1 /* SPDX-License-Identifier: GPL-2.0 */ 1 /* SPDX-License-Identifier: GPL-2.0 */ 2 #ifndef _LINUX_PAGEWALK_H 2 #ifndef _LINUX_PAGEWALK_H 3 #define _LINUX_PAGEWALK_H 3 #define _LINUX_PAGEWALK_H 4 4 5 #include <linux/mm.h> 5 #include <linux/mm.h> 6 6 7 struct mm_walk; 7 struct mm_walk; 8 8 9 /* Locking requirement during a page walk. */ << 10 enum page_walk_lock { << 11 /* mmap_lock should be locked for read << 12 PGWALK_RDLOCK = 0, << 13 /* vma will be write-locked during the << 14 PGWALK_WRLOCK = 1, << 15 /* vma is expected to be already write << 16 PGWALK_WRLOCK_VERIFY = 2, << 17 }; << 18 << 19 /** 9 /** 20 * struct mm_walk_ops - callbacks for walk_pag !! 10 * mm_walk_ops - callbacks for walk_page_range 21 * @pgd_entry: if set, called for eac !! 11 * @pud_entry: if set, called for each non-empty PUD (2nd-level) entry 22 * @p4d_entry: if set, called for eac !! 12 * this handler should only handle pud_trans_huge() puds. 23 * @pud_entry: if set, called for eac !! 13 * the pmd_entry or pte_entry callbacks will be used for 24 * @pmd_entry: if set, called for eac !! 14 * regular PUDs. >> 15 * @pmd_entry: if set, called for each non-empty PMD (3rd-level) entry 25 * this handler is requir 16 * this handler is required to be able to handle 26 * pmd_trans_huge() pmds. 17 * pmd_trans_huge() pmds. They may simply choose to 27 * split_huge_page() inst 18 * split_huge_page() instead of handling it explicitly. 28 * @pte_entry: if set, called for eac !! 19 * @pte_entry: if set, called for each non-empty PTE (4th-level) entry 29 * including empty ones !! 20 * @pte_hole: if set, called for each hole at all levels 30 * @pte_hole: if set, called for eac !! 21 * @hugetlb_entry: if set, called for each hugetlb entry 31 * depth is -1 if not kno << 32 * Any folded depths (whe << 33 * are skipped. << 34 * @hugetlb_entry: if set, called for eac << 35 * function is called wit << 36 * protect against a conc << 37 * the ptl. In some cases << 38 * and retake the vma loc << 39 * while calling other fu << 40 * function must either r << 41 * ptl after dropping the << 42 * those items after re-a << 43 * accessing them. << 44 * @test_walk: caller specific callba 22 * @test_walk: caller specific callback function to determine whether 45 * we walk over the curre 23 * we walk over the current vma or not. Returning 0 means 46 * "do page table walk ov 24 * "do page table walk over the current vma", returning 47 * a negative value means 25 * a negative value means "abort current page table walk 48 * right now" and returni 26 * right now" and returning 1 means "skip the current vma" 49 * Note that this callbac << 50 * passes in a single VMA << 51 * @pre_vma: if set, called before 27 * @pre_vma: if set, called before starting walk on a non-null vma. 52 * @post_vma: if set, called after a 28 * @post_vma: if set, called after a walk on a non-null vma, provided 53 * that @pre_vma and the 29 * that @pre_vma and the vma walk succeeded. 54 * << 55 * p?d_entry callbacks are called even if thos << 56 * particular architecture/configuration. << 57 */ 30 */ 58 struct mm_walk_ops { 31 struct mm_walk_ops { 59 int (*pgd_entry)(pgd_t *pgd, unsigned << 60 unsigned long next, s << 61 int (*p4d_entry)(p4d_t *p4d, unsigned << 62 unsigned long next, s << 63 int (*pud_entry)(pud_t *pud, unsigned 32 int (*pud_entry)(pud_t *pud, unsigned long addr, 64 unsigned long next, s 33 unsigned long next, struct mm_walk *walk); 65 int (*pmd_entry)(pmd_t *pmd, unsigned 34 int (*pmd_entry)(pmd_t *pmd, unsigned long addr, 66 unsigned long next, s 35 unsigned long next, struct mm_walk *walk); 67 int (*pte_entry)(pte_t *pte, unsigned 36 int (*pte_entry)(pte_t *pte, unsigned long addr, 68 unsigned long next, s 37 unsigned long next, struct mm_walk *walk); 69 int (*pte_hole)(unsigned long addr, un 38 int (*pte_hole)(unsigned long addr, unsigned long next, 70 int depth, struct mm_w !! 39 struct mm_walk *walk); 71 int (*hugetlb_entry)(pte_t *pte, unsig 40 int (*hugetlb_entry)(pte_t *pte, unsigned long hmask, 72 unsigned long add 41 unsigned long addr, unsigned long next, 73 struct mm_walk *w 42 struct mm_walk *walk); 74 int (*test_walk)(unsigned long addr, u 43 int (*test_walk)(unsigned long addr, unsigned long next, 75 struct mm_walk *walk); 44 struct mm_walk *walk); 76 int (*pre_vma)(unsigned long start, un 45 int (*pre_vma)(unsigned long start, unsigned long end, 77 struct mm_walk *walk); 46 struct mm_walk *walk); 78 void (*post_vma)(struct mm_walk *walk) 47 void (*post_vma)(struct mm_walk *walk); 79 enum page_walk_lock walk_lock; << 80 }; << 81 << 82 /* << 83 * Action for pud_entry / pmd_entry callbacks. << 84 * ACTION_SUBTREE is the default << 85 */ << 86 enum page_walk_action { << 87 /* Descend to next level, splitting hu << 88 ACTION_SUBTREE = 0, << 89 /* Continue to next entry at this leve << 90 ACTION_CONTINUE = 1, << 91 /* Call again for this entry */ << 92 ACTION_AGAIN = 2 << 93 }; 48 }; 94 49 95 /** 50 /** 96 * struct mm_walk - walk_page_range data !! 51 * mm_walk - walk_page_range data 97 * @ops: operation to call during the w 52 * @ops: operation to call during the walk 98 * @mm: mm_struct representing the tar 53 * @mm: mm_struct representing the target process of page table walk 99 * @pgd: pointer to PGD; only valid wit << 100 * @vma: vma currently walked (NULL if 54 * @vma: vma currently walked (NULL if walking outside vmas) 101 * @action: next action to perform (see en << 102 * @no_vma: walk ignoring vmas (vma will a << 103 * @private: private data for callbacks' us 55 * @private: private data for callbacks' usage 104 * 56 * 105 * (see the comment on walk_page_range() for m 57 * (see the comment on walk_page_range() for more details) 106 */ 58 */ 107 struct mm_walk { 59 struct mm_walk { 108 const struct mm_walk_ops *ops; 60 const struct mm_walk_ops *ops; 109 struct mm_struct *mm; 61 struct mm_struct *mm; 110 pgd_t *pgd; << 111 struct vm_area_struct *vma; 62 struct vm_area_struct *vma; 112 enum page_walk_action action; << 113 bool no_vma; << 114 void *private; 63 void *private; 115 }; 64 }; 116 65 117 int walk_page_range(struct mm_struct *mm, unsi 66 int walk_page_range(struct mm_struct *mm, unsigned long start, 118 unsigned long end, const struc 67 unsigned long end, const struct mm_walk_ops *ops, 119 void *private); 68 void *private); 120 int walk_page_range_novma(struct mm_struct *mm << 121 unsigned long end, c << 122 pgd_t *pgd, << 123 void *private); << 124 int walk_page_range_vma(struct vm_area_struct << 125 unsigned long end, con << 126 void *private); << 127 int walk_page_vma(struct vm_area_struct *vma, 69 int walk_page_vma(struct vm_area_struct *vma, const struct mm_walk_ops *ops, 128 void *private); 70 void *private); 129 int walk_page_mapping(struct address_space *ma 71 int walk_page_mapping(struct address_space *mapping, pgoff_t first_index, 130 pgoff_t nr, const struct 72 pgoff_t nr, const struct mm_walk_ops *ops, 131 void *private); 73 void *private); 132 << 133 typedef int __bitwise folio_walk_flags_t; << 134 << 135 /* << 136 * Walk migration entries as well. Careful: a << 137 * concurrently. << 138 */ << 139 #define FW_MIGRATION ((__fo << 140 << 141 /* Walk shared zeropages (small + huge) as wel << 142 #define FW_ZEROPAGE ((__fo << 143 << 144 enum folio_walk_level { << 145 FW_LEVEL_PTE, << 146 FW_LEVEL_PMD, << 147 FW_LEVEL_PUD, << 148 }; << 149 << 150 /** << 151 * struct folio_walk - folio_walk_start() / fo << 152 * @page: exact folio page referenced (i << 153 * @level: page table level identifying t << 154 * @pte: pointer to the page table entr << 155 * @pmd: pointer to the page table entr << 156 * @pud: pointer to the page table entr << 157 * @ptl: pointer to the page table lock << 158 * << 159 * (see folio_walk_start() documentation for m << 160 */ << 161 struct folio_walk { << 162 /* public */ << 163 struct page *page; << 164 enum folio_walk_level level; << 165 union { << 166 pte_t *ptep; << 167 pud_t *pudp; << 168 pmd_t *pmdp; << 169 }; << 170 union { << 171 pte_t pte; << 172 pud_t pud; << 173 pmd_t pmd; << 174 }; << 175 /* private */ << 176 struct vm_area_struct *vma; << 177 spinlock_t *ptl; << 178 }; << 179 << 180 struct folio *folio_walk_start(struct folio_wa << 181 struct vm_area_struct *vma, un << 182 folio_walk_flags_t flags); << 183 << 184 #define folio_walk_end(__fw, __vma) do { \ << 185 spin_unlock((__fw)->ptl); \ << 186 if (likely((__fw)->level == FW_LEVEL_P << 187 pte_unmap((__fw)->ptep); \ << 188 vma_pgtable_walk_end(__vma); \ << 189 } while (0) << 190 74 191 #endif /* _LINUX_PAGEWALK_H */ 75 #endif /* _LINUX_PAGEWALK_H */ 192 76
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.