[PATCH v3 0/2] mm: batch mm counter updating in filemap_map_pages()

[PATCH v3 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

v3: 
- make is_cow to bool in patch1 and fix null folio in patch2
- retest, improvement same with v1

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 | 17 ++++++++++++-----
 mm/memory.c  |  8 +++++---
 2 files changed, 17 insertions(+), 8 deletions(-)

-- 
2.41.0

[PATCH v3 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 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..fdfe965f32e4 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;
+	bool 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 v3 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(), since folios type are same(MM_SHMEMPAGES or
MM_FILEPAGES) in filemap_map_pages(), only check the first folio type
is enough, 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 | 21 ++++++++++++---------
 1 file changed, 12 insertions(+), 9 deletions(-)

diff --git a/mm/filemap.c b/mm/filemap.c
index 04b813f0146c..531af4acc667 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;
@@ -3610,7 +3609,8 @@ 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 int nr_pages = 0, mmap_miss = 0, mmap_miss_saved;
+	unsigned long rss = 0;
+	unsigned int nr_pages = 0, mmap_miss = 0, mmap_miss_saved, folio_type;
 
 	rcu_read_lock();
 	folio = next_uptodate_folio(&xas, mapping, end_pgoff);
@@ -3629,6 +3629,8 @@ vm_fault_t filemap_map_pages(struct vm_fault *vmf,
 		folio_put(folio);
 		goto out;
 	}
+
+	folio_type = mm_counter_file(folio);
 	do {
 		unsigned long end;
 
@@ -3640,15 +3642,16 @@ 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, folio_type, rss);
 	pte_unmap_unlock(vmf->pte, vmf->ptl);
 out:
 	rcu_read_unlock();
-- 
2.41.0

Re: [PATCH v3 0/2] mm: batch mm counter updating in filemap_map_pages()

[Andrew Morton]

On Fri, 12 Apr 2024 14:47:49 +0800 Kefeng Wang <wangkefeng.wang@huawei.com> wrote:

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

Are any performance testing results available?

Re: [PATCH v3 0/2] mm: batch mm counter updating in filemap_map_pages()

[Matthew Wilcox]

On Fri, Apr 12, 2024 at 04:12:17PM -0700, Andrew Morton wrote:
> On Fri, 12 Apr 2024 14:47:49 +0800 Kefeng Wang <wangkefeng.wang@huawei.com> wrote:
> 
> > Let's batch mm counter updating to accelerate filemap_map_pages(). 
> 
> Are any performance testing results available?

Patch 2/2 says 12% improvement

Re: [PATCH v3 0/2] mm: batch mm counter updating in filemap_map_pages()

[Kefeng Wang]

On 2024/4/13 9:05, Matthew Wilcox wrote:
> On Fri, Apr 12, 2024 at 04:12:17PM -0700, Andrew Morton wrote:
>> On Fri, 12 Apr 2024 14:47:49 +0800 Kefeng Wang <wangkefeng.wang@huawei.com> wrote:
>>
>>> Let's batch mm counter updating to accelerate filemap_map_pages().
>>
>> Are any performance testing results available?
> 
> Patch 2/2 says 12% improvement

Yes, lat_pagefault with 512M file from lmbench, thanks.