xref: /linux/include/linux/pagewalk.h (revision aa39ca69)
1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef _LINUX_PAGEWALK_H
3 #define _LINUX_PAGEWALK_H
4 
5 #include <linux/mm.h>
6 
7 struct mm_walk;
8 
9 /* Locking requirement during a page walk. */
10 enum page_walk_lock {
11 	/* mmap_lock should be locked for read to stabilize the vma tree */
12 	PGWALK_RDLOCK = 0,
13 	/* vma will be write-locked during the walk */
14 	PGWALK_WRLOCK = 1,
15 	/* vma is expected to be already write-locked during the walk */
16 	PGWALK_WRLOCK_VERIFY = 2,
17 };
18 
19 /**
20  * struct mm_walk_ops - callbacks for walk_page_range
21  * @pgd_entry:		if set, called for each non-empty PGD (top-level) entry
22  * @p4d_entry:		if set, called for each non-empty P4D entry
23  * @pud_entry:		if set, called for each non-empty PUD entry
24  * @pmd_entry:		if set, called for each non-empty PMD entry
25  *			this handler is required to be able to handle
26  *			pmd_trans_huge() pmds.  They may simply choose to
27  *			split_huge_page() instead of handling it explicitly.
28  * @pte_entry:		if set, called for each PTE (lowest-level) entry,
29  *			including empty ones
30  * @pte_hole:		if set, called for each hole at all levels,
31  *			depth is -1 if not known, 0:PGD, 1:P4D, 2:PUD, 3:PMD.
32  *			Any folded depths (where PTRS_PER_P?D is equal to 1)
33  *			are skipped.
34  * @hugetlb_entry:	if set, called for each hugetlb entry. This hook
35  *			function is called with the vma lock held, in order to
36  *			protect against a concurrent freeing of the pte_t* or
37  *			the ptl. In some cases, the hook function needs to drop
38  *			and retake the vma lock in order to avoid deadlocks
39  *			while calling other functions. In such cases the hook
40  *			function must either refrain from accessing the pte or
41  *			ptl after dropping the vma lock, or else revalidate
42  *			those items after re-acquiring the vma lock and before
43  *			accessing them.
44  * @test_walk:		caller specific callback function to determine whether
45  *			we walk over the current vma or not. Returning 0 means
46  *			"do page table walk over the current vma", returning
47  *			a negative value means "abort current page table walk
48  *			right now" and returning 1 means "skip the current vma"
49  *			Note that this callback is not called when the caller
50  *			passes in a single VMA as for walk_page_vma().
51  * @pre_vma:            if set, called before starting walk on a non-null vma.
52  * @post_vma:           if set, called after a walk on a non-null vma, provided
53  *                      that @pre_vma and the vma walk succeeded.
54  *
55  * p?d_entry callbacks are called even if those levels are folded on a
56  * particular architecture/configuration.
57  */
58 struct mm_walk_ops {
59 	int (*pgd_entry)(pgd_t *pgd, unsigned long addr,
60 			 unsigned long next, struct mm_walk *walk);
61 	int (*p4d_entry)(p4d_t *p4d, unsigned long addr,
62 			 unsigned long next, struct mm_walk *walk);
63 	int (*pud_entry)(pud_t *pud, unsigned long addr,
64 			 unsigned long next, struct mm_walk *walk);
65 	int (*pmd_entry)(pmd_t *pmd, unsigned long addr,
66 			 unsigned long next, struct mm_walk *walk);
67 	int (*pte_entry)(pte_t *pte, unsigned long addr,
68 			 unsigned long next, struct mm_walk *walk);
69 	int (*pte_hole)(unsigned long addr, unsigned long next,
70 			int depth, struct mm_walk *walk);
71 	int (*hugetlb_entry)(pte_t *pte, unsigned long hmask,
72 			     unsigned long addr, unsigned long next,
73 			     struct mm_walk *walk);
74 	int (*test_walk)(unsigned long addr, unsigned long next,
75 			struct mm_walk *walk);
76 	int (*pre_vma)(unsigned long start, unsigned long end,
77 		       struct mm_walk *walk);
78 	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 huge pages if needed and possible */
88 	ACTION_SUBTREE = 0,
89 	/* Continue to next entry at this level (ignoring any subtree) */
90 	ACTION_CONTINUE = 1,
91 	/* Call again for this entry */
92 	ACTION_AGAIN = 2
93 };
94 
95 /**
96  * struct mm_walk - walk_page_range data
97  * @ops:	operation to call during the walk
98  * @mm:		mm_struct representing the target process of page table walk
99  * @pgd:	pointer to PGD; only valid with no_vma (otherwise set to NULL)
100  * @vma:	vma currently walked (NULL if walking outside vmas)
101  * @action:	next action to perform (see enum page_walk_action)
102  * @no_vma:	walk ignoring vmas (vma will always be NULL)
103  * @private:	private data for callbacks' usage
104  *
105  * (see the comment on walk_page_range() for more details)
106  */
107 struct mm_walk {
108 	const struct mm_walk_ops *ops;
109 	struct mm_struct *mm;
110 	pgd_t *pgd;
111 	struct vm_area_struct *vma;
112 	enum page_walk_action action;
113 	bool no_vma;
114 	void *private;
115 };
116 
117 int walk_page_range(struct mm_struct *mm, unsigned long start,
118 		unsigned long end, const struct mm_walk_ops *ops,
119 		void *private);
120 int walk_page_range_novma(struct mm_struct *mm, unsigned long start,
121 			  unsigned long end, const struct mm_walk_ops *ops,
122 			  pgd_t *pgd,
123 			  void *private);
124 int walk_page_range_vma(struct vm_area_struct *vma, unsigned long start,
125 			unsigned long end, const struct mm_walk_ops *ops,
126 			void *private);
127 int walk_page_vma(struct vm_area_struct *vma, const struct mm_walk_ops *ops,
128 		void *private);
129 int walk_page_mapping(struct address_space *mapping, pgoff_t first_index,
130 		      pgoff_t nr, const struct mm_walk_ops *ops,
131 		      void *private);
132 
133 typedef int __bitwise folio_walk_flags_t;
134 
135 /*
136  * Walk migration entries as well. Careful: a large folio might get split
137  * concurrently.
138  */
139 #define FW_MIGRATION			((__force folio_walk_flags_t)BIT(0))
140 
141 /* Walk shared zeropages (small + huge) as well. */
142 #define FW_ZEROPAGE			((__force folio_walk_flags_t)BIT(1))
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() / folio_walk_end() data
152  * @page:	exact folio page referenced (if applicable)
153  * @level:	page table level identifying the entry type
154  * @pte:	pointer to the page table entry (FW_LEVEL_PTE).
155  * @pmd:	pointer to the page table entry (FW_LEVEL_PMD).
156  * @pud:	pointer to the page table entry (FW_LEVEL_PUD).
157  * @ptl:	pointer to the page table lock.
158  *
159  * (see folio_walk_start() documentation for more details)
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_walk *fw,
181 		struct vm_area_struct *vma, unsigned long addr,
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_PTE)) \
187 		pte_unmap((__fw)->ptep); \
188 	vma_pgtable_walk_end(__vma); \
189 } while (0)
190 
191 #endif /* _LINUX_PAGEWALK_H */
192