Skip to contentSkip to conversation

mm: Fix for PG_reserved page flag clearing

PATCH In review Masayoshi Mizuma 2 participants 4 messages Feb 19, 2018

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 mm: Fix for PG_reserved page flag clearing

From: Masayoshi Mizuma <[email protected]>

struct page is inizialized as zero in __init_single_page(). If the page is offlined page, PG_reserved flag is set in early boot time before __init_single_page(), so we should not clear the flag.

The real problem is that we can not online the offlined page through following sysfs operation because offlined page is expected PG_reserved flag is set.
It is not needed the initialization, so remove it simply.

Code:

  static int online_pages_range(unsigned long start_pfn, 
  ...
          if (PageReserved(pfn_to_page(start_pfn))) <= HERE!!
                  for (i = 0; i < nr_pages; i++) {
                          page = pfn_to_page(start_pfn + i);
                          (*online_page_callback)(page);
                          onlined_pages++;
  sysfs operation:
  # echo online > /sys/devices/system/node/node2/memory12288/online
  # cat /sys/devices/system/node/node2/memory12288/online 
  1
  # cat /sys/devices/system/node/node2/meminfo 
  Node 2 MemTotal:              0 kB
Fixes f7f99100d8d9 ("mm: stop zeroing memory during allocation in vmemmap")
---
 mm/page_alloc.c | 1 -
 1 file changed, 1 deletion(-)
mm/page_alloc.c +0 -1
index 76c9688..3260cd2 100644
@@ -1179,7 +1179,6 @@ static void free_one_page(struct zone *zone,
 static void __meminit __init_single_page(struct page *page, unsigned long pfn,
 				unsigned long zone, int nid)
 {
-	mm_zero_struct_page(page);
 	set_page_links(page, zone, nid, pfn);
 	init_page_count(page);
 	page_mapcount_reset(page);
Signature · 9 lines
2.16.1

- Masayoshi

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to [email protected].  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>
reviewer
#1 lore raw
Quoted 22 lines · Masayoshi Mizuma

From: Masayoshi Mizuma <[email protected]>

struct page is inizialized as zero in __init_single_page(). If the page is offlined page, PG_reserved flag is set in early boot time before __init_single_page(), so we should not clear the flag.

The real problem is that we can not online the offlined page through following sysfs operation because offlined page is expected PG_reserved flag is set.
It is not needed the initialization, so remove it simply.

Code:

  static int online_pages_range(unsigned long start_pfn, 
  ...
          if (PageReserved(pfn_to_page(start_pfn))) <= HERE!!
                  for (i = 0; i < nr_pages; i++) {
                          page = pfn_to_page(start_pfn + i);
                          (*online_page_callback)(page);
                          onlined_pages++;
  sysfs operation:
Quoted 5 lines
  # echo online > /sys/devices/system/node/node2/memory12288/online
  # cat /sys/devices/system/node/node2/memory12288/online 
  1
  # cat /sys/devices/system/node/node2/meminfo 
  Node 2 MemTotal:              0 kB

Nack. The patch is simply wrong. We do need to zero page for the boot pages. I believe the fix you are looking for is 9bb5a391f9a5 ("mm,
memory_hotplug: fix memmap initialization"). Or do you still see a problem with this patch applied?

Quoted 22 lines
Fixes f7f99100d8d9 ("mm: stop zeroing memory during allocation in vmemmap")
---
 mm/page_alloc.c | 1 -
 1 file changed, 1 deletion(-)
mm/page_alloc.c +0 -1
index 76c9688..3260cd2 100644
@@ -1179,7 +1179,6 @@ static void free_one_page(struct zone *zone,
 static void __meminit __init_single_page(struct page *page, unsigned long pfn,
 				unsigned long zone, int nid)
 {
-	mm_zero_struct_page(page);
 	set_page_links(page, zone, nid, pfn);
 	init_page_count(page);
 	page_mapcount_reset(page);
Signature · 3 lines
2.16.1

- Masayoshi
Signature · 8 lines
Michal Hocko
SUSE Labs

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to [email protected].  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>
#2 lore raw

Hello Michal,

Quoted 29 lines · Mon, 19 Feb 2018 18:19:16 +0100 Michal Hocko wrote:
Quoted 22 lines · Masayoshi Mizuma

From: Masayoshi Mizuma <[email protected]>

struct page is inizialized as zero in __init_single_page(). If the page is offlined page, PG_reserved flag is set in early boot time before __init_single_page(), so we should not clear the flag.

The real problem is that we can not online the offlined page through following sysfs operation because offlined page is expected PG_reserved flag is set.
It is not needed the initialization, so remove it simply.

Code:

  static int online_pages_range(unsigned long start_pfn, 
  ...
          if (PageReserved(pfn_to_page(start_pfn))) <= HERE!!
                  for (i = 0; i < nr_pages; i++) {
                          page = pfn_to_page(start_pfn + i);
                          (*online_page_callback)(page);
                          onlined_pages++;
  sysfs operation:
Quoted 5 lines
  # echo online > /sys/devices/system/node/node2/memory12288/online
  # cat /sys/devices/system/node/node2/memory12288/online 
  1
  # cat /sys/devices/system/node/node2/meminfo 
  Node 2 MemTotal:              0 kB
Quoted 4 lines

Nack. The patch is simply wrong. We do need to zero page for the boot pages. I believe the fix you are looking for is 9bb5a391f9a5 ("mm,
memory_hotplug: fix memmap initialization"). Or do you still see a problem with this patch applied?

I have confirmed the problem is fixed by your patch 9bb5a391f9a5. (I had tested it in 4.15.2, so I did not notice your patch, sorry)

Thank you for the fix!

- Masayoshi

Quoted 22 lines
Quoted 22 lines
Fixes f7f99100d8d9 ("mm: stop zeroing memory during allocation in vmemmap")
---
 mm/page_alloc.c | 1 -
 1 file changed, 1 deletion(-)
mm/page_alloc.c +0 -1
index 76c9688..3260cd2 100644
@@ -1179,7 +1179,6 @@ static void free_one_page(struct zone *zone,
 static void __meminit __init_single_page(struct page *page, unsigned long pfn,
 				unsigned long zone, int nid)
 {
-	mm_zero_struct_page(page);
 	set_page_links(page, zone, nid, pfn);
 	init_page_count(page);
 	page_mapcount_reset(page);
Signature · 1 line
2.16.1
Quoted 2 lines
Quoted 1 line

- Masayoshi

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>

reviewer
#3 lore raw
Quoted 32 lines · Masayoshi Mizuma

Hello Michal,

Quoted 29 lines · Mon, 19 Feb 2018 18:19:16 +0100 Michal Hocko wrote:
Quoted 27 lines · Masayoshi Mizuma

From: Masayoshi Mizuma <[email protected]>

struct page is inizialized as zero in __init_single_page(). If the page is offlined page, PG_reserved flag is set in early boot time before __init_single_page(), so we should not clear the flag.

The real problem is that we can not online the offlined page through following sysfs operation because offlined page is expected PG_reserved flag is set.
It is not needed the initialization, so remove it simply.

Code:

  static int online_pages_range(unsigned long start_pfn, 
  ...
          if (PageReserved(pfn_to_page(start_pfn))) <= HERE!!
                  for (i = 0; i < nr_pages; i++) {
                          page = pfn_to_page(start_pfn + i);
                          (*online_page_callback)(page);
                          onlined_pages++;
  sysfs operation:
  # echo online > /sys/devices/system/node/node2/memory12288/online
  # cat /sys/devices/system/node/node2/memory12288/online 
  1
  # cat /sys/devices/system/node/node2/meminfo 
  Node 2 MemTotal:              0 kB
Quoted 7 lines
Quoted 4 lines

Nack. The patch is simply wrong. We do need to zero page for the boot pages. I believe the fix you are looking for is 9bb5a391f9a5 ("mm,
memory_hotplug: fix memmap initialization"). Or do you still see a problem with this patch applied?

I have confirmed the problem is fixed by your patch 9bb5a391f9a5. (I had tested it in 4.15.2, so I did not notice your patch, sorry)

I have posted the backport for the 4.15 stable tree just an hour ago so it should appear in the next stable release.

Thanks for double checking.

Signature · 8 lines
Michal Hocko
SUSE Labs

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to [email protected].  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>