diff mbox series

mm/vmstat: add events for THP max_ptes_* exceeds

Message ID 20211225094036.574157-1-yang.yang29@zte.com.cn (mailing list archive)
State New
Headers show
Series mm/vmstat: add events for THP max_ptes_* exceeds | expand

Commit Message

CGEL Dec. 25, 2021, 9:40 a.m. UTC 
From: Yang Yang <yang.yang29@zte.com.cn>

There are interfaces to adjust max_ptes_none, max_ptes_swap,
max_ptes_shared values, see
/sys/kernel/mm/transparent_hugepage/khugepaged/.

But system administrator maynot know which value is the best. So
Add those events to support adjusting max_ptes_* to suitable values.

For example, if default max_ptes_swap value causes too much failures,
and system uses zram whose IO is fast, administrator could increase
max_ptes_swap until THP_SCAN_EXCEED_SWAP_PTE not increase anymore.

Signed-off-by: Yang Yang <yang.yang29@zte.com.cn>
---
 include/linux/vm_event_item.h | 3 +++
 mm/khugepaged.c               | 7 +++++++
 mm/vmstat.c                   | 3 +++
 3 files changed, 13 insertions(+)

Comments

Huang, Ying Dec. 27, 2021, 3:19 a.m. UTC | #1 
cgel.zte@gmail.com writes:

> From: Yang Yang <yang.yang29@zte.com.cn>
>
> There are interfaces to adjust max_ptes_none, max_ptes_swap,
> max_ptes_shared values, see
> /sys/kernel/mm/transparent_hugepage/khugepaged/.
>
> But system administrator maynot know which value is the best. So
> Add those events to support adjusting max_ptes_* to suitable values.
>
> For example, if default max_ptes_swap value causes too much failures,
> and system uses zram whose IO is fast, administrator could increase
> max_ptes_swap until THP_SCAN_EXCEED_SWAP_PTE not increase anymore.
>
> Signed-off-by: Yang Yang <yang.yang29@zte.com.cn>
> ---
>  include/linux/vm_event_item.h | 3 +++
>  mm/khugepaged.c               | 7 +++++++
>  mm/vmstat.c                   | 3 +++
>  3 files changed, 13 insertions(+)
>
> diff --git a/include/linux/vm_event_item.h b/include/linux/vm_event_item.h
> index a185cc75ff52..7b2363388bfa 100644
> --- a/include/linux/vm_event_item.h
> +++ b/include/linux/vm_event_item.h
> @@ -98,6 +98,9 @@ enum vm_event_item { PGPGIN, PGPGOUT, PSWPIN, PSWPOUT,
>  		THP_SPLIT_PAGE_FAILED,
>  		THP_DEFERRED_SPLIT_PAGE,
>  		THP_SPLIT_PMD,
> +		THP_SCAN_EXCEED_NONE_PTE,
> +		THP_SCAN_EXCEED_SWAP_PTE,
> +		THP_SCAN_EXCEED_SHARED_PTE,

"SCAN" appears too general, how about

  THP_COLLAPSE_EXCEED_SWAP_PTE

?

Best Regards,
Huang, Ying

[snip]
CGEL Dec. 27, 2021, 3:59 a.m. UTC | #2 
On Mon, Dec 27, 2021 at 11:19:32AM +0800, Huang, Ying wrote:
> cgel.zte@gmail.com writes:
> 
> > From: Yang Yang <yang.yang29@zte.com.cn>
> >
> > There are interfaces to adjust max_ptes_none, max_ptes_swap,
> > max_ptes_shared values, see
> > /sys/kernel/mm/transparent_hugepage/khugepaged/.
> >
> > But system administrator maynot know which value is the best. So
> > Add those events to support adjusting max_ptes_* to suitable values.
> >
> > For example, if default max_ptes_swap value causes too much failures,
> > and system uses zram whose IO is fast, administrator could increase
> > max_ptes_swap until THP_SCAN_EXCEED_SWAP_PTE not increase anymore.
> >
> > Signed-off-by: Yang Yang <yang.yang29@zte.com.cn>
> > ---
> >  include/linux/vm_event_item.h | 3 +++
> >  mm/khugepaged.c               | 7 +++++++
> >  mm/vmstat.c                   | 3 +++
> >  3 files changed, 13 insertions(+)
> >
> > diff --git a/include/linux/vm_event_item.h b/include/linux/vm_event_item.h
> > index a185cc75ff52..7b2363388bfa 100644
> > --- a/include/linux/vm_event_item.h
> > +++ b/include/linux/vm_event_item.h
> > @@ -98,6 +98,9 @@ enum vm_event_item { PGPGIN, PGPGOUT, PSWPIN, PSWPOUT,
> >  		THP_SPLIT_PAGE_FAILED,
> >  		THP_DEFERRED_SPLIT_PAGE,
> >  		THP_SPLIT_PMD,
> > +		THP_SCAN_EXCEED_NONE_PTE,
> > +		THP_SCAN_EXCEED_SWAP_PTE,
> > +		THP_SCAN_EXCEED_SHARED_PTE,
> 
> "SCAN" appears too general, how about
> 
>   THP_COLLAPSE_EXCEED_SWAP_PTE
> 
> ?
>
Thanks for your suggesting.

But I think SCAN is easy to understand, it is SCAN behaviour cause the event,
and we know this event is comes from SCAN_EXCEED_*_PTE.

> Best Regards,
> Huang, Ying
> 
> [snip]
diff mbox series

Patch

diff --git a/include/linux/vm_event_item.h b/include/linux/vm_event_item.h
index a185cc75ff52..7b2363388bfa 100644
--- a/include/linux/vm_event_item.h
+++ b/include/linux/vm_event_item.h
@@ -98,6 +98,9 @@  enum vm_event_item { PGPGIN, PGPGOUT, PSWPIN, PSWPOUT,
 		THP_SPLIT_PAGE_FAILED,
 		THP_DEFERRED_SPLIT_PAGE,
 		THP_SPLIT_PMD,
+		THP_SCAN_EXCEED_NONE_PTE,
+		THP_SCAN_EXCEED_SWAP_PTE,
+		THP_SCAN_EXCEED_SHARED_PTE,
 #ifdef CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD
 		THP_SPLIT_PUD,
 #endif
diff --git a/mm/khugepaged.c b/mm/khugepaged.c
index 7720189a2da7..abcd6bd8b97a 100644
--- a/mm/khugepaged.c
+++ b/mm/khugepaged.c
@@ -618,6 +618,7 @@  static int __collapse_huge_page_isolate(struct vm_area_struct *vma,
 				continue;
 			} else {
 				result = SCAN_EXCEED_NONE_PTE;
+				count_vm_event(THP_SCAN_EXCEED_NONE_PTE);
 				goto out;
 			}
 		}
@@ -636,6 +637,7 @@  static int __collapse_huge_page_isolate(struct vm_area_struct *vma,
 		if (page_mapcount(page) > 1 &&
 				++shared > khugepaged_max_ptes_shared) {
 			result = SCAN_EXCEED_SHARED_PTE;
+			count_vm_event(THP_SCAN_EXCEED_SHARED_PTE);
 			goto out;
 		}
 
@@ -1253,6 +1255,7 @@  static int khugepaged_scan_pmd(struct mm_struct *mm,
 				continue;
 			} else {
 				result = SCAN_EXCEED_SWAP_PTE;
+				count_vm_event(THP_SCAN_EXCEED_SWAP_PTE);
 				goto out_unmap;
 			}
 		}
@@ -1262,6 +1265,7 @@  static int khugepaged_scan_pmd(struct mm_struct *mm,
 				continue;
 			} else {
 				result = SCAN_EXCEED_NONE_PTE;
+				count_vm_event(THP_SCAN_EXCEED_NONE_PTE);
 				goto out_unmap;
 			}
 		}
@@ -1290,6 +1294,7 @@  static int khugepaged_scan_pmd(struct mm_struct *mm,
 		if (page_mapcount(page) > 1 &&
 				++shared > khugepaged_max_ptes_shared) {
 			result = SCAN_EXCEED_SHARED_PTE;
+			count_vm_event(THP_SCAN_EXCEED_SHARED_PTE);
 			goto out_unmap;
 		}
 
@@ -2006,6 +2011,7 @@  static void khugepaged_scan_file(struct mm_struct *mm,
 		if (xa_is_value(page)) {
 			if (++swap > khugepaged_max_ptes_swap) {
 				result = SCAN_EXCEED_SWAP_PTE;
+				count_vm_event(THP_SCAN_EXCEED_SWAP_PTE);
 				break;
 			}
 			continue;
@@ -2056,6 +2062,7 @@  static void khugepaged_scan_file(struct mm_struct *mm,
 	if (result == SCAN_SUCCEED) {
 		if (present < HPAGE_PMD_NR - khugepaged_max_ptes_none) {
 			result = SCAN_EXCEED_NONE_PTE;
+			count_vm_event(THP_SCAN_EXCEED_NONE_PTE);
 		} else {
 			node = khugepaged_find_target_node();
 			collapse_file(mm, file, start, hpage, node);
diff --git a/mm/vmstat.c b/mm/vmstat.c
index d701c335628c..4057372745d0 100644
--- a/mm/vmstat.c
+++ b/mm/vmstat.c
@@ -1353,6 +1353,9 @@  const char * const vmstat_text[] = {
 	"thp_split_page_failed",
 	"thp_deferred_split_page",
 	"thp_split_pmd",
+	"thp_scan_exceed_none_pte",
+	"thp_scan_exceed_swap_pte",
+	"thp_scan_exceed_share_pte",
 #ifdef CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD
 	"thp_split_pud",
 #endif