Skip to contentSkip to conversation

mm: change may_enter_fs check condition

PATCH In review yalin wang 3 participants 4 messages Nov 13, 2015 – Nov 16, 2015

Subsystems (3)
lore
Share & tools
lore

yalin wang

author
#0 lore raw

PATCH 1/1 mm: change may_enter_fs check condition

Add page_is_file_cache() for __GFP_FS check,
otherwise, a Pageswapcache() && PageDirty() page can always be write back if the gfp flag is __GFP_FS, this is not the expected behavior.

---
 mm/vmscan.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
mm/vmscan.c +1 -1
index bd2918e..f8fc8c1 100644
@@ -930,7 +930,7 @@ static unsigned long shrink_page_list(struct list_head *page_list,
 		if (page_mapped(page) || PageSwapCache(page))
 			sc->nr_scanned++;
 
-		may_enter_fs = (sc->gfp_mask & __GFP_FS) ||
+		may_enter_fs = (page_is_file_cache(page) && (sc->gfp_mask & __GFP_FS)) ||
 			(PageSwapCache(page) && (sc->gfp_mask & __GFP_IO));
 
 		/*
Signature · 7 lines
1.9.1

--
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>
#1 lore raw
Quoted 3 lines · yalin wang

Add page_is_file_cache() for __GFP_FS check,
otherwise, a Pageswapcache() && PageDirty() page can always be write back if the gfp flag is __GFP_FS, this is not the expected behavior.

I'm not sure I understand your point correctly *), but you seem to imply that there would be an allocation that has __GFP_FS but doesn't have __GFP_IO? Are there such allocations and does it make sense?

*) It helps to state which problem you actually observed and are trying to fix. Or was this found by code inspection? In that case describe the theoretical problem, as "expected behavior" isn't always understood by everyone the same.

Quoted 11 lines
---
  mm/vmscan.c | 2 +-
  1 file changed, 1 insertion(+), 1 deletion(-)
mm/vmscan.c +0 -0
index bd2918e..f8fc8c1 100644

mm/vmscan.c static unsigned long shrink_page_list(struct list_head *page_list, · line 936

Quoted 8 lines
mm/vmscan.c +1 -1
@@ -930,7 +930,7 @@ static unsigned long shrink_page_list(struct list_head *page_list,
  		if (page_mapped(page) || PageSwapCache(page))
  			sc->nr_scanned++;
 
-		may_enter_fs = (sc->gfp_mask & __GFP_FS) ||
+		may_enter_fs = (page_is_file_cache(page) && (sc->gfp_mask & __GFP_FS)) ||
  			(PageSwapCache(page) && (sc->gfp_mask & __GFP_IO));
 
  		/*

--
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
#2 lore raw
Quoted 8 lines · Vlastimil Babka
Quoted 3 lines · yalin wang

Add page_is_file_cache() for __GFP_FS check,
otherwise, a Pageswapcache() && PageDirty() page can always be write back if the gfp flag is __GFP_FS, this is not the expected behavior.

I'm not sure I understand your point correctly *), but you seem to imply that there would be an allocation that has __GFP_FS but doesn't have __GFP_IO? Are there such allocations and does it make sense?

No it doesn't. There is a natural layering here and __GFP_FS allocations should contain __GFP_IO.

The patch as is makes only little sense to me. Are you seeing any issue which this is trying to fix?

Quoted 24 lines

*) It helps to state which problem you actually observed and are trying to fix. Or was this found by code inspection? In that case describe the theoretical problem, as "expected behavior" isn't always understood by everyone the same.

Quoted 19 lines
---
 mm/vmscan.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
mm/vmscan.c +1 -1
index bd2918e..f8fc8c1 100644
@@ -930,7 +930,7 @@ static unsigned long shrink_page_list(struct list_head *page_list,
 		if (page_mapped(page) || PageSwapCache(page))
 			sc->nr_scanned++;
 
-		may_enter_fs = (sc->gfp_mask & __GFP_FS) ||
+		may_enter_fs = (page_is_file_cache(page) && (sc->gfp_mask & __GFP_FS)) ||
 			(PageSwapCache(page) && (sc->gfp_mask & __GFP_IO));
 
 		/*
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>
author
#3 lore raw
Quoted 13 lines · Michal Hocko
Quoted 8 lines · Vlastimil Babka
Quoted 3 lines · yalin wang

Add page_is_file_cache() for __GFP_FS check,
otherwise, a Pageswapcache() && PageDirty() page can always be write back if the gfp flag is __GFP_FS, this is not the expected behavior.

I'm not sure I understand your point correctly *), but you seem to imply that there would be an allocation that has __GFP_FS but doesn't have __GFP_IO? Are there such allocations and does it make sense?

Quoted 5 lines

No it doesn't. There is a natural layering here and __GFP_FS allocations should contain __GFP_IO.

The patch as is makes only little sense to me. Are you seeing any issue which this is trying to fix?

mm..
i don’t see issue for this part ,
just feel confuse when i see code about this part ,
then i make a patch for this .
i am not sure if __GFP_FS will make sure __GFP_IO flag must be always set. if it is , i think can add comment here to make people clear . :)

Thanks

--
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>