explicitly document overloaded page flags V2

With the recent page flag reorganisation we have a single enum which defines the valid page flags and their values, nice and clear. However there are a number of bits which are overloaded by different subsystems. Firstly there is PG_owner_priv_1 which is used by filesystems and by XEN. Secondly both SLOB and SLUB use a couple of extra page bits to manage internal state for pages they own; both overlay other bits. All of these "aliases" are scattered about the source making it very hard for a reader to know if the bits are safe to rely on in all contexts; confusion here is bad.

As we now have a single place where the bits are clearly assigned it makes sense to clarify the reuse of bits by making the aliases explicit and visible with the original bit assignments. This patch creates explicit aliases within the enum itself for the overloaded bits, creates standard bit accessors PageFoo etc. and uses those throughout.

This version pulls the bit manipulation out to standard named page bit accessors as suggested by Christoph, it retains the explicit mapping to the overlayed bits. A fusion of both ideas. This has been SLUB and SLOB have been compile tested on x86_64 only, and SLUB boot tested. If people feel this is worth doing then I can run a fuller set of testing.

-apw

--
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>

Some page flags are used for more than one purpose, for example PG_owner_priv_1. Currently there are individual accessors for each user, each built using the common flag name far away from the bit definitions. This makes it hard to see all possible uses of these bits.

Now that we have a single enum to generate the bit orders it makes sense to express overlays in the same place. So create per use aliases for this bit in the main page-flags enum and use those in the accessors.

---
 include/linux/page-flags.h |   12 +++++++++---
 1 files changed, 9 insertions(+), 3 deletions(-)

index 590cff3..2cc1fb1 100644
@@ -96,7 +96,13 @@ enum pageflags {
 #ifdef CONFIG_IA64_UNCACHED_ALLOCATOR
 	PG_uncached,		/* Page has been mapped as uncached */
 #endif
-	__NR_PAGEFLAGS
+	__NR_PAGEFLAGS,
+
+	/* Filesystems */
+	PG_checked = PG_owner_priv_1,
+
+	/* XEN */
+	PG_pinned = PG_owner_priv_1,
 };
 
 #ifndef __GENERATING_BOUNDS_H

@@ -155,8 +161,8 @@ PAGEFLAG(Dirty, dirty) TESTSCFLAG(Dirty, dirty) __CLEARPAGEFLAG(Dirty, dirty)
 PAGEFLAG(LRU, lru) __CLEARPAGEFLAG(LRU, lru)
 PAGEFLAG(Active, active) __CLEARPAGEFLAG(Active, active)
 __PAGEFLAG(Slab, slab)
-PAGEFLAG(Checked, owner_priv_1)		/* Used by some filesystems */
-PAGEFLAG(Pinned, owner_priv_1) TESTSCFLAG(Pinned, owner_priv_1) /* Xen */
+PAGEFLAG(Checked, checked)		/* Used by some filesystems */
+PAGEFLAG(Pinned, pinned) TESTSCFLAG(Pinned, pinned) /* Xen */
 PAGEFLAG(Reserved, reserved) __CLEARPAGEFLAG(Reserved, reserved)
 PAGEFLAG(Private, private) __CLEARPAGEFLAG(Private, private)
 	__SETPAGEFLAG(Private, private)

--
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>

SLUB reuses two page bits for internal purposes, it overlays PG_active and PG_error. This is hidden away in slub.c. Document these overlays explicitly in the main page-flags enum along with all the others.

---
 include/linux/page-flags.h |    7 +++++
 mm/slub.c                  |   65 +++++++++++--------------------------------
 2 files changed, 24 insertions(+), 48 deletions(-)

index 2cc1fb1..dfd0a26 100644
@@ -103,6 +103,10 @@ enum pageflags {
 
 	/* XEN */
 	PG_pinned = PG_owner_priv_1,
+
+	/* SLUB */
+	PG_slub_frozen = PG_active,
+	PG_slub_debug = PG_error,
 };
 
 #ifndef __GENERATING_BOUNDS_H

@@ -167,6 +171,9 @@ PAGEFLAG(Reserved, reserved) __CLEARPAGEFLAG(Reserved, reserved)
 PAGEFLAG(Private, private) __CLEARPAGEFLAG(Private, private)
 	__SETPAGEFLAG(Private, private)
 
+__PAGEFLAG(SlubFrozen, slub_frozen)
+__PAGEFLAG(SlubDebug, slub_debug)
+
 /*
  * Only test-and-set exist for PG_writeback.  The unconditional operators are
  * risky: they bypass page accounting.

--
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>

SLOB reuses two page bits for internal purposes, it overlays PG_active and PG_private. This is hidden away in slob.c. Document these overlays explicitly in the main page-flags enum along with all the others.

---
 include/linux/page-flags.h |    7 +++++++
 mm/slob.c                  |   12 ++++++------
 2 files changed, 13 insertions(+), 6 deletions(-)

index dfd0a26..43b3598 100644
@@ -104,6 +104,10 @@ enum pageflags {
 	/* XEN */
 	PG_pinned = PG_owner_priv_1,
 
+	/* SLOB */
+	PG_slob_page = PG_active,
+	PG_slob_free = PG_private,
+
 	/* SLUB */
 	PG_slub_frozen = PG_active,
 	PG_slub_debug = PG_error,

@@ -171,6 +175,9 @@ PAGEFLAG(Reserved, reserved) __CLEARPAGEFLAG(Reserved, reserved)
 PAGEFLAG(Private, private) __CLEARPAGEFLAG(Private, private)
 	__SETPAGEFLAG(Private, private)
 
+__PAGEFLAG(SlobPage, slob_page)
+__PAGEFLAG(SlobFree, slob_free)
+
 __PAGEFLAG(SlubFrozen, slub_frozen)
 __PAGEFLAG(SlubDebug, slub_debug)
 

index 6038cba..2d18d89 100644
@@ -130,17 +130,17 @@ static LIST_HEAD(free_slob_large);
  */
 static inline int slob_page(struct slob_page *sp)
 {
-	return test_bit(PG_active, &sp->flags);
+	return PageSlobPage((struct page *)sp);
 }
 
 static inline void set_slob_page(struct slob_page *sp)
 {
-	__set_bit(PG_active, &sp->flags);
+	__SetPageSlobPage((struct page *)sp);
 }
 
 static inline void clear_slob_page(struct slob_page *sp)
 {
-	__clear_bit(PG_active, &sp->flags);
+	__ClearPageSlobPage((struct page *)sp);
 }
 
 /*

@@ -148,19 +148,19 @@ static inline void clear_slob_page(struct slob_page *sp)
  */
 static inline int slob_page_free(struct slob_page *sp)
 {
-	return test_bit(PG_private, &sp->flags);
+	return PageSlobFree((struct page *)sp);
 }
 
 static void set_slob_page_free(struct slob_page *sp, struct list_head *list)
 {
 	list_add(&sp->list, list);
-	__set_bit(PG_private, &sp->flags);
+	__SetPageSlobFree((struct page *)sp);
 }
 
 static inline void clear_slob_page_free(struct slob_page *sp)
 {
 	list_del(&sp->list);
-	__clear_bit(PG_private, &sp->flags);
+	__ClearPageSlobFree((struct page *)sp);
 }
 
 #define SLOB_UNIT sizeof(slob_t)

--
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>

Looks good.

If we want to do the bit aliasing at the PG_xxx layer then there may be a way to simplify the page flag generation function because the xxx in PG_xxx is always the lower case of the PageXxx case. So we could remove one parameter if we can find some kind of cpp function that can lowercase a text.

--
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>

Thanks, I like this style of page-flags definition.

BTW, I have a quiestion as crash-dump user. With this 'enum' style, position of each flags in page->flags depends on configs. Can we know what a bit means from dump or bad_page()'s message ? (not a big problem now but..)

Thanks,
-Kame

Hi

Thank you nice patch.

My review found no bug.

--
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>

Hi

This patch works well on my box.
but I have one question.

I dont have SLOB box.
but My review found no bug.

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>

They were unconditional before as well. SlabDebug would always return 0 before the patch. The point being, to my reading, that if you asked for debug on the slab and debug was not compiled in you would still get told that it was not set; which it cannot without the support.

-apw

--
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>

Thank you explain!

--
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>