string: factorize skip_spaces and export it to be generally available
[safe/jmp/linux-2.6] / mm / pagewalk.c
index 0afd238..7b47a57 100644 (file)
@@ -1,16 +1,17 @@
 #include <linux/mm.h>
 #include <linux/highmem.h>
 #include <linux/sched.h>
+#include <linux/hugetlb.h>
 
 static int walk_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
-                         const struct mm_walk *walk, void *private)
+                         struct mm_walk *walk)
 {
        pte_t *pte;
        int err = 0;
 
        pte = pte_offset_map(pmd, addr);
        for (;;) {
-               err = walk->pte_entry(pte, addr, addr + PAGE_SIZE, private);
+               err = walk->pte_entry(pte, addr, addr + PAGE_SIZE, walk);
                if (err)
                       break;
                addr += PAGE_SIZE;
@@ -24,7 +25,7 @@ static int walk_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
 }
 
 static int walk_pmd_range(pud_t *pud, unsigned long addr, unsigned long end,
-                         const struct mm_walk *walk, void *private)
+                         struct mm_walk *walk)
 {
        pmd_t *pmd;
        unsigned long next;
@@ -35,15 +36,15 @@ static int walk_pmd_range(pud_t *pud, unsigned long addr, unsigned long end,
                next = pmd_addr_end(addr, end);
                if (pmd_none_or_clear_bad(pmd)) {
                        if (walk->pte_hole)
-                               err = walk->pte_hole(addr, next, private);
+                               err = walk->pte_hole(addr, next, walk);
                        if (err)
                                break;
                        continue;
                }
                if (walk->pmd_entry)
-                       err = walk->pmd_entry(pmd, addr, next, private);
+                       err = walk->pmd_entry(pmd, addr, next, walk);
                if (!err && walk->pte_entry)
-                       err = walk_pte_range(pmd, addr, next, walk, private);
+                       err = walk_pte_range(pmd, addr, next, walk);
                if (err)
                        break;
        } while (pmd++, addr = next, addr != end);
@@ -52,7 +53,7 @@ static int walk_pmd_range(pud_t *pud, unsigned long addr, unsigned long end,
 }
 
 static int walk_pud_range(pgd_t *pgd, unsigned long addr, unsigned long end,
-                         const struct mm_walk *walk, void *private)
+                         struct mm_walk *walk)
 {
        pud_t *pud;
        unsigned long next;
@@ -63,15 +64,15 @@ static int walk_pud_range(pgd_t *pgd, unsigned long addr, unsigned long end,
                next = pud_addr_end(addr, end);
                if (pud_none_or_clear_bad(pud)) {
                        if (walk->pte_hole)
-                               err = walk->pte_hole(addr, next, private);
+                               err = walk->pte_hole(addr, next, walk);
                        if (err)
                                break;
                        continue;
                }
                if (walk->pud_entry)
-                       err = walk->pud_entry(pud, addr, next, private);
+                       err = walk->pud_entry(pud, addr, next, walk);
                if (!err && (walk->pmd_entry || walk->pte_entry))
-                       err = walk_pmd_range(pud, addr, next, walk, private);
+                       err = walk_pmd_range(pud, addr, next, walk);
                if (err)
                        break;
        } while (pud++, addr = next, addr != end);
@@ -85,15 +86,15 @@ static int walk_pud_range(pgd_t *pgd, unsigned long addr, unsigned long end,
  * @addr: starting address
  * @end: ending address
  * @walk: set of callbacks to invoke for each level of the tree
- * @private: private data passed to the callback function
  *
  * Recursively walk the page table for the memory area in a VMA,
  * calling supplied callbacks. Callbacks are called in-order (first
  * PGD, first PUD, first PMD, first PTE, second PTE... second PMD,
  * etc.). If lower-level callbacks are omitted, walking depth is reduced.
  *
- * Each callback receives an entry pointer, the start and end of the
- * associated range, and a caller-supplied private data pointer.
+ * Each callback receives an entry pointer and the start and end of the
+ * associated range, and a copy of the original mm_walk for access to
+ * the ->private or ->mm fields.
  *
  * No locks are taken, but the bottom level iterator will map PTE
  * directories from highmem if necessary.
@@ -101,35 +102,66 @@ static int walk_pud_range(pgd_t *pgd, unsigned long addr, unsigned long end,
  * If any callback returns a non-zero value, the walk is aborted and
  * the return value is propagated back to the caller. Otherwise 0 is returned.
  */
-int walk_page_range(const struct mm_struct *mm,
-                   unsigned long addr, unsigned long end,
-                   const struct mm_walk *walk, void *private)
+int walk_page_range(unsigned long addr, unsigned long end,
+                   struct mm_walk *walk)
 {
        pgd_t *pgd;
        unsigned long next;
        int err = 0;
+       struct vm_area_struct *vma;
 
        if (addr >= end)
                return err;
 
-       pgd = pgd_offset(mm, addr);
+       if (!walk->mm)
+               return -EINVAL;
+
+       pgd = pgd_offset(walk->mm, addr);
        do {
                next = pgd_addr_end(addr, end);
+
+               /*
+                * handle hugetlb vma individually because pagetable walk for
+                * the hugetlb page is dependent on the architecture and
+                * we can't handled it in the same manner as non-huge pages.
+                */
+               vma = find_vma(walk->mm, addr);
+#ifdef CONFIG_HUGETLB_PAGE
+               if (vma && is_vm_hugetlb_page(vma)) {
+                       pte_t *pte;
+                       struct hstate *hs;
+
+                       if (vma->vm_end < next)
+                               next = vma->vm_end;
+                       hs = hstate_vma(vma);
+                       pte = huge_pte_offset(walk->mm,
+                                             addr & huge_page_mask(hs));
+                       if (pte && !huge_pte_none(huge_ptep_get(pte))
+                           && walk->hugetlb_entry)
+                               err = walk->hugetlb_entry(pte, addr,
+                                                         next, walk);
+                       if (err)
+                               break;
+                       continue;
+               }
+#endif
                if (pgd_none_or_clear_bad(pgd)) {
                        if (walk->pte_hole)
-                               err = walk->pte_hole(addr, next, private);
+                               err = walk->pte_hole(addr, next, walk);
                        if (err)
                                break;
+                       pgd++;
                        continue;
                }
                if (walk->pgd_entry)
-                       err = walk->pgd_entry(pgd, addr, next, private);
+                       err = walk->pgd_entry(pgd, addr, next, walk);
                if (!err &&
                    (walk->pud_entry || walk->pmd_entry || walk->pte_entry))
-                       err = walk_pud_range(pgd, addr, next, walk, private);
+                       err = walk_pud_range(pgd, addr, next, walk);
                if (err)
                        break;
-       } while (pgd++, addr = next, addr != end);
+               pgd++;
+       } while (addr = next, addr != end);
 
        return err;
 }