__isolate_lru_page:skip unneeded "not"

PATCH RESEND In review Bob Liu 2 participants 4 messages Apr 1, 2010 – Apr 4, 2010

Subsystems (3)

Memory management MGLRU MM reclaim

lore

Share & tools

lore

Bob Liu

author

2010-04-01 13:38 UTC #0 lore raw

PATCH RESEND 1/1 __isolate_lru_page:skip unneeded "not"

PageActive(page) will return int 0 or 1, mode is also int 0 or 1, they are comparible so "not" is unneeded to be sure to boolean values.
I also collected the ISOLATE_BOTH check together.

Signed-off-by Bob Liu <[email protected]>

---
 mm/vmscan.c |   15 +++++----------
 1 files changed, 5 insertions(+), 10 deletions(-)

mm/vmscan.c +5 ~~-10~~

index e0e5f15..ce9ee85 100644
@@ -862,16 +862,11 @@ int __isolate_lru_page(struct page *page, int mode, int file)
 	if (!PageLRU(page))
 		return ret;
 
-	/*
-	 * When checking the active state, we need to be sure we are
-	 * dealing with comparible boolean values.  Take the logical not
-	 * of each.
-	 */
-	if (mode != ISOLATE_BOTH && (!PageActive(page) != !mode))
-		return ret;
-
-	if (mode != ISOLATE_BOTH && page_is_file_cache(page) != file)
-		return ret;
+	if (mode != ISOLATE_BOTH) {
+		if ((PageActive(page) != mode) ||
+			(page_is_file_cache(page) != file))
+				return ret;
+	}
 
 	/*
 	 * When this function is being called for lumpy reclaim, we

Signature · 7 lines

1.5.6.3

--
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>

Andrew Morton

maintainer

2010-04-02 22:05 UTC #1 lore raw

Quoted 15 lines · On Thu, 1 Apr 2010 21:37:35 +0800 Bob Liu <[email protected]> wrote:

PageActive(page) will return int 0 or 1, mode is also int 0 or 1, they are comparible so "not" is unneeded to be sure to boolean values.
I also collected the ISOLATE_BOTH check together.
Signed-off-by Bob Liu <[email protected]>
---
 mm/vmscan.c |   15 +++++----------
 1 files changed, 5 insertions(+), 10 deletions(-)
mm/vmscan.c +0 -0
index e0e5f15..ce9ee85 100644

mm/vmscan.c int __isolate_lru_page(struct page *page, int mode, int file) · line 869

Quoted 8 lines

mm/vmscan.c +0 -5

@@ -862,16 +862,11 @@ int __isolate_lru_page(struct page *page, int mode, int file)
 	if (!PageLRU(page))
 		return ret;
 
-	/*
-	 * When checking the active state, we need to be sure we are
-	 * dealing with comparible boolean values.  Take the logical not
-	 * of each.
-	 */

You deleted a spelling mistake too!

mm/vmscan.c int __isolate_lru_page(struct page *page, int mode, int file) · line 869

Quoted 10 lines

mm/vmscan.c +5 -5

-	if (mode != ISOLATE_BOTH && (!PageActive(page) != !mode))
-		return ret;
-
-	if (mode != ISOLATE_BOTH && page_is_file_cache(page) != file)
-		return ret;
+	if (mode != ISOLATE_BOTH) {
+		if ((PageActive(page) != mode) ||
+			(page_is_file_cache(page) != file))
+				return ret;
+	}

The compiler should be able to avoid testing for ISOLATE_BOTH twice, and I think the previous code layout was superior:

	if (mode != ISOLATE_BOTH && (!PageActive(page) != !mode))
		return ret;

	if (mode != ISOLATE_BOTH && page_is_file_cache(page) != file)
		return ret;

Because it gives us nice places to put a comment explaining what the code is doing, whereas making it a more complex single expression:

	if (mode != ISOLATE_BOTH) {
		if ((PageActive(page) != mode) ||
			(page_is_file_cache(page) != file))
				return ret;
	}

makes clearly commenting each test more difficult.

Yeah, there's no comment there at present. But that's because we suck
- I'm sure someone is working on it ;)

--
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>

Bob Liu

author

2010-04-02 22:25 UTC #2 lore raw

Quoted 30 lines · Andrew Morton

Quoted 15 lines · On Thu, 1 Apr 2010 21:37:35 +0800 Bob Liu <[email protected]> wrote:
PageActive(page) will return int 0 or 1, mode is also int 0 or 1, they are comparible so "not" is unneeded to be sure to boolean values.
I also collected the ISOLATE_BOTH check together.
Signed-off-by Bob Liu <[email protected]>
---
 mm/vmscan.c |   15 +++++----------
 1 files changed, 5 insertions(+), 10 deletions(-)
mm/vmscan.c +0 -0
index e0e5f15..ce9ee85 100644
mm/vmscan.c int __isolate_lru_page(struct page *page, int mode, · line 870
Quoted 9 lines
mm/vmscan.c +0 -5
@@ -862,16 +862,11 @@ int __isolate_lru_page(struct page *page, int mode,
 int file)
 	if (!PageLRU(page))
 		return ret;
 
-	/*
-	 * When checking the active state, we need to be sure we are
-	 * dealing with comparible boolean values.  Take the logical not
-	 * of each.
-	 */
You deleted a spelling mistake too!

Quoted 12 lines

mm/vmscan.c int __isolate_lru_page(struct page *page, int mode, int file) · line 869
Quoted 10 lines
mm/vmscan.c +5 -5
-	if (mode != ISOLATE_BOTH && (!PageActive(page) != !mode))
-		return ret;
-
-	if (mode != ISOLATE_BOTH && page_is_file_cache(page) != file)
-		return ret;
+	if (mode != ISOLATE_BOTH) {
+		if ((PageActive(page) != mode) ||
+			(page_is_file_cache(page) != file))
+				return ret;
+	}
The compiler should be able to avoid testing for ISOLATE_BOTH twice,

Thanks for your kindly reply.
then is the two "not" able to avoid by the compiler ? if yes, this patch is meanless and should be ignore.

Quoted 23 lines

and I think the previous code layout was superior:
	if (mode != ISOLATE_BOTH && (!PageActive(page) != !mode))
		return ret;

	if (mode != ISOLATE_BOTH && page_is_file_cache(page) != file)
		return ret;
Because it gives us nice places to put a comment explaining what the code is doing, whereas making it a more complex single expression:
	if (mode != ISOLATE_BOTH) {
		if ((PageActive(page) != mode) ||
			(page_is_file_cache(page) != file))
				return ret;
	}
makes clearly commenting each test more difficult.
Yeah, there's no comment there at present. But that's because we suck
- I'm sure someone is working on it ;)

Signature · 8 lines

Regards,
--Bob

--
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>

Andrew Morton

maintainer

2010-04-04 15:08 UTC #3 lore raw

Quoted 21 lines · On Sat, 3 Apr 2010 06:25:08 +0800 Bob Liu <[email protected]> wrote:

Quoted 8 lines
mm/vmscan.c int __isolate_lru_page(struct page *page, int mode, int file) · line 869
Quoted 5 lines
mm/vmscan.c +0 -5
-	/*
-	 * When checking the active state, we need to be sure we are
-	 * dealing with comparible boolean values.  Take the logical not
-	 * of each.
-	 */
You deleted a spelling mistake too!
Quoted 12 lines
mm/vmscan.c int __isolate_lru_page(struct page *page, int mode, int file) · line 869
Quoted 10 lines
mm/vmscan.c +5 -5
-	if (mode != ISOLATE_BOTH && (!PageActive(page) != !mode))
-		return ret;
-
-	if (mode != ISOLATE_BOTH && page_is_file_cache(page) != file)
-		return ret;
+	if (mode != ISOLATE_BOTH) {
+		if ((PageActive(page) != mode) ||
+			(page_is_file_cache(page) != file))
+				return ret;
+	}
The compiler should be able to avoid testing for ISOLATE_BOTH twice,

Quoted 3 lines

Thanks for your kindly reply.
then is the two "not" able to avoid by the compiler ? if yes, this patch is meanless and should be ignore.

I very much doubt if the compiler knows that these two variables can only ever have values 0 or 1, so I expect that removing the !'s will reduce text size.

That being said, it wouldn't be a good and maintainable change - one point in using enumerations such as ISOLATE_* is to hide their real values. Adding code which implicitly "knows" that a particular enumerated identifier has a particular underlying value is rather grubby and fragile.

But the code's already doing that.

It's also a bit fragile to assume that a true/false-returning C function (PageActive) will always return 0 or 1. It's a common C idiom for such functions to return 0 or non-zero (not necessarily 1).

So a clean and maintainable implementation of

	if (mode != ISOLATE_BOTH && (!PageActive(page) != !mode))
		return ret;

would be

	if (mode != ISOLATE_BOTH &&
			((PageActive(page) && mode == ISOLATE_ACTIVE) ||
			 (!PageActive(page) && mode == ISOLATE_INACTIVE)))
		return ret;

which is just dying for an optimisation trick such as the one which is already there ;)