[PATCH 1/2] fix __set_page_dirty_no_writeback() return value

[PATCH 1/2] fix __set_page_dirty_no_writeback() return value

[Bob Liu]

__set_page_dirty_no_writeback() should return true if it actually transitioned
the page from a clean to dirty state although it seems nobody used its return
value now.

Signed-off-by: Bob Liu <lliubbo@gmail.com>
---
 mm/page-writeback.c |    4 +---
 1 files changed, 1 insertions(+), 3 deletions(-)

diff --git a/mm/page-writeback.c b/mm/page-writeback.c
index bf85062..e8f5f06 100644
--- a/mm/page-writeback.c
+++ b/mm/page-writeback.c
@@ -1157,9 +1157,7 @@ EXPORT_SYMBOL(write_one_page);
  */
 int __set_page_dirty_no_writeback(struct page *page)
 {
-	if (!PageDirty(page))
-		SetPageDirty(page);
-	return 0;
+	return !TestSetPageDirty(page);
 }
 
 /*
-- 
1.6.3.3

--
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 policy in Canada: sign http://dissolvethecrtc.ca/
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

[PATCH 2/2] clean up set_page_dirty()

[Bob Liu]

Use TestSetPageDirty() to clean up set_page_dirty().

Signed-off-by: Bob Liu <lliubbo@gmail.com>
---
 mm/page-writeback.c |    7 ++-----
 1 files changed, 2 insertions(+), 5 deletions(-)

diff --git a/mm/page-writeback.c b/mm/page-writeback.c
index e8f5f06..da86224 100644
--- a/mm/page-writeback.c
+++ b/mm/page-writeback.c
@@ -1268,11 +1268,8 @@ int set_page_dirty(struct page *page)
 #endif
 		return (*spd)(page);
 	}
-	if (!PageDirty(page)) {
-		if (!TestSetPageDirty(page))
-			return 1;
-	}
-	return 0;
+
+	return !TestSetPageDirty(page);
 }
 EXPORT_SYMBOL(set_page_dirty);
 
-- 
1.6.3.3

--
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 policy in Canada: sign http://dissolvethecrtc.ca/
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

Re: [PATCH 1/2] fix __set_page_dirty_no_writeback() return value

[Andrew Morton]

On Wed, 10 Nov 2010 17:00:27 +0800
Bob Liu <lliubbo@gmail.com> wrote:

> __set_page_dirty_no_writeback() should return true if it actually transitioned
> the page from a clean to dirty state although it seems nobody used its return
> value now.
> 
> Signed-off-by: Bob Liu <lliubbo@gmail.com>
> ---
>  mm/page-writeback.c |    4 +---
>  1 files changed, 1 insertions(+), 3 deletions(-)
> 
> diff --git a/mm/page-writeback.c b/mm/page-writeback.c
> index bf85062..e8f5f06 100644
> --- a/mm/page-writeback.c
> +++ b/mm/page-writeback.c
> @@ -1157,9 +1157,7 @@ EXPORT_SYMBOL(write_one_page);
>   */
>  int __set_page_dirty_no_writeback(struct page *page)
>  {
> -	if (!PageDirty(page))
> -		SetPageDirty(page);
> -	return 0;
> +	return !TestSetPageDirty(page);
>  }

The idea here is to avoid modifying the cacheline which contains the
pageframe if that page was already dirty.  So that a set_page_dirty()
against an already-dirty page doesn't result in the CPU having to
perform writeback of the cacheline.

The code as it stands assumes that a test_and_set_bit() will
unconditionally modify the target.  This might not be true of certain
CPUs - perhaps they optimise away the write in that case, I don't know.

Yes, you're right, __set_page_dirty_no_writeback() should return the
correct value.  But the way to do that while preserving this
optimisation is

	if (!PageDirty(page))
		return !TestSetPageDirty(page);
	return 0;


This optimisation is used in quite a few places and is done in
differeing ways depending upon what is being modified.  I've never
really seen any quantification of its effectiveness.

--
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 policy in Canada: sign http://dissolvethecrtc.ca/
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

Re: [PATCH 2/2] clean up set_page_dirty()

[Andrew Morton]

On Wed, 10 Nov 2010 17:00:28 +0800
Bob Liu <lliubbo@gmail.com> wrote:

> Use TestSetPageDirty() to clean up set_page_dirty().
> 
> Signed-off-by: Bob Liu <lliubbo@gmail.com>
> ---
>  mm/page-writeback.c |    7 ++-----
>  1 files changed, 2 insertions(+), 5 deletions(-)
> 
> diff --git a/mm/page-writeback.c b/mm/page-writeback.c
> index e8f5f06..da86224 100644
> --- a/mm/page-writeback.c
> +++ b/mm/page-writeback.c
> @@ -1268,11 +1268,8 @@ int set_page_dirty(struct page *page)
>  #endif
>  		return (*spd)(page);
>  	}
> -	if (!PageDirty(page)) {
> -		if (!TestSetPageDirty(page))
> -			return 1;
> -	}
> -	return 0;
> +
> +	return !TestSetPageDirty(page);
>  }
>  EXPORT_SYMBOL(set_page_dirty);

This just undoes the optimisation.

We could do

-		if (!TestSetPageDirty(page))
-			return 1;
+		return !TestSetPageDirty(page);

I suppose.

Or even


		return (!TestSetPageDirty(page) ^ 1);

if we're feeling stupid, and if TestSetPageDirty() reliably returns
1/0, and if that really is superior (by eliminating a test-n-branch).


--
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 policy in Canada: sign http://dissolvethecrtc.ca/
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

Re: [PATCH 2/2] clean up set_page_dirty()

[Michel Lespinasse]

On Wed, Nov 10, 2010 at 1:00 AM, Bob Liu <lliubbo@gmail.com> wrote:
> Use TestSetPageDirty() to clean up set_page_dirty().
>
> Signed-off-by: Bob Liu <lliubbo@gmail.com>
> ---
>  mm/page-writeback.c |    7 ++-----
>  1 files changed, 2 insertions(+), 5 deletions(-)
>
> diff --git a/mm/page-writeback.c b/mm/page-writeback.c
> index e8f5f06..da86224 100644
> --- a/mm/page-writeback.c
> +++ b/mm/page-writeback.c
> @@ -1268,11 +1268,8 @@ int set_page_dirty(struct page *page)
>  #endif
>                return (*spd)(page);
>        }
> -       if (!PageDirty(page)) {
> -               if (!TestSetPageDirty(page))
> -                       return 1;
> -       }
> -       return 0;
> +
> +       return !TestSetPageDirty(page);
>  }
>  EXPORT_SYMBOL(set_page_dirty);

TestSetPageDirty compiles to a locked bts instruction (on x86). This
will acquire the cache line for exclusive access, even when the bit is
already set. I think this is why we have an extra if
(!PageDirty(page)) test - we don't want to cause cache coherency
overhead if the page is already dirty.

-- 
Michel "Walken" Lespinasse
A program is never fully debugged until the last user dies.

--
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 policy in Canada: sign http://dissolvethecrtc.ca/
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>