mm/pagewalk.c

   1 #include <linux/mm.h>
   2 #include <linux/highmem.h>
   3 #include <linux/sched.h>
   4 #include <linux/hugetlb.h>
   5
   6 static int walk_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
   7                           struct mm_walk *walk)
   8 {
   9         pte_t *pte;
  10         int err = 0;
  11
  12         pte = pte_offset_map(pmd, addr);
  13         for (;;) {
  14                 err = walk->pte_entry(pte, addr, addr + PAGE_SIZE, walk);
  15                 if (err)
  16                        break;
  17                 addr += PAGE_SIZE;
  18                 if (addr == end)
  19                         break;
  20                 pte++;
  21         }
  22
  23         pte_unmap(pte);
  24         return err;
  25 }
  26
  27 static int walk_pmd_range(pud_t *pud, unsigned long addr, unsigned long end,
  28                           struct mm_walk *walk)
  29 {
  30         pmd_t *pmd;
  31         unsigned long next;
  32         int err = 0;
  33
  34         pmd = pmd_offset(pud, addr);
  35         do {
  36 again:
  37                 next = pmd_addr_end(addr, end);
  38                 if (pmd_none(*pmd)) {
  39                         if (walk->pte_hole)
  40                                 err = walk->pte_hole(addr, next, walk);
  41                         if (err)
  42                                 break;
  43                         continue;
  44                 }
  45                 /*
  46                  * This implies that each ->pmd_entry() handler
  47                  * needs to know about pmd_trans_huge() pmds
  48                  */
  49                 if (walk->pmd_entry)
  50                         err = walk->pmd_entry(pmd, addr, next, walk);
  51                 if (err)
  52                         break;
  53
  54                 /*
  55                  * Check this here so we only break down trans_huge
  56                  * pages when we _need_ to
  57                  */
  58                 if (!walk->pte_entry)
  59                         continue;
  60
  61                 split_huge_page_pmd(walk->mm, pmd);
  62                 if (pmd_none_or_trans_huge_or_clear_bad(pmd))
  63                         goto again;
  64                 err = walk_pte_range(pmd, addr, next, walk);
  65                 if (err)
  66                         break;
  67         } while (pmd++, addr = next, addr != end);
  68
  69         return err;
  70 }
  71
  72 static int walk_pud_range(pgd_t *pgd, unsigned long addr, unsigned long end,
  73                           struct mm_walk *walk)
  74 {
  75         pud_t *pud;
  76         unsigned long next;
  77         int err = 0;
  78
  79         pud = pud_offset(pgd, addr);
  80         do {
  81                 next = pud_addr_end(addr, end);
  82                 if (pud_none_or_clear_bad(pud)) {
  83                         if (walk->pte_hole)
  84                                 err = walk->pte_hole(addr, next, walk);
  85                         if (err)
  86                                 break;
  87                         continue;
  88                 }
  89                 if (walk->pud_entry)
  90                         err = walk->pud_entry(pud, addr, next, walk);
  91                 if (!err && (walk->pmd_entry || walk->pte_entry))
  92                         err = walk_pmd_range(pud, addr, next, walk);
  93                 if (err)
  94                         break;
  95         } while (pud++, addr = next, addr != end);
  96
  97         return err;
  98 }
  99
 100 #ifdef CONFIG_HUGETLB_PAGE
 101 static unsigned long hugetlb_entry_end(struct hstate *h, unsigned long addr,
 102                                        unsigned long end)
 103 {
 104         unsigned long boundary = (addr & huge_page_mask(h)) + huge_page_size(h);
 105         return boundary < end ? boundary : end;
 106 }
 107
 108 static int walk_hugetlb_range(struct vm_area_struct *vma,
 109                               unsigned long addr, unsigned long end,
 110                               struct mm_walk *walk)
 111 {
 112         struct hstate *h = hstate_vma(vma);
 113         unsigned long next;
 114         unsigned long hmask = huge_page_mask(h);
 115         pte_t *pte;
 116         int err = 0;
 117
 118         do {
 119                 next = hugetlb_entry_end(h, addr, end);
 120                 pte = huge_pte_offset(walk->mm, addr & hmask);
 121                 if (pte && walk->hugetlb_entry)
 122                         err = walk->hugetlb_entry(pte, hmask, addr, next, walk);
 123                 if (err)
 124                         return err;
 125         } while (addr = next, addr != end);
 126
 127         return 0;
 128 }
 129 #endif
 130
 131 /**
 132  * walk_page_range - walk a memory map's page tables with a callback
 133  * @mm: memory map to walk
 134  * @addr: starting address
 135  * @end: ending address
 136  * @walk: set of callbacks to invoke for each level of the tree
 137  *
 138  * Recursively walk the page table for the memory area in a VMA,
 139  * calling supplied callbacks. Callbacks are called in-order (first
 140  * PGD, first PUD, first PMD, first PTE, second PTE... second PMD,
 141  * etc.). If lower-level callbacks are omitted, walking depth is reduced.
 142  *
 143  * Each callback receives an entry pointer and the start and end of the
 144  * associated range, and a copy of the original mm_walk for access to
 145  * the ->private or ->mm fields.
 146  *
 147  * No locks are taken, but the bottom level iterator will map PTE
 148  * directories from highmem if necessary.
 149  *
 150  * If any callback returns a non-zero value, the walk is aborted and
 151  * the return value is propagated back to the caller. Otherwise 0 is returned.
 152  */
 153 int walk_page_range(unsigned long addr, unsigned long end,
 154                     struct mm_walk *walk)
 155 {
 156         pgd_t *pgd;
 157         unsigned long next;
 158         int err = 0;
 159
 160         if (addr >= end)
 161                 return err;
 162
 163         if (!walk->mm)
 164                 return -EINVAL;
 165
 166         pgd = pgd_offset(walk->mm, addr);
 167         do {
 168                 struct vm_area_struct *uninitialized_var(vma);
 169
 170                 next = pgd_addr_end(addr, end);
 171
 172 #ifdef CONFIG_HUGETLB_PAGE
 173                 /*
 174                  * handle hugetlb vma individually because pagetable walk for
 175                  * the hugetlb page is dependent on the architecture and
 176                  * we can't handled it in the same manner as non-huge pages.
 177                  */
 178                 vma = find_vma(walk->mm, addr);
 179                 if (vma && is_vm_hugetlb_page(vma)) {
 180                         if (vma->vm_end < next)
 181                                 next = vma->vm_end;
 182                         /*
 183                          * Hugepage is very tightly coupled with vma, so
 184                          * walk through hugetlb entries within a given vma.
 185                          */
 186                         err = walk_hugetlb_range(vma, addr, next, walk);
 187                         if (err)
 188                                 break;
 189                         pgd = pgd_offset(walk->mm, next);
 190                         continue;
 191                 }
 192 #endif
 193                 if (pgd_none_or_clear_bad(pgd)) {
 194                         if (walk->pte_hole)
 195                                 err = walk->pte_hole(addr, next, walk);
 196                         if (err)
 197                                 break;
 198                         pgd++;
 199                         continue;
 200                 }
 201                 if (walk->pgd_entry)
 202                         err = walk->pgd_entry(pgd, addr, next, walk);
 203                 if (!err &&
 204                     (walk->pud_entry || walk->pmd_entry || walk->pte_entry))
 205                         err = walk_pud_range(pgd, addr, next, walk);
 206                 if (err)
 207                         break;
 208                 pgd++;
 209         } while (addr = next, addr != end);
 210
 211         return err;
 212 }