1 // SPDX-License-Identifier: GPL-2.0
3 #include <linux/rmap.h>
4 #include <linux/hugetlb.h>
5 #include <linux/swap.h>
6 #include <linux/swapops.h>
10 static inline bool not_found(struct page_vma_mapped_walk *pvmw)
12 page_vma_mapped_walk_done(pvmw);
16 static bool map_pte(struct page_vma_mapped_walk *pvmw)
18 pvmw->pte = pte_offset_map(pvmw->pmd, pvmw->address);
19 if (!(pvmw->flags & PVMW_SYNC)) {
20 if (pvmw->flags & PVMW_MIGRATION) {
21 if (!is_swap_pte(*pvmw->pte))
25 * We get here when we are trying to unmap a private
26 * device page from the process address space. Such
27 * page is not CPU accessible and thus is mapped as
28 * a special swap entry, nonetheless it still does
29 * count as a valid regular mapping for the page (and
30 * is accounted as such in page maps count).
32 * So handle this special case as if it was a normal
33 * page mapping ie lock CPU page table and returns
36 * For more details on device private memory see HMM
37 * (include/linux/hmm.h or mm/hmm.c).
39 if (is_swap_pte(*pvmw->pte)) {
42 /* Handle un-addressable ZONE_DEVICE memory */
43 entry = pte_to_swp_entry(*pvmw->pte);
44 if (!is_device_private_entry(entry))
46 } else if (!pte_present(*pvmw->pte))
50 pvmw->ptl = pte_lockptr(pvmw->vma->vm_mm, pvmw->pmd);
55 static inline bool pfn_in_hpage(struct page *hpage, unsigned long pfn)
57 unsigned long hpage_pfn = page_to_pfn(hpage);
59 /* THP can be referenced by any subpage */
60 return pfn >= hpage_pfn && pfn - hpage_pfn < hpage_nr_pages(hpage);
64 * check_pte - check if @pvmw->page is mapped at the @pvmw->pte
66 * page_vma_mapped_walk() found a place where @pvmw->page is *potentially*
67 * mapped. check_pte() has to validate this.
69 * @pvmw->pte may point to empty PTE, swap PTE or PTE pointing to arbitrary
72 * If PVMW_MIGRATION flag is set, returns true if @pvmw->pte contains migration
73 * entry that points to @pvmw->page or any subpage in case of THP.
75 * If PVMW_MIGRATION flag is not set, returns true if @pvmw->pte points to
76 * @pvmw->page or any subpage in case of THP.
78 * Otherwise, return false.
81 static bool check_pte(struct page_vma_mapped_walk *pvmw)
85 if (pvmw->flags & PVMW_MIGRATION) {
87 if (!is_swap_pte(*pvmw->pte))
89 entry = pte_to_swp_entry(*pvmw->pte);
91 if (!is_migration_entry(entry))
94 pfn = migration_entry_to_pfn(entry);
95 } else if (is_swap_pte(*pvmw->pte)) {
98 /* Handle un-addressable ZONE_DEVICE memory */
99 entry = pte_to_swp_entry(*pvmw->pte);
100 if (!is_device_private_entry(entry))
103 pfn = device_private_entry_to_pfn(entry);
105 if (!pte_present(*pvmw->pte))
108 pfn = pte_pfn(*pvmw->pte);
111 return pfn_in_hpage(pvmw->page, pfn);
114 static void step_forward(struct page_vma_mapped_walk *pvmw, unsigned long size)
116 pvmw->address = (pvmw->address + size) & ~(size - 1);
118 pvmw->address = ULONG_MAX;
122 * page_vma_mapped_walk - check if @pvmw->page is mapped in @pvmw->vma at
124 * @pvmw: pointer to struct page_vma_mapped_walk. page, vma, address and flags
125 * must be set. pmd, pte and ptl must be NULL.
127 * Returns true if the page is mapped in the vma. @pvmw->pmd and @pvmw->pte point
128 * to relevant page table entries. @pvmw->ptl is locked. @pvmw->address is
129 * adjusted if needed (for PTE-mapped THPs).
131 * If @pvmw->pmd is set but @pvmw->pte is not, you have found PMD-mapped page
132 * (usually THP). For PTE-mapped THP, you should run page_vma_mapped_walk() in
133 * a loop to find all PTEs that map the THP.
135 * For HugeTLB pages, @pvmw->pte is set to the relevant page table entry
136 * regardless of which page table level the page is mapped at. @pvmw->pmd is
139 * Retruns false if there are no more page table entries for the page in
140 * the vma. @pvmw->ptl is unlocked and @pvmw->pte is unmapped.
142 * If you need to stop the walk before page_vma_mapped_walk() returned false,
143 * use page_vma_mapped_walk_done(). It will do the housekeeping.
145 bool page_vma_mapped_walk(struct page_vma_mapped_walk *pvmw)
147 struct mm_struct *mm = pvmw->vma->vm_mm;
148 struct page *page = pvmw->page;
155 /* The only possible pmd mapping has been handled on last iteration */
156 if (pvmw->pmd && !pvmw->pte)
157 return not_found(pvmw);
159 if (unlikely(PageHuge(page))) {
160 /* The only possible mapping was handled on last iteration */
162 return not_found(pvmw);
164 /* when pud is not present, pte will be NULL */
165 pvmw->pte = huge_pte_offset(mm, pvmw->address,
166 PAGE_SIZE << compound_order(page));
170 pvmw->ptl = huge_pte_lockptr(page_hstate(page), mm, pvmw->pte);
171 spin_lock(pvmw->ptl);
172 if (!check_pte(pvmw))
173 return not_found(pvmw);
178 * Seek to next pte only makes sense for THP.
179 * But more important than that optimization, is to filter out
180 * any PageKsm page: whose page->index misleads vma_address()
181 * and vma_address_end() to disaster.
183 end = PageTransCompound(page) ?
184 vma_address_end(page, pvmw->vma) :
185 pvmw->address + PAGE_SIZE;
190 pgd = pgd_offset(mm, pvmw->address);
191 if (!pgd_present(*pgd)) {
192 step_forward(pvmw, PGDIR_SIZE);
195 p4d = p4d_offset(pgd, pvmw->address);
196 if (!p4d_present(*p4d)) {
197 step_forward(pvmw, P4D_SIZE);
200 pud = pud_offset(p4d, pvmw->address);
201 if (!pud_present(*pud)) {
202 step_forward(pvmw, PUD_SIZE);
206 pvmw->pmd = pmd_offset(pud, pvmw->address);
208 * Make sure the pmd value isn't cached in a register by the
209 * compiler and used as a stale value after we've observed a
212 pmde = READ_ONCE(*pvmw->pmd);
214 if (pmd_trans_huge(pmde) || is_pmd_migration_entry(pmde)) {
215 pvmw->ptl = pmd_lock(mm, pvmw->pmd);
217 if (likely(pmd_trans_huge(pmde))) {
218 if (pvmw->flags & PVMW_MIGRATION)
219 return not_found(pvmw);
220 if (pmd_page(pmde) != page)
221 return not_found(pvmw);
224 if (!pmd_present(pmde)) {
227 if (!thp_migration_supported() ||
228 !(pvmw->flags & PVMW_MIGRATION))
229 return not_found(pvmw);
230 entry = pmd_to_swp_entry(pmde);
231 if (!is_migration_entry(entry) ||
232 migration_entry_to_page(entry) != page)
233 return not_found(pvmw);
236 /* THP pmd was split under us: handle on pte level */
237 spin_unlock(pvmw->ptl);
239 } else if (!pmd_present(pmde)) {
241 * If PVMW_SYNC, take and drop THP pmd lock so that we
242 * cannot return prematurely, while zap_huge_pmd() has
243 * cleared *pmd but not decremented compound_mapcount().
245 if ((pvmw->flags & PVMW_SYNC) &&
246 PageTransCompound(page)) {
247 spinlock_t *ptl = pmd_lock(mm, pvmw->pmd);
251 step_forward(pvmw, PMD_SIZE);
261 pvmw->address += PAGE_SIZE;
262 if (pvmw->address >= end)
263 return not_found(pvmw);
264 /* Did we cross page table boundary? */
265 if ((pvmw->address & (PMD_SIZE - PAGE_SIZE)) == 0) {
267 spin_unlock(pvmw->ptl);
270 pte_unmap(pvmw->pte);
275 if ((pvmw->flags & PVMW_SYNC) && !pvmw->ptl) {
276 pvmw->ptl = pte_lockptr(mm, pvmw->pmd);
277 spin_lock(pvmw->ptl);
279 } while (pte_none(*pvmw->pte));
282 pvmw->ptl = pte_lockptr(mm, pvmw->pmd);
283 spin_lock(pvmw->ptl);
286 } while (pvmw->address < end);
292 * page_mapped_in_vma - check whether a page is really mapped in a VMA
293 * @page: the page to test
294 * @vma: the VMA to test
296 * Returns 1 if the page is mapped into the page tables of the VMA, 0
297 * if the page is not mapped into the page tables of this VMA. Only
298 * valid for normal file or anonymous VMAs.
300 int page_mapped_in_vma(struct page *page, struct vm_area_struct *vma)
302 struct page_vma_mapped_walk pvmw = {
308 pvmw.address = vma_address(page, vma);
309 if (pvmw.address == -EFAULT)
311 if (!page_vma_mapped_walk(&pvmw))
313 page_vma_mapped_walk_done(&pvmw);