xref: /openbmc/linux/mm/page_vma_mapped.c (revision 6aa7de05)
1 #include <linux/mm.h>
2 #include <linux/rmap.h>
3 #include <linux/hugetlb.h>
4 #include <linux/swap.h>
5 #include <linux/swapops.h>
6 
7 #include "internal.h"
8 
9 static inline bool not_found(struct page_vma_mapped_walk *pvmw)
10 {
11 	page_vma_mapped_walk_done(pvmw);
12 	return false;
13 }
14 
15 static bool map_pte(struct page_vma_mapped_walk *pvmw)
16 {
17 	pvmw->pte = pte_offset_map(pvmw->pmd, pvmw->address);
18 	if (!(pvmw->flags & PVMW_SYNC)) {
19 		if (pvmw->flags & PVMW_MIGRATION) {
20 			if (!is_swap_pte(*pvmw->pte))
21 				return false;
22 		} else {
23 			if (!pte_present(*pvmw->pte))
24 				return false;
25 		}
26 	}
27 	pvmw->ptl = pte_lockptr(pvmw->vma->vm_mm, pvmw->pmd);
28 	spin_lock(pvmw->ptl);
29 	return true;
30 }
31 
32 static bool check_pte(struct page_vma_mapped_walk *pvmw)
33 {
34 	if (pvmw->flags & PVMW_MIGRATION) {
35 #ifdef CONFIG_MIGRATION
36 		swp_entry_t entry;
37 		if (!is_swap_pte(*pvmw->pte))
38 			return false;
39 		entry = pte_to_swp_entry(*pvmw->pte);
40 
41 		if (!is_migration_entry(entry))
42 			return false;
43 		if (migration_entry_to_page(entry) - pvmw->page >=
44 				hpage_nr_pages(pvmw->page)) {
45 			return false;
46 		}
47 		if (migration_entry_to_page(entry) < pvmw->page)
48 			return false;
49 #else
50 		WARN_ON_ONCE(1);
51 #endif
52 	} else {
53 		if (is_swap_pte(*pvmw->pte)) {
54 			swp_entry_t entry;
55 
56 			entry = pte_to_swp_entry(*pvmw->pte);
57 			if (is_device_private_entry(entry) &&
58 			    device_private_entry_to_page(entry) == pvmw->page)
59 				return true;
60 		}
61 
62 		if (!pte_present(*pvmw->pte))
63 			return false;
64 
65 		/* THP can be referenced by any subpage */
66 		if (pte_page(*pvmw->pte) - pvmw->page >=
67 				hpage_nr_pages(pvmw->page)) {
68 			return false;
69 		}
70 		if (pte_page(*pvmw->pte) < pvmw->page)
71 			return false;
72 	}
73 
74 	return true;
75 }
76 
77 /**
78  * page_vma_mapped_walk - check if @pvmw->page is mapped in @pvmw->vma at
79  * @pvmw->address
80  * @pvmw: pointer to struct page_vma_mapped_walk. page, vma, address and flags
81  * must be set. pmd, pte and ptl must be NULL.
82  *
83  * Returns true if the page is mapped in the vma. @pvmw->pmd and @pvmw->pte point
84  * to relevant page table entries. @pvmw->ptl is locked. @pvmw->address is
85  * adjusted if needed (for PTE-mapped THPs).
86  *
87  * If @pvmw->pmd is set but @pvmw->pte is not, you have found PMD-mapped page
88  * (usually THP). For PTE-mapped THP, you should run page_vma_mapped_walk() in
89  * a loop to find all PTEs that map the THP.
90  *
91  * For HugeTLB pages, @pvmw->pte is set to the relevant page table entry
92  * regardless of which page table level the page is mapped at. @pvmw->pmd is
93  * NULL.
94  *
95  * Retruns false if there are no more page table entries for the page in
96  * the vma. @pvmw->ptl is unlocked and @pvmw->pte is unmapped.
97  *
98  * If you need to stop the walk before page_vma_mapped_walk() returned false,
99  * use page_vma_mapped_walk_done(). It will do the housekeeping.
100  */
101 bool page_vma_mapped_walk(struct page_vma_mapped_walk *pvmw)
102 {
103 	struct mm_struct *mm = pvmw->vma->vm_mm;
104 	struct page *page = pvmw->page;
105 	pgd_t *pgd;
106 	p4d_t *p4d;
107 	pud_t *pud;
108 	pmd_t pmde;
109 
110 	/* The only possible pmd mapping has been handled on last iteration */
111 	if (pvmw->pmd && !pvmw->pte)
112 		return not_found(pvmw);
113 
114 	if (pvmw->pte)
115 		goto next_pte;
116 
117 	if (unlikely(PageHuge(pvmw->page))) {
118 		/* when pud is not present, pte will be NULL */
119 		pvmw->pte = huge_pte_offset(mm, pvmw->address,
120 					    PAGE_SIZE << compound_order(page));
121 		if (!pvmw->pte)
122 			return false;
123 
124 		pvmw->ptl = huge_pte_lockptr(page_hstate(page), mm, pvmw->pte);
125 		spin_lock(pvmw->ptl);
126 		if (!check_pte(pvmw))
127 			return not_found(pvmw);
128 		return true;
129 	}
130 restart:
131 	pgd = pgd_offset(mm, pvmw->address);
132 	if (!pgd_present(*pgd))
133 		return false;
134 	p4d = p4d_offset(pgd, pvmw->address);
135 	if (!p4d_present(*p4d))
136 		return false;
137 	pud = pud_offset(p4d, pvmw->address);
138 	if (!pud_present(*pud))
139 		return false;
140 	pvmw->pmd = pmd_offset(pud, pvmw->address);
141 	/*
142 	 * Make sure the pmd value isn't cached in a register by the
143 	 * compiler and used as a stale value after we've observed a
144 	 * subsequent update.
145 	 */
146 	pmde = READ_ONCE(*pvmw->pmd);
147 	if (pmd_trans_huge(pmde) || is_pmd_migration_entry(pmde)) {
148 		pvmw->ptl = pmd_lock(mm, pvmw->pmd);
149 		if (likely(pmd_trans_huge(*pvmw->pmd))) {
150 			if (pvmw->flags & PVMW_MIGRATION)
151 				return not_found(pvmw);
152 			if (pmd_page(*pvmw->pmd) != page)
153 				return not_found(pvmw);
154 			return true;
155 		} else if (!pmd_present(*pvmw->pmd)) {
156 			if (thp_migration_supported()) {
157 				if (!(pvmw->flags & PVMW_MIGRATION))
158 					return not_found(pvmw);
159 				if (is_migration_entry(pmd_to_swp_entry(*pvmw->pmd))) {
160 					swp_entry_t entry = pmd_to_swp_entry(*pvmw->pmd);
161 
162 					if (migration_entry_to_page(entry) != page)
163 						return not_found(pvmw);
164 					return true;
165 				}
166 			}
167 			return not_found(pvmw);
168 		} else {
169 			/* THP pmd was split under us: handle on pte level */
170 			spin_unlock(pvmw->ptl);
171 			pvmw->ptl = NULL;
172 		}
173 	} else if (!pmd_present(pmde)) {
174 		return false;
175 	}
176 	if (!map_pte(pvmw))
177 		goto next_pte;
178 	while (1) {
179 		if (check_pte(pvmw))
180 			return true;
181 next_pte:
182 		/* Seek to next pte only makes sense for THP */
183 		if (!PageTransHuge(pvmw->page) || PageHuge(pvmw->page))
184 			return not_found(pvmw);
185 		do {
186 			pvmw->address += PAGE_SIZE;
187 			if (pvmw->address >= pvmw->vma->vm_end ||
188 			    pvmw->address >=
189 					__vma_address(pvmw->page, pvmw->vma) +
190 					hpage_nr_pages(pvmw->page) * PAGE_SIZE)
191 				return not_found(pvmw);
192 			/* Did we cross page table boundary? */
193 			if (pvmw->address % PMD_SIZE == 0) {
194 				pte_unmap(pvmw->pte);
195 				if (pvmw->ptl) {
196 					spin_unlock(pvmw->ptl);
197 					pvmw->ptl = NULL;
198 				}
199 				goto restart;
200 			} else {
201 				pvmw->pte++;
202 			}
203 		} while (pte_none(*pvmw->pte));
204 
205 		if (!pvmw->ptl) {
206 			pvmw->ptl = pte_lockptr(mm, pvmw->pmd);
207 			spin_lock(pvmw->ptl);
208 		}
209 	}
210 }
211 
212 /**
213  * page_mapped_in_vma - check whether a page is really mapped in a VMA
214  * @page: the page to test
215  * @vma: the VMA to test
216  *
217  * Returns 1 if the page is mapped into the page tables of the VMA, 0
218  * if the page is not mapped into the page tables of this VMA.  Only
219  * valid for normal file or anonymous VMAs.
220  */
221 int page_mapped_in_vma(struct page *page, struct vm_area_struct *vma)
222 {
223 	struct page_vma_mapped_walk pvmw = {
224 		.page = page,
225 		.vma = vma,
226 		.flags = PVMW_SYNC,
227 	};
228 	unsigned long start, end;
229 
230 	start = __vma_address(page, vma);
231 	end = start + PAGE_SIZE * (hpage_nr_pages(page) - 1);
232 
233 	if (unlikely(end < vma->vm_start || start >= vma->vm_end))
234 		return 0;
235 	pvmw.address = max(start, vma->vm_start);
236 	if (!page_vma_mapped_walk(&pvmw))
237 		return 0;
238 	page_vma_mapped_walk_done(&pvmw);
239 	return 1;
240 }
241