thp: add x86 32bit support
Johannes Weiner [Thu, 13 Jan 2011 23:47:01 +0000 (15:47 -0800)]
Add support for transparent hugepages to x86 32bit.

Share the same VM_ bitflag for VM_MAPPED_COPY.  mm/nommu.c will never
support transparent hugepages.

Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Signed-off-by: Andrea Arcangeli <aarcange@redhat.com>
Reviewed-by: Rik van Riel <riel@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

arch/x86/include/asm/pgtable-2level.h
arch/x86/include/asm/pgtable-3level.h
arch/x86/include/asm/pgtable.h
arch/x86/include/asm/pgtable_64.h
arch/x86/mm/pgtable.c
include/linux/mm.h
mm/Kconfig

index 2334982..98391db 100644 (file)
@@ -46,6 +46,15 @@ static inline pte_t native_ptep_get_and_clear(pte_t *xp)
 #define native_ptep_get_and_clear(xp) native_local_ptep_get_and_clear(xp)
 #endif
 
+#ifdef CONFIG_SMP
+static inline pmd_t native_pmdp_get_and_clear(pmd_t *xp)
+{
+       return __pmd(xchg((pmdval_t *)xp, 0));
+}
+#else
+#define native_pmdp_get_and_clear(xp) native_local_pmdp_get_and_clear(xp)
+#endif
+
 /*
  * Bits _PAGE_BIT_PRESENT, _PAGE_BIT_FILE and _PAGE_BIT_PROTNONE are taken,
  * split up the 29 bits of offset into this range:
index 177b016..94b979d 100644 (file)
@@ -104,6 +104,29 @@ static inline pte_t native_ptep_get_and_clear(pte_t *ptep)
 #define native_ptep_get_and_clear(xp) native_local_ptep_get_and_clear(xp)
 #endif
 
+#ifdef CONFIG_SMP
+union split_pmd {
+       struct {
+               u32 pmd_low;
+               u32 pmd_high;
+       };
+       pmd_t pmd;
+};
+static inline pmd_t native_pmdp_get_and_clear(pmd_t *pmdp)
+{
+       union split_pmd res, *orig = (union split_pmd *)pmdp;
+
+       /* xchg acts as a barrier before setting of the high bits */
+       res.pmd_low = xchg(&orig->pmd_low, 0);
+       res.pmd_high = orig->pmd_high;
+       orig->pmd_high = 0;
+
+       return res.pmd;
+}
+#else
+#define native_pmdp_get_and_clear(xp) native_local_pmdp_get_and_clear(xp)
+#endif
+
 /*
  * Bits 0, 6 and 7 are taken in the low part of the pte,
  * put the 32 bits of offset into the high part.
index 3278038..001a383 100644 (file)
@@ -97,6 +97,11 @@ static inline int pte_young(pte_t pte)
        return pte_flags(pte) & _PAGE_ACCESSED;
 }
 
+static inline int pmd_young(pmd_t pmd)
+{
+       return pmd_flags(pmd) & _PAGE_ACCESSED;
+}
+
 static inline int pte_write(pte_t pte)
 {
        return pte_flags(pte) & _PAGE_RW;
@@ -145,6 +150,18 @@ static inline int pmd_large(pmd_t pte)
                (_PAGE_PSE | _PAGE_PRESENT);
 }
 
+#ifdef CONFIG_TRANSPARENT_HUGEPAGE
+static inline int pmd_trans_splitting(pmd_t pmd)
+{
+       return pmd_val(pmd) & _PAGE_SPLITTING;
+}
+
+static inline int pmd_trans_huge(pmd_t pmd)
+{
+       return pmd_val(pmd) & _PAGE_PSE;
+}
+#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
+
 static inline pte_t pte_set_flags(pte_t pte, pteval_t set)
 {
        pteval_t v = native_pte_val(pte);
@@ -219,6 +236,55 @@ static inline pte_t pte_mkspecial(pte_t pte)
        return pte_set_flags(pte, _PAGE_SPECIAL);
 }
 
+static inline pmd_t pmd_set_flags(pmd_t pmd, pmdval_t set)
+{
+       pmdval_t v = native_pmd_val(pmd);
+
+       return __pmd(v | set);
+}
+
+static inline pmd_t pmd_clear_flags(pmd_t pmd, pmdval_t clear)
+{
+       pmdval_t v = native_pmd_val(pmd);
+
+       return __pmd(v & ~clear);
+}
+
+static inline pmd_t pmd_mkold(pmd_t pmd)
+{
+       return pmd_clear_flags(pmd, _PAGE_ACCESSED);
+}
+
+static inline pmd_t pmd_wrprotect(pmd_t pmd)
+{
+       return pmd_clear_flags(pmd, _PAGE_RW);
+}
+
+static inline pmd_t pmd_mkdirty(pmd_t pmd)
+{
+       return pmd_set_flags(pmd, _PAGE_DIRTY);
+}
+
+static inline pmd_t pmd_mkhuge(pmd_t pmd)
+{
+       return pmd_set_flags(pmd, _PAGE_PSE);
+}
+
+static inline pmd_t pmd_mkyoung(pmd_t pmd)
+{
+       return pmd_set_flags(pmd, _PAGE_ACCESSED);
+}
+
+static inline pmd_t pmd_mkwrite(pmd_t pmd)
+{
+       return pmd_set_flags(pmd, _PAGE_RW);
+}
+
+static inline pmd_t pmd_mknotpresent(pmd_t pmd)
+{
+       return pmd_clear_flags(pmd, _PAGE_PRESENT);
+}
+
 /*
  * Mask out unsupported bits in a present pgprot.  Non-present pgprots
  * can use those bits for other purposes, so leave them be.
@@ -527,6 +593,14 @@ static inline pte_t native_local_ptep_get_and_clear(pte_t *ptep)
        return res;
 }
 
+static inline pmd_t native_local_pmdp_get_and_clear(pmd_t *pmdp)
+{
+       pmd_t res = *pmdp;
+
+       native_pmd_clear(pmdp);
+       return res;
+}
+
 static inline void native_set_pte_at(struct mm_struct *mm, unsigned long addr,
                                     pte_t *ptep , pte_t pte)
 {
@@ -616,6 +690,49 @@ static inline void ptep_set_wrprotect(struct mm_struct *mm,
 
 #define flush_tlb_fix_spurious_fault(vma, address)
 
+#define mk_pmd(page, pgprot)   pfn_pmd(page_to_pfn(page), (pgprot))
+
+#define  __HAVE_ARCH_PMDP_SET_ACCESS_FLAGS
+extern int pmdp_set_access_flags(struct vm_area_struct *vma,
+                                unsigned long address, pmd_t *pmdp,
+                                pmd_t entry, int dirty);
+
+#define __HAVE_ARCH_PMDP_TEST_AND_CLEAR_YOUNG
+extern int pmdp_test_and_clear_young(struct vm_area_struct *vma,
+                                    unsigned long addr, pmd_t *pmdp);
+
+#define __HAVE_ARCH_PMDP_CLEAR_YOUNG_FLUSH
+extern int pmdp_clear_flush_young(struct vm_area_struct *vma,
+                                 unsigned long address, pmd_t *pmdp);
+
+
+#define __HAVE_ARCH_PMDP_SPLITTING_FLUSH
+extern void pmdp_splitting_flush(struct vm_area_struct *vma,
+                                unsigned long addr, pmd_t *pmdp);
+
+#define __HAVE_ARCH_PMD_WRITE
+static inline int pmd_write(pmd_t pmd)
+{
+       return pmd_flags(pmd) & _PAGE_RW;
+}
+
+#define __HAVE_ARCH_PMDP_GET_AND_CLEAR
+static inline pmd_t pmdp_get_and_clear(struct mm_struct *mm, unsigned long addr,
+                                      pmd_t *pmdp)
+{
+       pmd_t pmd = native_pmdp_get_and_clear(pmdp);
+       pmd_update(mm, addr, pmdp);
+       return pmd;
+}
+
+#define __HAVE_ARCH_PMDP_SET_WRPROTECT
+static inline void pmdp_set_wrprotect(struct mm_struct *mm,
+                                     unsigned long addr, pmd_t *pmdp)
+{
+       clear_bit(_PAGE_BIT_RW, (unsigned long *)pmdp);
+       pmd_update(mm, addr, pmdp);
+}
+
 /*
  * clone_pgd_range(pgd_t *dst, pgd_t *src, int count);
  *
index b2df039..975f709 100644 (file)
@@ -182,115 +182,6 @@ extern void cleanup_highmap(void);
 
 #define __HAVE_ARCH_PTE_SAME
 
-#ifdef CONFIG_TRANSPARENT_HUGEPAGE
-static inline int pmd_trans_splitting(pmd_t pmd)
-{
-       return pmd_val(pmd) & _PAGE_SPLITTING;
-}
-
-static inline int pmd_trans_huge(pmd_t pmd)
-{
-       return pmd_val(pmd) & _PAGE_PSE;
-}
-#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
-
-#define mk_pmd(page, pgprot)   pfn_pmd(page_to_pfn(page), (pgprot))
-
-#define  __HAVE_ARCH_PMDP_SET_ACCESS_FLAGS
-extern int pmdp_set_access_flags(struct vm_area_struct *vma,
-                                unsigned long address, pmd_t *pmdp,
-                                pmd_t entry, int dirty);
-
-#define __HAVE_ARCH_PMDP_TEST_AND_CLEAR_YOUNG
-extern int pmdp_test_and_clear_young(struct vm_area_struct *vma,
-                                    unsigned long addr, pmd_t *pmdp);
-
-#define __HAVE_ARCH_PMDP_CLEAR_YOUNG_FLUSH
-extern int pmdp_clear_flush_young(struct vm_area_struct *vma,
-                                 unsigned long address, pmd_t *pmdp);
-
-
-#define __HAVE_ARCH_PMDP_SPLITTING_FLUSH
-extern void pmdp_splitting_flush(struct vm_area_struct *vma,
-                                unsigned long addr, pmd_t *pmdp);
-
-#define __HAVE_ARCH_PMD_WRITE
-static inline int pmd_write(pmd_t pmd)
-{
-       return pmd_flags(pmd) & _PAGE_RW;
-}
-
-#define __HAVE_ARCH_PMDP_GET_AND_CLEAR
-static inline pmd_t pmdp_get_and_clear(struct mm_struct *mm, unsigned long addr,
-                                      pmd_t *pmdp)
-{
-       pmd_t pmd = native_pmdp_get_and_clear(pmdp);
-       pmd_update(mm, addr, pmdp);
-       return pmd;
-}
-
-#define __HAVE_ARCH_PMDP_SET_WRPROTECT
-static inline void pmdp_set_wrprotect(struct mm_struct *mm,
-                                     unsigned long addr, pmd_t *pmdp)
-{
-       clear_bit(_PAGE_BIT_RW, (unsigned long *)&pmdp->pmd);
-       pmd_update(mm, addr, pmdp);
-}
-
-static inline int pmd_young(pmd_t pmd)
-{
-       return pmd_flags(pmd) & _PAGE_ACCESSED;
-}
-
-static inline pmd_t pmd_set_flags(pmd_t pmd, pmdval_t set)
-{
-       pmdval_t v = native_pmd_val(pmd);
-
-       return native_make_pmd(v | set);
-}
-
-static inline pmd_t pmd_clear_flags(pmd_t pmd, pmdval_t clear)
-{
-       pmdval_t v = native_pmd_val(pmd);
-
-       return native_make_pmd(v & ~clear);
-}
-
-static inline pmd_t pmd_mkold(pmd_t pmd)
-{
-       return pmd_clear_flags(pmd, _PAGE_ACCESSED);
-}
-
-static inline pmd_t pmd_wrprotect(pmd_t pmd)
-{
-       return pmd_clear_flags(pmd, _PAGE_RW);
-}
-
-static inline pmd_t pmd_mkdirty(pmd_t pmd)
-{
-       return pmd_set_flags(pmd, _PAGE_DIRTY);
-}
-
-static inline pmd_t pmd_mkhuge(pmd_t pmd)
-{
-       return pmd_set_flags(pmd, _PAGE_PSE);
-}
-
-static inline pmd_t pmd_mkyoung(pmd_t pmd)
-{
-       return pmd_set_flags(pmd, _PAGE_ACCESSED);
-}
-
-static inline pmd_t pmd_mkwrite(pmd_t pmd)
-{
-       return pmd_set_flags(pmd, _PAGE_RW);
-}
-
-static inline pmd_t pmd_mknotpresent(pmd_t pmd)
-{
-       return pmd_clear_flags(pmd, _PAGE_PRESENT);
-}
-
 #endif /* !__ASSEMBLY__ */
 
 #endif /* _ASM_X86_PGTABLE_64_H */
index 65e92d5..500242d 100644 (file)
@@ -362,7 +362,7 @@ int pmdp_test_and_clear_young(struct vm_area_struct *vma,
 
        if (pmd_young(*pmdp))
                ret = test_and_clear_bit(_PAGE_BIT_ACCESSED,
-                                        (unsigned long *) &pmdp->pmd);
+                                        (unsigned long *)pmdp);
 
        if (ret)
                pmd_update(vma->vm_mm, addr, pmdp);
@@ -404,7 +404,7 @@ void pmdp_splitting_flush(struct vm_area_struct *vma,
        int set;
        VM_BUG_ON(address & ~HPAGE_PMD_MASK);
        set = !test_and_set_bit(_PAGE_BIT_SPLITTING,
-                               (unsigned long *)&pmdp->pmd);
+                               (unsigned long *)pmdp);
        if (set) {
                pmd_update(vma->vm_mm, address, pmdp);
                /* need tlb flush only to serialize against gup-fast */
index 7ab7d2b..9c2695b 100644 (file)
@@ -102,7 +102,11 @@ extern unsigned int kobjsize(const void *objp);
 #define VM_NORESERVE   0x00200000      /* should the VM suppress accounting */
 #define VM_HUGETLB     0x00400000      /* Huge TLB Page VM */
 #define VM_NONLINEAR   0x00800000      /* Is non-linear (remap_file_pages) */
+#ifndef CONFIG_TRANSPARENT_HUGEPAGE
 #define VM_MAPPED_COPY 0x01000000      /* T if mapped copy of data (nommu mmap) */
+#else
+#define VM_HUGEPAGE    0x01000000      /* MADV_HUGEPAGE marked this vma */
+#endif
 #define VM_INSERTPAGE  0x02000000      /* The vma has had "vm_insert_page()" done on it */
 #define VM_ALWAYSDUMP  0x04000000      /* Always include in core dumps */
 
@@ -111,9 +115,6 @@ extern unsigned int kobjsize(const void *objp);
 #define VM_SAO         0x20000000      /* Strong Access Ordering (powerpc) */
 #define VM_PFN_AT_MMAP 0x40000000      /* PFNMAP vma that is fully mapped at mmap time */
 #define VM_MERGEABLE   0x80000000      /* KSM may merge identical pages */
-#if BITS_PER_LONG > 32
-#define VM_HUGEPAGE    0x100000000UL   /* MADV_HUGEPAGE marked this vma */
-#endif
 
 /* Bits set in the VMA until the stack is in its final location */
 #define VM_STACK_INCOMPLETE_SETUP      (VM_RAND_READ | VM_SEQ_READ)
index 3982be2..d774f77 100644 (file)
@@ -304,7 +304,7 @@ config NOMMU_INITIAL_TRIM_EXCESS
 
 config TRANSPARENT_HUGEPAGE
        bool "Transparent Hugepage Support" if EMBEDDED
-       depends on X86_64 && MMU
+       depends on X86 && MMU
        default y
        help
          Transparent Hugepages allows the kernel to use huge pages and