[PATCH 1/2] folded page table walkers

[PATCH 1/2] folded page table walkers

[Nick Piggin]

[-- Attachment #1: Type: text/plain, Size: 698 bytes --]

(Note these are all obviously just RFCs, until after 2.6.11,
at which time I'll send them off to Andrew if they've proven
OK).

Anyway, here is the patch to fold the page table walkers
nicely for 2 and 3 level implementations... now we are getting
some good reasons why people should convert to the -nop?d.h
headers :)


We cut 35% off lmbench fork time with 2 level UP i386. This
patch gets most of the improvement.

                        plain  +ptwalk +folded +funit-at-a-time for mm/
mm/memory.o text size  10935  10679    9799    9519
lmb fork                 335    327     223     218
lmb exec                1417   1445    1228    1198
lmb shell               5483   5507    5195    5087

[-- Attachment #2: vm-folded-walkers.patch --]
[-- Type: text/plain, Size: 2888 bytes --]




---

 linux-2.6-npiggin/include/asm-generic/pgtable-nopmd.h |    9 +++++++++
 linux-2.6-npiggin/include/asm-generic/pgtable-nopud.h |    8 ++++++++
 linux-2.6-npiggin/include/asm-generic/pgtable.h       |    4 ++++
 3 files changed, 21 insertions(+)

diff -puN include/asm-generic/pgtable.h~vm-folded-walkers include/asm-generic/pgtable.h
--- linux-2.6/include/asm-generic/pgtable.h~vm-folded-walkers	2005-02-18 01:54:50.000000000 +1100
+++ linux-2.6-npiggin/include/asm-generic/pgtable.h	2005-02-18 01:54:50.000000000 +1100
@@ -175,6 +175,7 @@ static inline void ptep_mkdirty(pte_t *p
  *
  * see for_each_pgd
  */
+#ifndef for_each_pud
 #define for_each_pud(pgd, start, end, pud, pud_start, pud_end)		\
 	for (	pud = pud_offset(pgd, start),				\
 		  pud_start = start;					\
@@ -183,12 +184,14 @@ static inline void ptep_mkdirty(pte_t *p
 		  pud <= pud_offset(pgd, end-1);			\
 		pud_start = pud_end,					\
 		  pud++ )
+#endif
 
 /*
  * for_each_pmd - iterate through pmd entries in a given pud
  *
  * see for_each_pgd
  */
+#ifndef for_each_pmd
 #define for_each_pmd(pud, start, end, pmd, pmd_start, pmd_end)		\
 	for (	pmd = pmd_offset(pud, start),				\
 		  pmd_start = start;					\
@@ -197,6 +200,7 @@ static inline void ptep_mkdirty(pte_t *p
 		  pmd <= pmd_offset(pud, end-1);			\
 		pmd_start = pmd_end,					\
 		  pmd++ )
+#endif
 
 /*
  * for_each_pte_map - iterate through pte entries in a given pmd
diff -puN include/asm-generic/pgtable-nopud.h~vm-folded-walkers include/asm-generic/pgtable-nopud.h
--- linux-2.6/include/asm-generic/pgtable-nopud.h~vm-folded-walkers	2005-02-18 01:54:50.000000000 +1100
+++ linux-2.6-npiggin/include/asm-generic/pgtable-nopud.h	2005-02-18 02:22:43.000000000 +1100
@@ -52,5 +52,13 @@ static inline pud_t * pud_offset(pgd_t *
 #define pud_free(x)				do { } while (0)
 #define __pud_free_tlb(tlb, x)			do { } while (0)
 
+#undef for_each_pud
+#define for_each_pud(pgd, start, end, pud, pud_start, pud_end)	\
+	for (	pud = (pud_t *)pgd,				\
+		  pud_start = start,				\
+		  pud_end = end;				\
+		pud_start == start;				\
+		pud_start++ )
+
 #endif /* __ASSEMBLY__ */
 #endif /* _PGTABLE_NOPUD_H */
diff -puN include/asm-generic/pgtable-nopmd.h~vm-folded-walkers include/asm-generic/pgtable-nopmd.h
--- linux-2.6/include/asm-generic/pgtable-nopmd.h~vm-folded-walkers	2005-02-18 01:54:50.000000000 +1100
+++ linux-2.6-npiggin/include/asm-generic/pgtable-nopmd.h	2005-02-18 02:22:43.000000000 +1100
@@ -55,6 +55,15 @@ static inline pmd_t * pmd_offset(pud_t *
 #define pmd_free(x)				do { } while (0)
 #define __pmd_free_tlb(tlb, x)			do { } while (0)
 
+#undef for_each_pmd
+#define for_each_pmd(pud, start, end, pmd, pmd_start, pmd_end)	\
+	for (	pmd = (pmd_t *)pud,				\
+		  pmd_start = start,				\
+		  pmd_end = end;				\
+		pmd_start == start;				\
+		pmd_start++ )
+
+
 #endif /* __ASSEMBLY__ */
 
 #endif /* _PGTABLE_NOPMD_H */

_

[PATCH 2/2] trim unused functions

[Nick Piggin]

[-- Attachment #1: Type: text/plain, Size: 95 bytes --]

And lastly, redo this vital optimisation that David
had to remove earlier. Saves a few bytes.


[-- Attachment #2: vm-trim-unused.patch --]
[-- Type: text/plain, Size: 2206 bytes --]




---

 linux-2.6-npiggin/include/asm-generic/pgtable-nopmd.h |    2 ++
 linux-2.6-npiggin/include/asm-generic/pgtable-nopud.h |    2 ++
 linux-2.6-npiggin/mm/memory.c                         |    8 +++++++-
 3 files changed, 11 insertions(+), 1 deletion(-)

diff -puN include/asm-generic/pgtable-nopud.h~vm-trim-unused include/asm-generic/pgtable-nopud.h
--- linux-2.6/include/asm-generic/pgtable-nopud.h~vm-trim-unused	2005-02-18 02:14:30.000000000 +1100
+++ linux-2.6-npiggin/include/asm-generic/pgtable-nopud.h	2005-02-18 02:15:41.000000000 +1100
@@ -3,6 +3,8 @@
 
 #ifndef __ASSEMBLY__
 
+#define __PAGETABLE_PUD_FOLDED
+
 /*
  * Having the pud type consist of a pgd gets the size right, and allows
  * us to conceptually access the pgd entry that this pud is folded into
diff -puN include/asm-generic/pgtable-nopmd.h~vm-trim-unused include/asm-generic/pgtable-nopmd.h
--- linux-2.6/include/asm-generic/pgtable-nopmd.h~vm-trim-unused	2005-02-18 02:14:32.000000000 +1100
+++ linux-2.6-npiggin/include/asm-generic/pgtable-nopmd.h	2005-02-18 02:15:31.000000000 +1100
@@ -5,6 +5,8 @@
 
 #include <asm-generic/pgtable-nopud.h>
 
+#define __PAGETABLE_PMD_FOLDED
+
 /*
  * Having the pmd type consist of a pud gets the size right, and allows
  * us to conceptually access the pud entry that this pmd is folded into
diff -puN mm/memory.c~vm-trim-unused mm/memory.c
--- linux-2.6/mm/memory.c~vm-trim-unused	2005-02-18 02:14:44.000000000 +1100
+++ linux-2.6-npiggin/mm/memory.c	2005-02-18 02:17:34.000000000 +1100
@@ -2006,6 +2006,8 @@ int handle_mm_fault(struct mm_struct *mm
 }
 
 #ifndef __ARCH_HAS_4LEVEL_HACK
+
+#ifndef __PAGETABLE_PUD_FOLDED
 /*
  * Allocate page upper directory.
  *
@@ -2037,7 +2039,9 @@ pud_t fastcall *__pud_alloc(struct mm_st
  out:
 	return pud_offset(pgd, address);
 }
+#endif /* __PAGETABLE_PUD_FOLDED */
 
+#ifndef __PAGETABLE_PMD_FOLDED
 /*
  * Allocate page middle directory.
  *
@@ -2069,7 +2073,9 @@ pmd_t fastcall *__pmd_alloc(struct mm_st
  out:
 	return pmd_offset(pud, address);
 }
-#else
+#endif /* __PAGETABLE_PMD_FOLDED */
+
+#else /* __ARCH_HAS_4LEVEL_HACK */
 pmd_t fastcall *__pmd_alloc(struct mm_struct *mm, pud_t *pud, unsigned long address)
 {
 	pmd_t *new;

_