[PATCH] mm: move buffer_heads_over_limit check up

[PATCH] mm: move buffer_heads_over_limit check up

[Hugh Dickins]

Not a functional change, just a minor internal cleanup: move the
buffer_heads_over_limit processing up from move_active_pages_to_lru()
(where it has to drop lock and reloop) to its caller shrink_active_list().

Signed-off-by: Hugh Dickins <hughd@google.com>
---
 mm/vmscan.c |   19 +++++++------------
 1 file changed, 7 insertions(+), 12 deletions(-)

--- mmotm.orig/mm/vmscan.c	2012-02-07 16:59:13.000000000 -0800
+++ mmotm/mm/vmscan.c	2012-02-07 17:07:23.800524771 -0800
@@ -1641,18 +1641,6 @@ static void move_active_pages_to_lru(str
 	unsigned long pgmoved = 0;
 	struct page *page;
 
-	if (buffer_heads_over_limit) {
-		spin_unlock_irq(&zone->lru_lock);
-		list_for_each_entry(page, list, lru) {
-			if (page_has_private(page) && trylock_page(page)) {
-				if (page_has_private(page))
-					try_to_release_page(page, 0);
-				unlock_page(page);
-			}
-		}
-		spin_lock_irq(&zone->lru_lock);
-	}
-
 	while (!list_empty(list)) {
 		struct lruvec *lruvec;
 
@@ -1734,6 +1722,13 @@ static void shrink_active_list(unsigned
 			continue;
 		}
 
+		if (buffer_heads_over_limit &&
+		    page_has_private(page) && trylock_page(page)) {
+			if (page_has_private(page))
+				try_to_release_page(page, 0);
+			unlock_page(page);
+		}
+
 		if (page_referenced(page, 0, mz->mem_cgroup, &vm_flags)) {
 			nr_rotated += hpage_nr_pages(page);
 			/*

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Fight unfair telecom internet charges in Canada: sign http://stopthemeter.ca/
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

Re: [PATCH] mm: move buffer_heads_over_limit check up

[Andrew Morton]

On Fri, 17 Feb 2012 16:00:14 -0800 (PST)
Hugh Dickins <hughd@google.com> wrote:

> Not a functional change, just a minor internal cleanup: move the
> buffer_heads_over_limit processing up from move_active_pages_to_lru()
> (where it has to drop lock and reloop) to its caller shrink_active_list().
> 
> Signed-off-by: Hugh Dickins <hughd@google.com>
> ---
>  mm/vmscan.c |   19 +++++++------------
>  1 file changed, 7 insertions(+), 12 deletions(-)
> 
> --- mmotm.orig/mm/vmscan.c	2012-02-07 16:59:13.000000000 -0800
> +++ mmotm/mm/vmscan.c	2012-02-07 17:07:23.800524771 -0800
> @@ -1641,18 +1641,6 @@ static void move_active_pages_to_lru(str
>  	unsigned long pgmoved = 0;
>  	struct page *page;
>  
> -	if (buffer_heads_over_limit) {
> -		spin_unlock_irq(&zone->lru_lock);
> -		list_for_each_entry(page, list, lru) {
> -			if (page_has_private(page) && trylock_page(page)) {
> -				if (page_has_private(page))
> -					try_to_release_page(page, 0);
> -				unlock_page(page);
> -			}
> -		}
> -		spin_lock_irq(&zone->lru_lock);
> -	}
> -
>  	while (!list_empty(list)) {
>  		struct lruvec *lruvec;
>  
> @@ -1734,6 +1722,13 @@ static void shrink_active_list(unsigned
>  			continue;
>  		}
>  
> +		if (buffer_heads_over_limit &&
> +		    page_has_private(page) && trylock_page(page)) {
> +			if (page_has_private(page))
> +				try_to_release_page(page, 0);
> +			unlock_page(page);
> +		}
> +
>  		if (page_referenced(page, 0, mz->mem_cgroup, &vm_flags)) {
>  			nr_rotated += hpage_nr_pages(page);
>  			/*

yup.

I don't think there's a lot of point in trying to micro-optimise the
buffer_heads_over_limit==true case, either.  But I suppose that
pointlessly locking 1000000 anon pages is indeed pointless.  Hopefully
there _is_ a point in micro-optimising the actual test for
buffer_heads_over_limit==true.  So...

--- a/mm/vmscan.c~mm-vmscan-forcibly-scan-highmem-if-there-are-too-many-buffer_heads-pinning-highmem-fix-fix
+++ a/mm/vmscan.c
@@ -1723,11 +1723,12 @@ static void shrink_active_list(unsigned 
 			continue;
 		}
 
-		if (buffer_heads_over_limit &&
-		    page_has_private(page) && trylock_page(page)) {
-			if (page_has_private(page))
-				try_to_release_page(page, 0);
-			unlock_page(page);
+		if (unlikely(buffer_heads_over_limit)) {
+			if (page_has_private(page) && trylock_page(page)) {
+				if (page_has_private(page))
+					try_to_release_page(page, 0);
+				unlock_page(page);
+			}
 		}
 
 		if (page_referenced(page, 0, mz->mem_cgroup, &vm_flags)) {
_

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Fight unfair telecom internet charges in Canada: sign http://stopthemeter.ca/
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

Re: [PATCH] mm: move buffer_heads_over_limit check up

[Hugh Dickins]

On Fri, 17 Feb 2012, Andrew Morton wrote:
> 
> I don't think there's a lot of point in trying to micro-optimise the
> buffer_heads_over_limit==true case, either.  But I suppose that
> pointlessly locking 1000000 anon pages is indeed pointless.  Hopefully
> there _is_ a point in micro-optimising the actual test for
> buffer_heads_over_limit==true.  So...

That's fine, yes, and thanks for putting this in.  I just want to
exonerate Mel: it's not a fix to his useful work on buffer_heads_over_limit,
but I don't think he'll mind terribly that you've named it thus.

If it's a fix to anything, it's to my 3.3 free_hot_cold_page_list-ification
of shrink_active_list(), which was silly to leave the buffer_heads business
down there in move_active_pages_to_lru().

And the only reason I'm concerned to get it in, is that it's in an area
which I then trample over in the per-memcg per-zone locking series (as
is Hillf's rearrangement around update_isolated_counts()), so it's
convenient for me to have both of those in the base, instead of
having to put them in a prologue.

If I were to worry about the buffer_heads_over_limit situation itself,
I might worry about the unevictable pages which we never scan.

Hugh

> 
> --- a/mm/vmscan.c~mm-vmscan-forcibly-scan-highmem-if-there-are-too-many-buffer_heads-pinning-highmem-fix-fix
> +++ a/mm/vmscan.c
> @@ -1723,11 +1723,12 @@ static void shrink_active_list(unsigned 
>  			continue;
>  		}
>  
> -		if (buffer_heads_over_limit &&
> -		    page_has_private(page) && trylock_page(page)) {
> -			if (page_has_private(page))
> -				try_to_release_page(page, 0);
> -			unlock_page(page);
> +		if (unlikely(buffer_heads_over_limit)) {
> +			if (page_has_private(page) && trylock_page(page)) {
> +				if (page_has_private(page))
> +					try_to_release_page(page, 0);
> +				unlock_page(page);
> +			}
>  		}
>  
>  		if (page_referenced(page, 0, mz->mem_cgroup, &vm_flags)) {
> _

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Fight unfair telecom internet charges in Canada: sign http://stopthemeter.ca/
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>