)


base-commit: 238650ef6c7c7cca08e032527329424c9fbd70e5
-- 
2.43.0

[PATCH v2 1/2] mm/memory: reuse 16 PTEs of an exclusive large folio on a write fault

[Yuan-Hao Hsu]

fork() maps the anonymous pages of the parent read-only and clears
PageAnonExclusive on them.  Once the child has exec'ed or exited, the
parent's write fault takes the reuse path of do_wp_page():
wp_can_reuse_anon_folio() finds that all references to the folio come
from this MM, the page is marked exclusive again and its PTE is made
writable.

For a large folio that check is about the folio and holds for every
page of it, but only the page that faulted is marked exclusive and
made writable.  Each other page takes a write fault of its own and
takes the large mapcount lock to find out the same thing again: 16
faults for a 64K folio.  The same THP mapped by a PMD is reused by one
fault in do_huge_pmd_wp_page(), and do_swap_page() maps all PTEs of an
exclusive large folio writable at once.

Barry proposed reusing the whole mTHP from one fault in 2024 [1].  The
reservations then were the latency of the individual write fault and
how far to go around the faulting PTE: a contpte-sized block was fine,
anything bigger not yet convincing (David's replies, linked below).
Commit 1da190f4d0a6 ("mm: Copy-on-Write (COW) reuse support for
PTE-mapped THP") then added the per-folio check and left faulting
around for later.

This is the contpte-sized part.  Once the folio is known to be
exclusive, walk the aligned block of 16 PTEs around the fault, within
the folio and the VMA.  PTEs that still map the folio read-only are
batched with folio_pte_batch_flags(), their pages are marked
exclusive, and where can_change_pte_writable() agrees the batch is
made writable with modify_prot_start_ptes()/modify_prot_commit_ptes(),
as mprotect() does it.  That leaves NUMA hinting and uffd-wp PTEs
alone, keeps soft-dirty tracking exact, and on arm64 writes a contpte
block back as a whole where ptep_set_access_flags() on one PTE has to
unfold it.  Pages that cannot be made writable are still marked
exclusive, so their own fault skips the folio check.  The PTE that
faulted is in one of the batches and is completed by wp_page_reuse()
as before.  Small folios, PMD-mapped THPs, unsharing faults and the
copy path are unchanged.

The cost is bounded by the block: at most 16 PTEs read once by
folio_pte_batch_flags(), the scan fork() and mprotect() already run
over them, 16 pages marked exclusive and 16 PTEs written.  On x86-64
(i7-12700KF; medians of 15 runs, two boots of each kernel taken
alternately) the fault that does that for a 64K folio takes 750 ns,
against 420-440 ns for a reuse fault today and 4,600-5,000 ns for the
fault that allocated the folio.

256 MiB of 64K folios after fork() and the child's exit, faults and
time of the pass:

                            v7.3-rc3+             patched
  one byte per page, seq    65,601  30.8/30.5 ms   4,164   5.7/ 5.7 ms
  one byte per page, random 65,601  39.5/41.6 ms   4,161   7.4/ 7.6 ms
  memset()                  65,541  57.4/59.5 ms   4,102  39.5/39.8 ms
  8 threads, random order   65,541   6.1/ 6.2 ms   4,155   0.9/ 0.9 ms
  one store per 64K folio    4,101   1.9/ 2.0 ms   4,101   3.5/ 3.3 ms

Redis 7.0.15, 1.3 M keys of 512 bytes on 64K mTHP, BGSAVE and then
1,000,000 SETs: the faults of redis-server during the SETs go from
262,100 to 18,550 and its CPU time from 1.48-1.53 s to 1.31-1.41 s;
the requests per second stay within the boot-to-boot spread.  With
THP off nothing changes.

arm64, under QEMU for the counters: the first pass after fork()
unfolds every contpte block of the 64K folios today (512
contpte_convert() calls for 512 blocks, and nothing folds them again);
with this patch none is unfolded.

[1] https://lore.kernel.org/r/20240831092339.66085-1-21cnbao@gmail.com

Link: https://lore.kernel.org/all/20240831092339.66085-1-21cnbao@gmail.com/
Link: https://lore.kernel.org/all/b7853f0f-7044-4c49-931c-c61900229b19@redhat.com/
Link: https://lore.kernel.org/all/36933711-ae0f-468c-93bd-d6a67d974c9d@redhat.com/
Assisted-by: LLM sparse
Signed-off-by: Yuan-Hao Hsu <aa9736195201@gmail.com>
---
 mm/memory.c | 75 +++++++++++++++++++++++++++++++++++++++++++++++++++--
 1 file changed, 73 insertions(+), 2 deletions(-)

diff --git a/mm/memory.c b/mm/memory.c
index 8b0c2c735d3d..73e5691b3ed8 100644
--- a/mm/memory.c
+++ b/mm/memory.c
@@ -4355,6 +4355,71 @@ static bool wp_can_reuse_anon_folio(struct folio *folio,
 	return true;
 }
 
+/*
+ * The pages of the folio around the one that faulted are handled in aligned
+ * blocks of this many PTEs: a 64K folio with 4K pages, and the size of a
+ * contpte block on arm64.
+ */
+#define WP_REUSE_NR_PTES	16
+
+/*
+ * wp_can_reuse_anon_folio() found a large folio to be exclusive to this MM.
+ * That holds for all of its pages and not only for the one that faulted: mark
+ * the ones in the same block exclusive as well and map them writable, like
+ * mprotect() would. Each of them would otherwise take a write fault of its own
+ * that repeats the check on the very same folio.
+ *
+ * The PTE that faulted is among them; wp_page_reuse() completes it.
+ */
+static void wp_reuse_large_anon_folio(struct vm_fault *vmf,
+		struct folio *folio)
+{
+	const fpb_t flags = FPB_RESPECT_WRITE | FPB_RESPECT_SOFT_DIRTY;
+	const unsigned long idx = folio_page_idx(folio, vmf->page);
+	struct vm_area_struct *vma = vmf->vma;
+	unsigned long addr = vmf->address;
+	unsigned long block = ALIGN_DOWN(addr, WP_REUSE_NR_PTES * PAGE_SIZE);
+	unsigned long nr_before, nr_after, end;
+	struct page *page;
+	unsigned int nr, i;
+	pte_t *ptep, pte;
+
+	/* Stay within the folio, the VMA and the block. */
+	nr_before = min3(idx, (addr - block) >> PAGE_SHIFT,
+			 (addr - vma->vm_start) >> PAGE_SHIFT);
+	nr_after = min3(folio_nr_pages(folio) - idx,
+			(block + WP_REUSE_NR_PTES * PAGE_SIZE - addr) >> PAGE_SHIFT,
+			(vma->vm_end - addr) >> PAGE_SHIFT);
+	end = addr + (nr_after << PAGE_SHIFT);
+	addr -= nr_before << PAGE_SHIFT;
+	ptep = vmf->pte - nr_before;
+	page = vmf->page - nr_before;
+
+	for (; addr != end; addr += nr * PAGE_SIZE, ptep += nr, page += nr) {
+		pte = ptep_get(ptep);
+		nr = 1;
+
+		/* Unmapped or replaced since, or writable already. */
+		if (!pte_present(pte) || pte_pfn(pte) != page_to_pfn(page) ||
+		    pte_write(pte))
+			continue;
+
+		nr = folio_pte_batch_flags(folio, NULL, ptep, &pte,
+					   (end - addr) >> PAGE_SHIFT, flags);
+		for (i = 0; i < nr; i++)
+			if (!PageAnonExclusive(page + i))
+				SetPageAnonExclusive(page + i);
+
+		/* The PTEs of a batch agree on everything this looks at. */
+		if (!can_change_pte_writable(vma, addr, pte))
+			continue;
+
+		pte = modify_prot_start_ptes(vma, addr, ptep, nr);
+		modify_prot_commit_ptes(vma, addr, ptep, pte,
+					pte_mkwrite(pte, vma), nr);
+	}
+}
+
 /*
  * This routine handles present pages, when
  * * users try to write to a shared page (FAULT_FLAG_WRITE)
@@ -4449,8 +4514,14 @@ static vm_fault_t do_wp_page(struct vm_fault *vmf)
 	 */
 	if (folio && folio_test_anon(folio) &&
 	    (PageAnonExclusive(vmf->page) || wp_can_reuse_anon_folio(folio, vma))) {
-		if (!PageAnonExclusive(vmf->page))
-			SetPageAnonExclusive(vmf->page);
+		if (!PageAnonExclusive(vmf->page)) {
+			if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) &&
+			    folio_test_large(folio) && likely(!unshare) &&
+			    likely(vma->vm_flags & VM_WRITE))
+				wp_reuse_large_anon_folio(vmf, folio);
+			else
+				SetPageAnonExclusive(vmf->page);
+		}
 		if (unlikely(unshare)) {
 			pte_unmap_unlock(vmf->pte, vmf->ptl);
 			return 0;
-- 
2.43.0

[PATCH v2 2/2] mm/memory: reuse the whole exclusive large folio on a write fault

[Yuan-Hao Hsu]

With blocks of 16 PTEs a PTE-mapped 2M THP still takes 32 reuse faults
and a 1M folio 16.  Lift the bound to the page table: the walk then
covers the PTEs that map the folio in this page table, at most
PTRS_PER_PTE of them, and one fault does the work for the folio.

Its cost, measured as the time of the store that takes it, against
420 ns for a reuse fault today (x86-64, i7-12700KF, medians, ns):

                          reuse fault   fault that     COW fault that
                          (patched)     allocated it   copies 4K
      1M mTHP            5,400- 5,700   63,000-68,000          1,500
      2M THP, PTE-mapped  10,000-10,500         126,000          1,500

That is 14-20 ns per PTE, about 2 ns of it the scan.  Builds that
differ only by NOPs in front of the function take either 10,100 or
7,500 ns for the 2M folio, with a period of 32 bytes: the loop of
modify_prot_commit_ptes() changes speed with its address.

What it buys, 256 MiB after fork() and the child's exit, medians of 15
runs, two boots of each kernel:

                                   16 PTEs         whole folio
  one byte per page, seq
      1M mTHP                     5.7 /  5.8 ms    4.1 /  4.0 ms
      2M THP, PTE-mapped          5.7 /  5.8 ms    3.9 /  3.9 ms
  one byte per page, random order
      2M THP, PTE-mapped          7.4 /  7.5 ms    4.9 /  5.1 ms
  memset()
      2M THP, PTE-mapped         40.2 / 39.4 ms   38.1 / 37.2 ms
  8 threads, random order
      2M THP, PTE-mapped          0.9 /  1.0 ms    0.7 /  0.7 ms
  one store per 64K, 2M folios    3.3 /  3.3 ms    1.5 /  1.6 ms
  one store per 2M, 2M folios     0.1 /  0.1 ms    1.4 /  1.4 ms

The last row is the pattern where the whole-folio fault is pure cost:
511 pages made writable that nobody writes, 1.3 ms more per 256 MiB,
paid once per fork().  Anything that stores to more than one page per
64K comes out ahead.

Link: https://lore.kernel.org/all/36933711-ae0f-468c-93bd-d6a67d974c9d@redhat.com/
Assisted-by: LLM sparse
Signed-off-by: Yuan-Hao Hsu <aa9736195201@gmail.com>
---
 mm/memory.c | 21 +++++++--------------
 1 file changed, 7 insertions(+), 14 deletions(-)

diff --git a/mm/memory.c b/mm/memory.c
index 73e5691b3ed8..85c883d1e558 100644
--- a/mm/memory.c
+++ b/mm/memory.c
@@ -4355,19 +4355,12 @@ static bool wp_can_reuse_anon_folio(struct folio *folio,
 	return true;
 }
 
-/*
- * The pages of the folio around the one that faulted are handled in aligned
- * blocks of this many PTEs: a 64K folio with 4K pages, and the size of a
- * contpte block on arm64.
- */
-#define WP_REUSE_NR_PTES	16
-
 /*
  * wp_can_reuse_anon_folio() found a large folio to be exclusive to this MM.
  * That holds for all of its pages and not only for the one that faulted: mark
- * the ones in the same block exclusive as well and map them writable, like
- * mprotect() would. Each of them would otherwise take a write fault of its own
- * that repeats the check on the very same folio.
+ * the ones that this page table maps exclusive as well and map them writable,
+ * like mprotect() would. Each of them would otherwise take a write fault of
+ * its own that repeats the check on the very same folio.
  *
  * The PTE that faulted is among them; wp_page_reuse() completes it.
  */
@@ -4378,17 +4371,17 @@ static void wp_reuse_large_anon_folio(struct vm_fault *vmf,
 	const unsigned long idx = folio_page_idx(folio, vmf->page);
 	struct vm_area_struct *vma = vmf->vma;
 	unsigned long addr = vmf->address;
-	unsigned long block = ALIGN_DOWN(addr, WP_REUSE_NR_PTES * PAGE_SIZE);
+	unsigned long pt_start = ALIGN_DOWN(addr, PMD_SIZE);
 	unsigned long nr_before, nr_after, end;
 	struct page *page;
 	unsigned int nr, i;
 	pte_t *ptep, pte;
 
-	/* Stay within the folio, the VMA and the block. */
-	nr_before = min3(idx, (addr - block) >> PAGE_SHIFT,
+	/* Stay within the folio, the VMA and the page table. */
+	nr_before = min3(idx, (addr - pt_start) >> PAGE_SHIFT,
 			 (addr - vma->vm_start) >> PAGE_SHIFT);
 	nr_after = min3(folio_nr_pages(folio) - idx,
-			(block + WP_REUSE_NR_PTES * PAGE_SIZE - addr) >> PAGE_SHIFT,
+			(pt_start + PMD_SIZE - addr) >> PAGE_SHIFT,
 			(vma->vm_end - addr) >> PAGE_SHIFT);
 	end = addr + (nr_after << PAGE_SHIFT);
 	addr -= nr_before << PAGE_SHIFT;
-- 
2.43.0

Re: [PATCH v2 0/2] mm/memory: reuse the whole exclusive large folio on a write fault

[David Hildenbrand (Arm)]

On 9/19/26 09:31, Yuan-Hao Hsu wrote:
> After fork() and the child's exit, the parent's write faults reuse a
> large anonymous folio one PTE at a time, although
> wp_can_reuse_anon_folio() has already found the whole folio exclusive.
> 
> v2, after the comments on v1 [1]:

Please don't send out a new version

* Unless a maintainer asked you to
* At least one week passed

-- 
Cheers,

David

Re: [PATCH v2 2/2] mm/memory: reuse the whole exclusive large folio on a write fault

[David Hildenbrand (Arm)]

On 9/19/26 09:31, Yuan-Hao Hsu wrote:
> With blocks of 16 PTEs a PTE-mapped 2M THP still takes 32 reuse faults
> and a 1M folio 16.  Lift the bound to the page table: the walk then
> covers the PTEs that map the folio in this page table, at most
> PTRS_PER_PTE of them, and one fault does the work for the folio.
> 
> Its cost, measured as the time of the store that takes it, against
> 420 ns for a reuse fault today (x86-64, i7-12700KF, medians, ns):
> 
>                           reuse fault   fault that     COW fault that
>                           (patched)     allocated it   copies 4K
>       1M mTHP            5,400- 5,700   63,000-68,000          1,500
>       2M THP, PTE-mapped  10,000-10,500         126,000          1,500
> 
> That is 14-20 ns per PTE, about 2 ns of it the scan.  Builds that
> differ only by NOPs in front of the function take either 10,100 or
> 7,500 ns for the 2M folio, with a period of 32 bytes: the loop of
> modify_prot_commit_ptes() changes speed with its address.
> 
> What it buys, 256 MiB after fork() and the child's exit, medians of 15
> runs, two boots of each kernel:
> 
>                                    16 PTEs         whole folio
>   one byte per page, seq
>       1M mTHP                     5.7 /  5.8 ms    4.1 /  4.0 ms
>       2M THP, PTE-mapped          5.7 /  5.8 ms    3.9 /  3.9 ms
>   one byte per page, random order
>       2M THP, PTE-mapped          7.4 /  7.5 ms    4.9 /  5.1 ms
>   memset()
>       2M THP, PTE-mapped         40.2 / 39.4 ms   38.1 / 37.2 ms
>   8 threads, random order
>       2M THP, PTE-mapped          0.9 /  1.0 ms    0.7 /  0.7 ms
>   one store per 64K, 2M folios    3.3 /  3.3 ms    1.5 /  1.6 ms
>   one store per 2M, 2M folios     0.1 /  0.1 ms    1.4 /  1.4 ms
> 

Are you measuring the worst-case latency of a single write fault or only the
average over all writes?

-- 
Cheers,

David

Re: [PATCH v2 0/2] mm/memory: reuse the whole exclusive large folio on a write fault

[Yuan-Hao Hsu]

On Sat, 19 Sep 2026 12:08:35 +0200, David Hildenbrand (Arm) wrote:
> Please don't send out a new version
>
> * Unless a maintainer asked you to
> * At least one week passed

Understood, sorry about that.  I'll sit tight and wait for the review
of this version.

Re: [PATCH v2 2/2] mm/memory: reuse the whole exclusive large folio on a write fault

[Yuan-Hao Hsu]

On Sat, 19 Sep 2026 12:10:56 +0200, David Hildenbrand (Arm) wrote:
> Are you measuring the worst-case latency of a single write fault or only the
> average over all writes?

The single fault.  Each store is timed by itself (clock_gettime()
before and after), and the latency numbers come from a pattern where
every store takes one of these faults (one store per folio), so there
is nothing else in the average.  The tables show the median of those
faults; I also have p99 and max (i7-12700KF, in a VM, us):

                     v7.3-rc3+              patched
                     p50   p99   max        p50        p99        max
  64K, 1 fault      0.42  0.57  25-26      0.71-0.74  0.96-1.10  28-60
  2M, 1 fault       0.50  0.67  0.7-1.2    10.0-10.2  12.2-15.8  13-22

max is the slowest store of a run, median over 15 runs.  The 25 us on
the base kernel are interrupts; the VM has those on both sides.  The
2M row is the worst case by construction, every one of those faults
walks all 512 PTEs.