1 // SPDX-License-Identifier: GPL-2.0 2 #include <linux/mm.h> 3 #include <linux/rmap.h> 4 #include <linux/hugetlb.h> 5 #include <linux/swap.h> 6 #include <linux/leafops.h> 7 8 #include "internal.h" 9 10 static inline bool not_found(struct page_vma_mapped_walk *pvmw) 11 { 12 page_vma_mapped_walk_done(pvmw); 13 return false; 14 } 15 16 static bool map_pte(struct page_vma_mapped_walk *pvmw, pmd_t *pmdvalp, 17 spinlock_t **ptlp) 18 { 19 bool is_migration; 20 pte_t ptent; 21 22 if (pvmw->flags & PVMW_SYNC) { 23 /* Use the stricter lookup */ 24 pvmw->pte = pte_offset_map_lock(pvmw->vma->vm_mm, pvmw->pmd, 25 pvmw->address, &pvmw->ptl); 26 *ptlp = pvmw->ptl; 27 return !!pvmw->pte; 28 } 29 30 is_migration = pvmw->flags & PVMW_MIGRATION; 31 again: 32 /* 33 * It is important to return the ptl corresponding to pte, 34 * in case *pvmw->pmd changes underneath us; so we need to 35 * return it even when choosing not to lock, in case caller 36 * proceeds to loop over next ptes, and finds a match later. 37 * Though, in most cases, page lock already protects this. 38 */ 39 pvmw->pte = pte_offset_map_rw_nolock(pvmw->vma->vm_mm, pvmw->pmd, 40 pvmw->address, pmdvalp, ptlp); 41 if (!pvmw->pte) 42 return false; 43 44 ptent = ptep_get_lockless(pvmw->pte); 45 46 if (pte_none(ptent)) { 47 return false; 48 } else if (pte_present(ptent)) { 49 if (is_migration) 50 return false; 51 } else if (!is_migration) { 52 softleaf_t entry; 53 54 /* 55 * Handle un-addressable ZONE_DEVICE memory. 56 * 57 * We get here when we are trying to unmap a private 58 * device page from the process address space. Such 59 * page is not CPU accessible and thus is mapped as 60 * a special swap entry, nonetheless it still does 61 * count as a valid regular mapping for the page 62 * (and is accounted as such in page maps count). 63 * 64 * So handle this special case as if it was a normal 65 * page mapping ie lock CPU page table and return true. 66 * 67 * For more details on device private memory see HMM 68 * (include/linux/hmm.h or mm/hmm.c). 69 */ 70 entry = softleaf_from_pte(ptent); 71 if (!softleaf_is_device_private(entry) && 72 !softleaf_is_device_exclusive(entry)) 73 return false; 74 } 75 spin_lock(*ptlp); 76 if (unlikely(!pmd_same(*pmdvalp, pmdp_get_lockless(pvmw->pmd)))) { 77 pte_unmap_unlock(pvmw->pte, *ptlp); 78 goto again; 79 } 80 pvmw->ptl = *ptlp; 81 82 return true; 83 } 84 85 /** 86 * check_pte - check if [pvmw->pfn, @pvmw->pfn + @pvmw->nr_pages) is 87 * mapped at the @pvmw->pte 88 * @pvmw: page_vma_mapped_walk struct, includes a pair pte and pfn range 89 * for checking 90 * @pte_nr: the number of small pages described by @pvmw->pte. 91 * 92 * page_vma_mapped_walk() found a place where pfn range is *potentially* 93 * mapped. check_pte() has to validate this. 94 * 95 * pvmw->pte may point to empty PTE, swap PTE or PTE pointing to 96 * arbitrary page. 97 * 98 * If PVMW_MIGRATION flag is set, returns true if @pvmw->pte contains migration 99 * entry that points to [pvmw->pfn, @pvmw->pfn + @pvmw->nr_pages) 100 * 101 * If PVMW_MIGRATION flag is not set, returns true if pvmw->pte points to 102 * [pvmw->pfn, @pvmw->pfn + @pvmw->nr_pages) 103 * 104 * Otherwise, return false. 105 * 106 */ 107 static bool check_pte(struct page_vma_mapped_walk *pvmw, unsigned long pte_nr) 108 { 109 unsigned long pfn; 110 pte_t ptent; 111 112 if (is_vm_hugetlb_page(pvmw->vma)) 113 ptent = huge_ptep_get(pvmw->vma->vm_mm, pvmw->address, 114 pvmw->pte); 115 else 116 ptent = ptep_get(pvmw->pte); 117 118 if (pvmw->flags & PVMW_MIGRATION) { 119 const softleaf_t entry = softleaf_from_pte(ptent); 120 121 if (!softleaf_is_migration(entry)) 122 return false; 123 124 pfn = softleaf_to_pfn(entry); 125 } else if (pte_present(ptent)) { 126 pfn = pte_pfn(ptent); 127 } else { 128 const softleaf_t entry = softleaf_from_pte(ptent); 129 130 /* Handle un-addressable ZONE_DEVICE memory */ 131 if (!softleaf_is_device_private(entry) && 132 !softleaf_is_device_exclusive(entry)) 133 return false; 134 135 pfn = softleaf_to_pfn(entry); 136 } 137 138 if ((pfn + pte_nr - 1) < pvmw->pfn) 139 return false; 140 if (pfn > (pvmw->pfn + pvmw->nr_pages - 1)) 141 return false; 142 return true; 143 } 144 145 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 146 /* Returns true if the two ranges overlap. Careful to not overflow. */ 147 static bool check_pmd(unsigned long pfn, struct page_vma_mapped_walk *pvmw) 148 { 149 if ((pfn + HPAGE_PMD_NR - 1) < pvmw->pfn) 150 return false; 151 if (pfn > pvmw->pfn + pvmw->nr_pages - 1) 152 return false; 153 return true; 154 } 155 #else 156 static bool check_pmd(unsigned long pfn, struct page_vma_mapped_walk *pvmw) 157 { 158 return false; 159 } 160 #endif 161 162 static void step_forward(struct page_vma_mapped_walk *pvmw, unsigned long size) 163 { 164 pvmw->address = (pvmw->address + size) & ~(size - 1); 165 if (!pvmw->address) 166 pvmw->address = ULONG_MAX; 167 } 168 169 /** 170 * page_vma_mapped_walk - check if @pvmw->pfn is mapped in @pvmw->vma at 171 * @pvmw->address 172 * @pvmw: pointer to struct page_vma_mapped_walk. page, vma, address and flags 173 * must be set. pmd, pte and ptl must be NULL. 174 * 175 * Returns true if the page is mapped in the vma. @pvmw->pmd and @pvmw->pte point 176 * to relevant page table entries. @pvmw->ptl is locked. @pvmw->address is 177 * adjusted if needed (for PTE-mapped THPs). 178 * 179 * If @pvmw->pmd is set but @pvmw->pte is not, you have found PMD-mapped page 180 * (usually THP). For PTE-mapped THP, you should run page_vma_mapped_walk() in 181 * a loop to find all PTEs that map the THP. 182 * 183 * For HugeTLB pages, @pvmw->pte is set to the relevant page table entry 184 * regardless of which page table level the page is mapped at. @pvmw->pmd is 185 * NULL. 186 * 187 * Returns false if there are no more page table entries for the page in 188 * the vma. @pvmw->ptl is unlocked and @pvmw->pte is unmapped. 189 * 190 * If you need to stop the walk before page_vma_mapped_walk() returned false, 191 * use page_vma_mapped_walk_done(). It will do the housekeeping. 192 */ 193 bool page_vma_mapped_walk(struct page_vma_mapped_walk *pvmw) 194 { 195 struct vm_area_struct *vma = pvmw->vma; 196 struct mm_struct *mm = vma->vm_mm; 197 unsigned long end; 198 spinlock_t *ptl; 199 pte_t pteval; 200 pgd_t *pgd; 201 p4d_t *p4d; 202 pud_t *pud; 203 pmd_t pmde; 204 205 /* The only possible pmd mapping has been handled on last iteration */ 206 if (pvmw->pmd && !pvmw->pte) 207 return not_found(pvmw); 208 209 if (unlikely(is_vm_hugetlb_page(vma))) { 210 struct hstate *hstate = hstate_vma(vma); 211 unsigned long size = huge_page_size(hstate); 212 /* The only possible mapping was handled on last iteration */ 213 if (pvmw->pte) 214 return not_found(pvmw); 215 /* 216 * All callers that get here will already hold the 217 * i_mmap_rwsem. Therefore, no additional locks need to be 218 * taken before calling hugetlb_walk(). 219 */ 220 pvmw->pte = hugetlb_walk(vma, pvmw->address, size); 221 if (!pvmw->pte) 222 return false; 223 224 pvmw->ptl = huge_pte_lock(hstate, mm, pvmw->pte); 225 if (!check_pte(pvmw, pages_per_huge_page(hstate))) 226 return not_found(pvmw); 227 return true; 228 } 229 230 end = vma_address_end(pvmw); 231 if (pvmw->pte) 232 goto next_pte; 233 restart: 234 do { 235 pgd = pgd_offset(mm, pvmw->address); 236 if (!pgd_present(*pgd)) { 237 step_forward(pvmw, PGDIR_SIZE); 238 continue; 239 } 240 p4d = p4d_offset(pgd, pvmw->address); 241 if (!p4d_present(*p4d)) { 242 step_forward(pvmw, P4D_SIZE); 243 continue; 244 } 245 pud = pud_offset(p4d, pvmw->address); 246 if (!pud_present(*pud)) { 247 step_forward(pvmw, PUD_SIZE); 248 continue; 249 } 250 251 pvmw->pmd = pmd_offset(pud, pvmw->address); 252 /* 253 * Make sure the pmd value isn't cached in a register by the 254 * compiler and used as a stale value after we've observed a 255 * subsequent update. 256 */ 257 pmde = pmdp_get_lockless(pvmw->pmd); 258 259 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) && 260 (pmd_trans_huge(pmde) || pmd_is_migration_entry(pmde) || 261 pmd_is_device_private_entry(pmde))) { 262 pvmw->ptl = pmd_lock(mm, pvmw->pmd); 263 pmde = *pvmw->pmd; 264 if (pmd_is_migration_entry(pmde)) { 265 softleaf_t entry; 266 267 if (!(pvmw->flags & PVMW_MIGRATION)) 268 return not_found(pvmw); 269 entry = softleaf_from_pmd(pmde); 270 if (!check_pmd(softleaf_to_pfn(entry), pvmw)) 271 return not_found(pvmw); 272 return true; 273 } else if (pmd_is_device_private_entry(pmde)) { 274 softleaf_t entry; 275 276 if (pvmw->flags & PVMW_MIGRATION) 277 return not_found(pvmw); 278 entry = softleaf_from_pmd(pmde); 279 if (!check_pmd(softleaf_to_pfn(entry), pvmw)) 280 return not_found(pvmw); 281 return true; 282 } else if (!pmd_present(pmde)) { 283 return not_found(pvmw); 284 } 285 if (likely(pmd_trans_huge(pmde))) { 286 if (pvmw->flags & PVMW_MIGRATION) 287 return not_found(pvmw); 288 if (!check_pmd(pmd_pfn(pmde), pvmw)) 289 return not_found(pvmw); 290 return true; 291 } 292 /* THP/device-private pmd was split under us: handle on pte level */ 293 spin_unlock(pvmw->ptl); 294 pvmw->ptl = NULL; 295 } else if (!pmd_present(pmde)) { 296 if ((pvmw->flags & PVMW_SYNC) && 297 thp_vma_suitable_order(vma, pvmw->address, 298 PMD_ORDER) && 299 (pvmw->nr_pages >= HPAGE_PMD_NR)) 300 sync_with_folio_pmd_zap(mm, pvmw->pmd); 301 302 step_forward(pvmw, PMD_SIZE); 303 continue; 304 } 305 if (!map_pte(pvmw, &pmde, &ptl)) { 306 if (!pvmw->pte) 307 goto restart; 308 goto next_pte; 309 } 310 this_pte: 311 if (check_pte(pvmw, 1)) 312 return true; 313 next_pte: 314 do { 315 pvmw->address += PAGE_SIZE; 316 if (pvmw->address >= end) 317 return not_found(pvmw); 318 /* Did we cross page table boundary? */ 319 if ((pvmw->address & (PMD_SIZE - PAGE_SIZE)) == 0) { 320 if (pvmw->ptl) { 321 spin_unlock(pvmw->ptl); 322 pvmw->ptl = NULL; 323 } 324 pte_unmap(pvmw->pte); 325 pvmw->pte = NULL; 326 pvmw->flags |= PVMW_PGTABLE_CROSSED; 327 goto restart; 328 } 329 pvmw->pte++; 330 if (!pvmw->ptl) 331 pteval = ptep_get_lockless(pvmw->pte); 332 else 333 pteval = ptep_get(pvmw->pte); 334 } while (pte_none(pteval)); 335 336 if (!pvmw->ptl) { 337 spin_lock(ptl); 338 if (unlikely(!pmd_same(pmde, pmdp_get_lockless(pvmw->pmd)))) { 339 pte_unmap_unlock(pvmw->pte, ptl); 340 pvmw->pte = NULL; 341 goto restart; 342 } 343 pvmw->ptl = ptl; 344 } 345 goto this_pte; 346 } while (pvmw->address < end); 347 348 return false; 349 } 350 351 #ifdef CONFIG_MEMORY_FAILURE 352 /** 353 * page_mapped_in_vma - check whether a page is really mapped in a VMA 354 * @page: the page to test 355 * @vma: the VMA to test 356 * 357 * Return: The address the page is mapped at if the page is in the range 358 * covered by the VMA and present in the page table. If the page is 359 * outside the VMA or not present, returns -EFAULT. 360 * Only valid for normal file or anonymous VMAs. 361 */ 362 unsigned long page_mapped_in_vma(const struct page *page, 363 struct vm_area_struct *vma) 364 { 365 const struct folio *folio = page_folio(page); 366 const pgoff_t pgoff = page_pgoff(folio, page); 367 struct page_vma_mapped_walk pvmw = { 368 .pfn = page_to_pfn(page), 369 .nr_pages = 1, 370 .vma = vma, 371 .flags = PVMW_SYNC, 372 }; 373 374 if (folio_test_anon(folio)) 375 pvmw.address = vma_anon_address(vma, pgoff, 1); 376 else 377 pvmw.address = vma_filebacked_address(vma, pgoff, 1); 378 if (pvmw.address == -EFAULT) 379 goto out; 380 if (!page_vma_mapped_walk(&pvmw)) 381 return -EFAULT; 382 page_vma_mapped_walk_done(&pvmw); 383 out: 384 return pvmw.address; 385 } 386 #endif 387