LXR linux/mm/pagewalk.c

   1#include <linux/mm.h>
   2#include <linux/highmem.h>
   3#include <linux/sched.h>
   4#include <linux/hugetlb.h>
   5
   6static 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
  27static 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 {
  36again:
  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_mm(walk->mm, addr, 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
  72static 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
 101static 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
 108static 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
 130#else /* CONFIG_HUGETLB_PAGE */
 131static int walk_hugetlb_range(struct vm_area_struct *vma,
 132                              unsigned long addr, unsigned long end,
 133                              struct mm_walk *walk)
 134{
 135        return 0;
 136}
 137
 138#endif /* CONFIG_HUGETLB_PAGE */
 139
 140
 141
 142/**
 143 * walk_page_range - walk a memory map's page tables with a callback
 144 * @addr: starting address
 145 * @end: ending address
 146 * @walk: set of callbacks to invoke for each level of the tree
 147 *
 148 * Recursively walk the page table for the memory area in a VMA,
 149 * calling supplied callbacks. Callbacks are called in-order (first
 150 * PGD, first PUD, first PMD, first PTE, second PTE... second PMD,
 151 * etc.). If lower-level callbacks are omitted, walking depth is reduced.
 152 *
 153 * Each callback receives an entry pointer and the start and end of the
 154 * associated range, and a copy of the original mm_walk for access to
 155 * the ->private or ->mm fields.
 156 *
 157 * Usually no locks are taken, but splitting transparent huge page may
 158 * take page table lock. And the bottom level iterator will map PTE
 159 * directories from highmem if necessary.
 160 *
 161 * If any callback returns a non-zero value, the walk is aborted and
 162 * the return value is propagated back to the caller. Otherwise 0 is returned.
 163 *
 164 * walk->mm->mmap_sem must be held for at least read if walk->hugetlb_entry
 165 * is !NULL.
 166 */
 167int walk_page_range(unsigned long addr, unsigned long end,
 168                    struct mm_walk *walk)
 169{
 170        pgd_t *pgd;
 171        unsigned long next;
 172        int err = 0;
 173
 174        if (addr >= end)
 175                return err;
 176
 177        if (!walk->mm)
 178                return -EINVAL;
 179
 180        VM_BUG_ON(!rwsem_is_locked(&walk->mm->mmap_sem));
 181
 182        pgd = pgd_offset(walk->mm, addr);
 183        do {
 184                struct vm_area_struct *vma = NULL;
 185
 186                next = pgd_addr_end(addr, end);
 187
 188                /*
 189                 * This function was not intended to be vma based.
 190                 * But there are vma special cases to be handled:
 191                 * - hugetlb vma's
 192                 * - VM_PFNMAP vma's
 193                 */
 194                vma = find_vma(walk->mm, addr);
 195                if (vma) {
 196                        /*
 197                         * There are no page structures backing a VM_PFNMAP
 198                         * range, so do not allow split_huge_page_pmd().
 199                         */
 200                        if ((vma->vm_start <= addr) &&
 201                            (vma->vm_flags & VM_PFNMAP)) {
 202                                next = vma->vm_end;
 203                                pgd = pgd_offset(walk->mm, next);
 204                                continue;
 205                        }
 206                        /*
 207                         * Handle hugetlb vma individually because pagetable
 208                         * walk for the hugetlb page is dependent on the
 209                         * architecture and we can't handled it in the same
 210                         * manner as non-huge pages.
 211                         */
 212                        if (walk->hugetlb_entry && (vma->vm_start <= addr) &&
 213                            is_vm_hugetlb_page(vma)) {
 214                                if (vma->vm_end < next)
 215                                        next = vma->vm_end;
 216                                /*
 217                                 * Hugepage is very tightly coupled with vma,
 218                                 * so walk through hugetlb entries within a
 219                                 * given vma.
 220                                 */
 221                                err = walk_hugetlb_range(vma, addr, next, walk);
 222                                if (err)
 223                                        break;
 224                                pgd = pgd_offset(walk->mm, next);
 225                                continue;
 226                        }
 227                }
 228
 229                if (pgd_none_or_clear_bad(pgd)) {
 230                        if (walk->pte_hole)
 231                                err = walk->pte_hole(addr, next, walk);
 232                        if (err)
 233                                break;
 234                        pgd++;
 235                        continue;
 236                }
 237                if (walk->pgd_entry)
 238                        err = walk->pgd_entry(pgd, addr, next, walk);
 239                if (!err &&
 240                    (walk->pud_entry || walk->pmd_entry || walk->pte_entry))
 241                        err = walk_pud_range(pgd, addr, next, walk);
 242                if (err)
 243                        break;
 244                pgd++;
 245        } while (addr = next, addr != end);
 246
 247        return err;
 248}
 249