Re: [PATCH 1/2] mm: move mm counter updating out of set_pte_range()

Re: [PATCH 1/2] mm: move mm counter updating out of set_pte_range()

[Matthew Wilcox]

On Thu, Apr 11, 2024 at 09:09:49PM +0800, Kefeng Wang wrote:
> In order to support batch mm counter updating in filemap_map_pages(),
> make set_pte_range() return the type of MM_COUNTERS and move mm counter
> updating out of set_pte_range().

I don't like this.  You're making set_pte_range() harder to use.
It's also rather overengineered; if you're calling set_pte_range()
from filemap.c, you already know the folios are MM_FILEPAGES.

[PATCH 1/2] mm: move mm counter updating out of set_pte_range()

[Kefeng Wang]

In order to support batch mm counter updating in filemap_map_pages(),
make set_pte_range() return the type of MM_COUNTERS and move mm counter
updating out of set_pte_range().

Signed-off-by: Kefeng Wang <wangkefeng.wang@huawei.com>
---
 include/linux/mm.h |  4 ++--
 mm/filemap.c       | 10 +++++++---
 mm/memory.c        | 16 +++++++++++-----
 3 files changed, 20 insertions(+), 10 deletions(-)

diff --git a/include/linux/mm.h b/include/linux/mm.h
index 0b4046b1e63d..6ad440ac3706 100644
--- a/include/linux/mm.h
+++ b/include/linux/mm.h
@@ -1366,8 +1366,8 @@ static inline pte_t maybe_mkwrite(pte_t pte, struct vm_area_struct *vma)
 }
 
 vm_fault_t do_set_pmd(struct vm_fault *vmf, struct page *page);
-void set_pte_range(struct vm_fault *vmf, struct folio *folio,
-		struct page *page, unsigned int nr, unsigned long addr);
+int set_pte_range(struct vm_fault *vmf, struct folio *folio,
+		  struct page *page, unsigned int nr, unsigned long addr);
 
 vm_fault_t finish_fault(struct vm_fault *vmf);
 #endif
diff --git a/mm/filemap.c b/mm/filemap.c
index 92e2d43e4c9d..2274e590bab4 100644
--- a/mm/filemap.c
+++ b/mm/filemap.c
@@ -3512,6 +3512,7 @@ static vm_fault_t filemap_map_folio_range(struct vm_fault *vmf,
 	struct page *page = folio_page(folio, start);
 	unsigned int count = 0;
 	pte_t *old_ptep = vmf->pte;
+	int type;
 
 	do {
 		if (PageHWPoison(page + count))
@@ -3539,7 +3540,8 @@ static vm_fault_t filemap_map_folio_range(struct vm_fault *vmf,
 		continue;
 skip:
 		if (count) {
-			set_pte_range(vmf, folio, page, count, addr);
+			type = set_pte_range(vmf, folio, page, count, addr);
+			add_mm_counter(vmf->vma->vm_mm, type, count);
 			folio_ref_add(folio, count);
 			if (in_range(vmf->address, addr, count * PAGE_SIZE))
 				ret = VM_FAULT_NOPAGE;
@@ -3553,7 +3555,8 @@ static vm_fault_t filemap_map_folio_range(struct vm_fault *vmf,
 	} while (--nr_pages > 0);
 
 	if (count) {
-		set_pte_range(vmf, folio, page, count, addr);
+		type = set_pte_range(vmf, folio, page, count, addr);
+		add_mm_counter(vmf->vma->vm_mm, type, count);
 		folio_ref_add(folio, count);
 		if (in_range(vmf->address, addr, count * PAGE_SIZE))
 			ret = VM_FAULT_NOPAGE;
@@ -3589,7 +3592,8 @@ static vm_fault_t filemap_map_order0_folio(struct vm_fault *vmf,
 	if (vmf->address == addr)
 		ret = VM_FAULT_NOPAGE;
 
-	set_pte_range(vmf, folio, page, 1, addr);
+	add_mm_counter(vmf->vma->vm_mm,
+		       set_pte_range(vmf, folio, page, 1, addr), 1);
 	folio_ref_inc(folio);
 
 	return ret;
diff --git a/mm/memory.c b/mm/memory.c
index 78422d1c7381..485ffec9d4c7 100644
--- a/mm/memory.c
+++ b/mm/memory.c
@@ -4661,15 +4661,18 @@ vm_fault_t do_set_pmd(struct vm_fault *vmf, struct page *page)
  * @page: The first page to create a PTE for.
  * @nr: The number of PTEs to create.
  * @addr: The first address to create a PTE for.
+ *
+ * Return: type of MM_COUNTERS to be updated
  */
-void set_pte_range(struct vm_fault *vmf, struct folio *folio,
-		struct page *page, unsigned int nr, unsigned long addr)
+int set_pte_range(struct vm_fault *vmf, struct folio *folio,
+		  struct page *page, unsigned int nr, unsigned long addr)
 {
 	struct vm_area_struct *vma = vmf->vma;
 	bool uffd_wp = vmf_orig_pte_uffd_wp(vmf);
 	bool write = vmf->flags & FAULT_FLAG_WRITE;
 	bool prefault = in_range(vmf->address, addr, nr * PAGE_SIZE);
 	pte_t entry;
+	int type;
 
 	flush_icache_pages(vma, page, nr);
 	entry = mk_pte(page, vma->vm_page_prot);
@@ -4685,18 +4688,20 @@ void set_pte_range(struct vm_fault *vmf, struct folio *folio,
 		entry = pte_mkuffd_wp(entry);
 	/* copy-on-write page */
 	if (write && !(vma->vm_flags & VM_SHARED)) {
-		add_mm_counter(vma->vm_mm, MM_ANONPAGES, nr);
+		type = MM_ANONPAGES;
 		VM_BUG_ON_FOLIO(nr != 1, folio);
 		folio_add_new_anon_rmap(folio, vma, addr);
 		folio_add_lru_vma(folio, vma);
 	} else {
-		add_mm_counter(vma->vm_mm, mm_counter_file(folio), nr);
+		type = mm_counter_file(folio);
 		folio_add_file_rmap_ptes(folio, page, nr, vma);
 	}
 	set_ptes(vma->vm_mm, addr, vmf->pte, entry, nr);
 
 	/* no need to invalidate: a not-present page won't be cached */
 	update_mmu_cache_range(vmf, vma, addr, vmf->pte, nr);
+
+	return type;
 }
 
 static bool vmf_pte_changed(struct vm_fault *vmf)
@@ -4765,8 +4770,9 @@ vm_fault_t finish_fault(struct vm_fault *vmf)
 	/* Re-check under ptl */
 	if (likely(!vmf_pte_changed(vmf))) {
 		struct folio *folio = page_folio(page);
+		int type = set_pte_range(vmf, folio, page, 1, vmf->address);
 
-		set_pte_range(vmf, folio, page, 1, vmf->address);
+		add_mm_counter(vmf->vma->vm_mm, type, 1);
 		ret = 0;
 	} else {
 		update_mmu_tlb(vma, vmf->address, vmf->pte);
-- 
2.41.0

Re: [PATCH 1/2] mm: move mm counter updating out of set_pte_range()

[Kefeng Wang]

On 2024/4/11 20:53, Matthew Wilcox wrote:
> On Thu, Apr 11, 2024 at 09:09:49PM +0800, Kefeng Wang wrote:
>> In order to support batch mm counter updating in filemap_map_pages(),
>> make set_pte_range() return the type of MM_COUNTERS and move mm counter
>> updating out of set_pte_range().
> 
> I don't like this.  You're making set_pte_range() harder to use.
> It's also rather overengineered; if you're calling set_pte_range()
> from filemap.c, you already know the folios are MM_FILEPAGES.

or MM_SHMEMS,  and another caller finish_fault(), which already check
vmf->flags and vma->vm_flags, we could use it to distinguish anon or 
file, I will try this way, thanks.

> 
> 

Re: [PATCH 1/2] mm: move mm counter updating out of set_pte_range()

[Kefeng Wang]

On 2024/4/12 10:57, Kefeng Wang wrote:
> In order to support batch mm counter updating in filemap_map_pages(),
> move mm counter updating out of set_pte_range(), the folios are file
> from filemap, and distinguish folios type by vmf->flags and vma->vm_flags
> from another caller finish_fault().
> 
> Signed-off-by: Kefeng Wang <wangkefeng.wang@huawei.com>
> ---
>   mm/filemap.c | 4 ++++
>   mm/memory.c  | 8 +++++---
>   2 files changed, 9 insertions(+), 3 deletions(-)
> 
> diff --git a/mm/filemap.c b/mm/filemap.c
> index 92e2d43e4c9d..04b813f0146c 100644
> --- a/mm/filemap.c
> +++ b/mm/filemap.c
> @@ -3540,6 +3540,8 @@ static vm_fault_t filemap_map_folio_range(struct vm_fault *vmf,
>   skip:
>   		if (count) {
>   			set_pte_range(vmf, folio, page, count, addr);
> +			add_mm_counter(vmf->vma->vm_mm, mm_counter_file(folio),
> +				       count);
>   			folio_ref_add(folio, count);
>   			if (in_range(vmf->address, addr, count * PAGE_SIZE))
>   				ret = VM_FAULT_NOPAGE;
> @@ -3554,6 +3556,7 @@ static vm_fault_t filemap_map_folio_range(struct vm_fault *vmf,
>   
>   	if (count) {
>   		set_pte_range(vmf, folio, page, count, addr);
> +		add_mm_counter(vmf->vma->vm_mm, mm_counter_file(folio), count);
>   		folio_ref_add(folio, count);
>   		if (in_range(vmf->address, addr, count * PAGE_SIZE))
>   			ret = VM_FAULT_NOPAGE;
> @@ -3590,6 +3593,7 @@ static vm_fault_t filemap_map_order0_folio(struct vm_fault *vmf,
>   		ret = VM_FAULT_NOPAGE;
>   
>   	set_pte_range(vmf, folio, page, 1, addr);
> +	add_mm_counter(vmf->vma->vm_mm, mm_counter_file(folio), 1);
>   	folio_ref_inc(folio);
>   
>   	return ret;
> diff --git a/mm/memory.c b/mm/memory.c
> index 78422d1c7381..69bc63a5d6c8 100644
> --- a/mm/memory.c
> +++ b/mm/memory.c
> @@ -4685,12 +4685,10 @@ void set_pte_range(struct vm_fault *vmf, struct folio *folio,
>   		entry = pte_mkuffd_wp(entry);
>   	/* copy-on-write page */
>   	if (write && !(vma->vm_flags & VM_SHARED)) {
> -		add_mm_counter(vma->vm_mm, MM_ANONPAGES, nr);
>   		VM_BUG_ON_FOLIO(nr != 1, folio);
>   		folio_add_new_anon_rmap(folio, vma, addr);
>   		folio_add_lru_vma(folio, vma);
>   	} else {
> -		add_mm_counter(vma->vm_mm, mm_counter_file(folio), nr);
>   		folio_add_file_rmap_ptes(folio, page, nr, vma);
>   	}
>   	set_ptes(vma->vm_mm, addr, vmf->pte, entry, nr);
> @@ -4727,9 +4725,11 @@ vm_fault_t finish_fault(struct vm_fault *vmf)
>   	struct vm_area_struct *vma = vmf->vma;
>   	struct page *page;
>   	vm_fault_t ret;
> +	int is_cow = (vmf->flags & FAULT_FLAG_WRITE) &&
> +		     !(vma->vm_flags & VM_SHARED);

oops, bool is enough.

>   
>   	/* Did we COW the page? */
> -	if ((vmf->flags & FAULT_FLAG_WRITE) && !(vma->vm_flags & VM_SHARED))
> +	if (is_cow)
>   		page = vmf->cow_page;
>   	else
>   		page = vmf->page;
> @@ -4765,8 +4765,10 @@ vm_fault_t finish_fault(struct vm_fault *vmf)
>   	/* Re-check under ptl */
>   	if (likely(!vmf_pte_changed(vmf))) {
>   		struct folio *folio = page_folio(page);
> +		int type = is_cow ? MM_ANONPAGES : mm_counter_file(folio);
>   
>   		set_pte_range(vmf, folio, page, 1, vmf->address);
> +		add_mm_counter(vma->vm_mm, type, 1);
>   		ret = 0;
>   	} else {
>   		update_mmu_tlb(vma, vmf->address, vmf->pte);

[PATCH v2 0/2] mm: batch mm counter updating in filemap_map_pages()

[Kefeng Wang]

Let's batch mm counter updating to accelerate filemap_map_pages(). 

v2:
- estimate folio type from caller and no need to return from
  set_pte_range()
- use unsigned long for rss

Kefeng Wang (2):
  mm: move mm counter updating out of set_pte_range()
  mm: filemap: batch mm counter updating in filemap_map_pages()

 mm/filemap.c | 14 ++++++++++----
 mm/memory.c  |  8 +++++---
 2 files changed, 15 insertions(+), 7 deletions(-)

-- 
2.41.0

[PATCH 1/2] mm: move mm counter updating out of set_pte_range()

[Kefeng Wang]

In order to support batch mm counter updating in filemap_map_pages(),
move mm counter updating out of set_pte_range(), the folios are file
from filemap, and distinguish folios type by vmf->flags and vma->vm_flags
from another caller finish_fault().

Signed-off-by: Kefeng Wang <wangkefeng.wang@huawei.com>
---
 mm/filemap.c | 4 ++++
 mm/memory.c  | 8 +++++---
 2 files changed, 9 insertions(+), 3 deletions(-)

diff --git a/mm/filemap.c b/mm/filemap.c
index 92e2d43e4c9d..04b813f0146c 100644
--- a/mm/filemap.c
+++ b/mm/filemap.c
@@ -3540,6 +3540,8 @@ static vm_fault_t filemap_map_folio_range(struct vm_fault *vmf,
 skip:
 		if (count) {
 			set_pte_range(vmf, folio, page, count, addr);
+			add_mm_counter(vmf->vma->vm_mm, mm_counter_file(folio),
+				       count);
 			folio_ref_add(folio, count);
 			if (in_range(vmf->address, addr, count * PAGE_SIZE))
 				ret = VM_FAULT_NOPAGE;
@@ -3554,6 +3556,7 @@ static vm_fault_t filemap_map_folio_range(struct vm_fault *vmf,
 
 	if (count) {
 		set_pte_range(vmf, folio, page, count, addr);
+		add_mm_counter(vmf->vma->vm_mm, mm_counter_file(folio), count);
 		folio_ref_add(folio, count);
 		if (in_range(vmf->address, addr, count * PAGE_SIZE))
 			ret = VM_FAULT_NOPAGE;
@@ -3590,6 +3593,7 @@ static vm_fault_t filemap_map_order0_folio(struct vm_fault *vmf,
 		ret = VM_FAULT_NOPAGE;
 
 	set_pte_range(vmf, folio, page, 1, addr);
+	add_mm_counter(vmf->vma->vm_mm, mm_counter_file(folio), 1);
 	folio_ref_inc(folio);
 
 	return ret;
diff --git a/mm/memory.c b/mm/memory.c
index 78422d1c7381..69bc63a5d6c8 100644
--- a/mm/memory.c
+++ b/mm/memory.c
@@ -4685,12 +4685,10 @@ void set_pte_range(struct vm_fault *vmf, struct folio *folio,
 		entry = pte_mkuffd_wp(entry);
 	/* copy-on-write page */
 	if (write && !(vma->vm_flags & VM_SHARED)) {
-		add_mm_counter(vma->vm_mm, MM_ANONPAGES, nr);
 		VM_BUG_ON_FOLIO(nr != 1, folio);
 		folio_add_new_anon_rmap(folio, vma, addr);
 		folio_add_lru_vma(folio, vma);
 	} else {
-		add_mm_counter(vma->vm_mm, mm_counter_file(folio), nr);
 		folio_add_file_rmap_ptes(folio, page, nr, vma);
 	}
 	set_ptes(vma->vm_mm, addr, vmf->pte, entry, nr);
@@ -4727,9 +4725,11 @@ vm_fault_t finish_fault(struct vm_fault *vmf)
 	struct vm_area_struct *vma = vmf->vma;
 	struct page *page;
 	vm_fault_t ret;
+	int is_cow = (vmf->flags & FAULT_FLAG_WRITE) &&
+		     !(vma->vm_flags & VM_SHARED);
 
 	/* Did we COW the page? */
-	if ((vmf->flags & FAULT_FLAG_WRITE) && !(vma->vm_flags & VM_SHARED))
+	if (is_cow)
 		page = vmf->cow_page;
 	else
 		page = vmf->page;
@@ -4765,8 +4765,10 @@ vm_fault_t finish_fault(struct vm_fault *vmf)
 	/* Re-check under ptl */
 	if (likely(!vmf_pte_changed(vmf))) {
 		struct folio *folio = page_folio(page);
+		int type = is_cow ? MM_ANONPAGES : mm_counter_file(folio);
 
 		set_pte_range(vmf, folio, page, 1, vmf->address);
+		add_mm_counter(vma->vm_mm, type, 1);
 		ret = 0;
 	} else {
 		update_mmu_tlb(vma, vmf->address, vmf->pte);
-- 
2.41.0

[PATCH 2/2] mm: filemap: batch mm counter updating in filemap_map_pages()

[Kefeng Wang]

Like copy_pte_range()/zap_pte_range(), make mm counter batch updating
in filemap_map_pages(), the 'lat_pagefault -P 1 file' test from lmbench
shows 12% improvement, and the percpu_counter_add_batch() is gone from
perf flame graph.

Signed-off-by: Kefeng Wang <wangkefeng.wang@huawei.com>
---
 mm/filemap.c | 18 ++++++++++--------
 1 file changed, 10 insertions(+), 8 deletions(-)

diff --git a/mm/filemap.c b/mm/filemap.c
index 04b813f0146c..c8d41ab5034b 100644
--- a/mm/filemap.c
+++ b/mm/filemap.c
@@ -3506,7 +3506,7 @@ static struct folio *next_uptodate_folio(struct xa_state *xas,
 static vm_fault_t filemap_map_folio_range(struct vm_fault *vmf,
 			struct folio *folio, unsigned long start,
 			unsigned long addr, unsigned int nr_pages,
-			unsigned int *mmap_miss)
+			unsigned long *rss, unsigned int *mmap_miss)
 {
 	vm_fault_t ret = 0;
 	struct page *page = folio_page(folio, start);
@@ -3540,8 +3540,7 @@ static vm_fault_t filemap_map_folio_range(struct vm_fault *vmf,
 skip:
 		if (count) {
 			set_pte_range(vmf, folio, page, count, addr);
-			add_mm_counter(vmf->vma->vm_mm, mm_counter_file(folio),
-				       count);
+			*rss += count;
 			folio_ref_add(folio, count);
 			if (in_range(vmf->address, addr, count * PAGE_SIZE))
 				ret = VM_FAULT_NOPAGE;
@@ -3556,7 +3555,7 @@ static vm_fault_t filemap_map_folio_range(struct vm_fault *vmf,
 
 	if (count) {
 		set_pte_range(vmf, folio, page, count, addr);
-		add_mm_counter(vmf->vma->vm_mm, mm_counter_file(folio), count);
+		*rss += count;
 		folio_ref_add(folio, count);
 		if (in_range(vmf->address, addr, count * PAGE_SIZE))
 			ret = VM_FAULT_NOPAGE;
@@ -3569,7 +3568,7 @@ static vm_fault_t filemap_map_folio_range(struct vm_fault *vmf,
 
 static vm_fault_t filemap_map_order0_folio(struct vm_fault *vmf,
 		struct folio *folio, unsigned long addr,
-		unsigned int *mmap_miss)
+		unsigned long *rss, unsigned int *mmap_miss)
 {
 	vm_fault_t ret = 0;
 	struct page *page = &folio->page;
@@ -3593,7 +3592,7 @@ static vm_fault_t filemap_map_order0_folio(struct vm_fault *vmf,
 		ret = VM_FAULT_NOPAGE;
 
 	set_pte_range(vmf, folio, page, 1, addr);
-	add_mm_counter(vmf->vma->vm_mm, mm_counter_file(folio), 1);
+	(*rss)++;
 	folio_ref_inc(folio);
 
 	return ret;
@@ -3610,6 +3609,7 @@ vm_fault_t filemap_map_pages(struct vm_fault *vmf,
 	XA_STATE(xas, &mapping->i_pages, start_pgoff);
 	struct folio *folio;
 	vm_fault_t ret = 0;
+	unsigned long rss = 0;
 	unsigned int nr_pages = 0, mmap_miss = 0, mmap_miss_saved;
 
 	rcu_read_lock();
@@ -3640,15 +3640,17 @@ vm_fault_t filemap_map_pages(struct vm_fault *vmf,
 
 		if (!folio_test_large(folio))
 			ret |= filemap_map_order0_folio(vmf,
-					folio, addr, &mmap_miss);
+					folio, addr, &rss, &mmap_miss);
 		else
 			ret |= filemap_map_folio_range(vmf, folio,
 					xas.xa_index - folio->index, addr,
-					nr_pages, &mmap_miss);
+					nr_pages, &rss, &mmap_miss);
 
 		folio_unlock(folio);
 		folio_put(folio);
 	} while ((folio = next_uptodate_folio(&xas, mapping, end_pgoff)) != NULL);
+
+	add_mm_counter(vma->vm_mm, mm_counter_file(folio), rss);
 	pte_unmap_unlock(vmf->pte, vmf->ptl);
 out:
 	rcu_read_unlock();
-- 
2.41.0

Re: [PATCH 2/2] mm: filemap: batch mm counter updating in filemap_map_pages()

[Matthew Wilcox]

On Fri, Apr 12, 2024 at 10:57:04AM +0800, Kefeng Wang wrote:
>  	} while ((folio = next_uptodate_folio(&xas, mapping, end_pgoff)) != NULL);
> +
> +	add_mm_counter(vma->vm_mm, mm_counter_file(folio), rss);

Can't folio be NULL here?

Re: [PATCH 2/2] mm: filemap: batch mm counter updating in filemap_map_pages()

[Kefeng Wang]

On 2024/4/12 11:17, Matthew Wilcox wrote:
> On Fri, Apr 12, 2024 at 10:57:04AM +0800, Kefeng Wang wrote:
>>   	} while ((folio = next_uptodate_folio(&xas, mapping, end_pgoff)) != NULL);
>> +
>> +	add_mm_counter(vma->vm_mm, mm_counter_file(folio), rss);
> 
> Can't folio be NULL here?
indeed, I need get mm counter type before while
>