1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef _LINUX_SWAPOPS_H
3 #define _LINUX_SWAPOPS_H
5 #include <linux/radix-tree.h>
7 #include <linux/mm_types.h>
12 #include <linux/swapfile.h>
13 #endif /* CONFIG_SWAP */
16 * swapcache pages are stored in the swapper_space radix tree. We want to
17 * get good packing density in that tree, so the index should be dense in
20 * We arrange the `type' and `offset' fields so that `type' is at the six
21 * high-order bits of the swp_entry_t and `offset' is right-aligned in the
22 * remaining bits. Although `type' itself needs only five bits, we allow for
23 * shmem/tmpfs to shift it all up a further one bit: see swp_to_radix_entry().
25 * swp_entry_t's are *never* stored anywhere in their arch-dependent format.
27 #define SWP_TYPE_SHIFT (BITS_PER_XA_VALUE - MAX_SWAPFILES_SHIFT)
28 #define SWP_OFFSET_MASK ((1UL << SWP_TYPE_SHIFT) - 1)
31 * Definitions only for PFN swap entries (see is_pfn_swap_entry()). To
32 * store PFN, we only need SWP_PFN_BITS bits. Each of the pfn swap entries
33 * can use the extra bits to store other information besides PFN.
35 #ifdef MAX_PHYSMEM_BITS
36 #define SWP_PFN_BITS (MAX_PHYSMEM_BITS - PAGE_SHIFT)
37 #else /* MAX_PHYSMEM_BITS */
38 #define SWP_PFN_BITS min_t(int, \
39 sizeof(phys_addr_t) * 8 - PAGE_SHIFT, \
41 #endif /* MAX_PHYSMEM_BITS */
42 #define SWP_PFN_MASK (BIT(SWP_PFN_BITS) - 1)
45 * Migration swap entry specific bitfield definitions. Layout:
47 * |----------+--------------------|
48 * | swp_type | swp_offset |
49 * |----------+--------+-+-+-------|
50 * | | resv |D|A| PFN |
51 * |----------+--------+-+-+-------|
53 * @SWP_MIG_YOUNG_BIT: Whether the page used to have young bit set (bit A)
54 * @SWP_MIG_DIRTY_BIT: Whether the page used to have dirty bit set (bit D)
56 * Note: A/D bits will be stored in migration entries iff there're enough
57 * free bits in arch specific swp offset. By default we'll ignore A/D bits
58 * when migrating a page. Please refer to migration_entry_supports_ad()
59 * for more information. If there're more bits besides PFN and A/D bits,
60 * they should be reserved and always be zeros.
62 #define SWP_MIG_YOUNG_BIT (SWP_PFN_BITS)
63 #define SWP_MIG_DIRTY_BIT (SWP_PFN_BITS + 1)
64 #define SWP_MIG_TOTAL_BITS (SWP_PFN_BITS + 2)
66 #define SWP_MIG_YOUNG BIT(SWP_MIG_YOUNG_BIT)
67 #define SWP_MIG_DIRTY BIT(SWP_MIG_DIRTY_BIT)
69 static inline bool is_pfn_swap_entry(swp_entry_t entry);
71 /* Clear all flags but only keep swp_entry_t related information */
72 static inline pte_t pte_swp_clear_flags(pte_t pte)
74 if (pte_swp_exclusive(pte))
75 pte = pte_swp_clear_exclusive(pte);
76 if (pte_swp_soft_dirty(pte))
77 pte = pte_swp_clear_soft_dirty(pte);
78 if (pte_swp_uffd_wp(pte))
79 pte = pte_swp_clear_uffd_wp(pte);
84 * Store a type+offset into a swp_entry_t in an arch-independent format
86 static inline swp_entry_t swp_entry(unsigned long type, pgoff_t offset)
90 ret.val = (type << SWP_TYPE_SHIFT) | (offset & SWP_OFFSET_MASK);
95 * Extract the `type' field from a swp_entry_t. The swp_entry_t is in
96 * arch-independent format
98 static inline unsigned swp_type(swp_entry_t entry)
100 return (entry.val >> SWP_TYPE_SHIFT);
104 * Extract the `offset' field from a swp_entry_t. The swp_entry_t is in
105 * arch-independent format
107 static inline pgoff_t swp_offset(swp_entry_t entry)
109 return entry.val & SWP_OFFSET_MASK;
113 * This should only be called upon a pfn swap entry to get the PFN stored
114 * in the swap entry. Please refers to is_pfn_swap_entry() for definition
117 static inline unsigned long swp_offset_pfn(swp_entry_t entry)
119 VM_BUG_ON(!is_pfn_swap_entry(entry));
120 return swp_offset(entry) & SWP_PFN_MASK;
123 /* check whether a pte points to a swap entry */
124 static inline int is_swap_pte(pte_t pte)
126 return !pte_none(pte) && !pte_present(pte);
130 * Convert the arch-dependent pte representation of a swp_entry_t into an
131 * arch-independent swp_entry_t.
133 static inline swp_entry_t pte_to_swp_entry(pte_t pte)
135 swp_entry_t arch_entry;
137 pte = pte_swp_clear_flags(pte);
138 arch_entry = __pte_to_swp_entry(pte);
139 return swp_entry(__swp_type(arch_entry), __swp_offset(arch_entry));
143 * Convert the arch-independent representation of a swp_entry_t into the
144 * arch-dependent pte representation.
146 static inline pte_t swp_entry_to_pte(swp_entry_t entry)
148 swp_entry_t arch_entry;
150 arch_entry = __swp_entry(swp_type(entry), swp_offset(entry));
151 return __swp_entry_to_pte(arch_entry);
154 static inline swp_entry_t radix_to_swp_entry(void *arg)
158 entry.val = xa_to_value(arg);
162 static inline void *swp_to_radix_entry(swp_entry_t entry)
164 return xa_mk_value(entry.val);
167 #if IS_ENABLED(CONFIG_DEVICE_PRIVATE)
168 static inline swp_entry_t make_readable_device_private_entry(pgoff_t offset)
170 return swp_entry(SWP_DEVICE_READ, offset);
173 static inline swp_entry_t make_writable_device_private_entry(pgoff_t offset)
175 return swp_entry(SWP_DEVICE_WRITE, offset);
178 static inline bool is_device_private_entry(swp_entry_t entry)
180 int type = swp_type(entry);
181 return type == SWP_DEVICE_READ || type == SWP_DEVICE_WRITE;
184 static inline bool is_writable_device_private_entry(swp_entry_t entry)
186 return unlikely(swp_type(entry) == SWP_DEVICE_WRITE);
189 static inline swp_entry_t make_device_exclusive_entry(pgoff_t offset)
191 return swp_entry(SWP_DEVICE_EXCLUSIVE, offset);
194 static inline bool is_device_exclusive_entry(swp_entry_t entry)
196 return swp_type(entry) == SWP_DEVICE_EXCLUSIVE;
199 #else /* CONFIG_DEVICE_PRIVATE */
200 static inline swp_entry_t make_readable_device_private_entry(pgoff_t offset)
202 return swp_entry(0, 0);
205 static inline swp_entry_t make_writable_device_private_entry(pgoff_t offset)
207 return swp_entry(0, 0);
210 static inline bool is_device_private_entry(swp_entry_t entry)
215 static inline bool is_writable_device_private_entry(swp_entry_t entry)
220 static inline swp_entry_t make_device_exclusive_entry(pgoff_t offset)
222 return swp_entry(0, 0);
225 static inline bool is_device_exclusive_entry(swp_entry_t entry)
230 #endif /* CONFIG_DEVICE_PRIVATE */
232 #ifdef CONFIG_MIGRATION
233 static inline int is_migration_entry(swp_entry_t entry)
235 return unlikely(swp_type(entry) == SWP_MIGRATION_READ ||
236 swp_type(entry) == SWP_MIGRATION_READ_EXCLUSIVE ||
237 swp_type(entry) == SWP_MIGRATION_WRITE);
240 static inline int is_writable_migration_entry(swp_entry_t entry)
242 return unlikely(swp_type(entry) == SWP_MIGRATION_WRITE);
245 static inline int is_readable_migration_entry(swp_entry_t entry)
247 return unlikely(swp_type(entry) == SWP_MIGRATION_READ);
250 static inline int is_readable_exclusive_migration_entry(swp_entry_t entry)
252 return unlikely(swp_type(entry) == SWP_MIGRATION_READ_EXCLUSIVE);
255 static inline swp_entry_t make_readable_migration_entry(pgoff_t offset)
257 return swp_entry(SWP_MIGRATION_READ, offset);
260 static inline swp_entry_t make_readable_exclusive_migration_entry(pgoff_t offset)
262 return swp_entry(SWP_MIGRATION_READ_EXCLUSIVE, offset);
265 static inline swp_entry_t make_writable_migration_entry(pgoff_t offset)
267 return swp_entry(SWP_MIGRATION_WRITE, offset);
271 * Returns whether the host has large enough swap offset field to support
272 * carrying over pgtable A/D bits for page migrations. The result is
273 * pretty much arch specific.
275 static inline bool migration_entry_supports_ad(void)
278 return swap_migration_ad_supported;
279 #else /* CONFIG_SWAP */
281 #endif /* CONFIG_SWAP */
284 static inline swp_entry_t make_migration_entry_young(swp_entry_t entry)
286 if (migration_entry_supports_ad())
287 return swp_entry(swp_type(entry),
288 swp_offset(entry) | SWP_MIG_YOUNG);
292 static inline bool is_migration_entry_young(swp_entry_t entry)
294 if (migration_entry_supports_ad())
295 return swp_offset(entry) & SWP_MIG_YOUNG;
296 /* Keep the old behavior of aging page after migration */
300 static inline swp_entry_t make_migration_entry_dirty(swp_entry_t entry)
302 if (migration_entry_supports_ad())
303 return swp_entry(swp_type(entry),
304 swp_offset(entry) | SWP_MIG_DIRTY);
308 static inline bool is_migration_entry_dirty(swp_entry_t entry)
310 if (migration_entry_supports_ad())
311 return swp_offset(entry) & SWP_MIG_DIRTY;
312 /* Keep the old behavior of clean page after migration */
316 extern void migration_entry_wait(struct mm_struct *mm, pmd_t *pmd,
317 unsigned long address);
318 extern void migration_entry_wait_huge(struct vm_area_struct *vma, unsigned long addr, pte_t *pte);
319 #else /* CONFIG_MIGRATION */
320 static inline swp_entry_t make_readable_migration_entry(pgoff_t offset)
322 return swp_entry(0, 0);
325 static inline swp_entry_t make_readable_exclusive_migration_entry(pgoff_t offset)
327 return swp_entry(0, 0);
330 static inline swp_entry_t make_writable_migration_entry(pgoff_t offset)
332 return swp_entry(0, 0);
335 static inline int is_migration_entry(swp_entry_t swp)
340 static inline void migration_entry_wait(struct mm_struct *mm, pmd_t *pmd,
341 unsigned long address) { }
342 static inline void migration_entry_wait_huge(struct vm_area_struct *vma,
343 unsigned long addr, pte_t *pte) { }
344 static inline int is_writable_migration_entry(swp_entry_t entry)
348 static inline int is_readable_migration_entry(swp_entry_t entry)
353 static inline swp_entry_t make_migration_entry_young(swp_entry_t entry)
358 static inline bool is_migration_entry_young(swp_entry_t entry)
363 static inline swp_entry_t make_migration_entry_dirty(swp_entry_t entry)
368 static inline bool is_migration_entry_dirty(swp_entry_t entry)
372 #endif /* CONFIG_MIGRATION */
374 #ifdef CONFIG_MEMORY_FAILURE
377 * Support for hardware poisoned pages
379 static inline swp_entry_t make_hwpoison_entry(struct page *page)
381 BUG_ON(!PageLocked(page));
382 return swp_entry(SWP_HWPOISON, page_to_pfn(page));
385 static inline int is_hwpoison_entry(swp_entry_t entry)
387 return swp_type(entry) == SWP_HWPOISON;
392 static inline swp_entry_t make_hwpoison_entry(struct page *page)
394 return swp_entry(0, 0);
397 static inline int is_hwpoison_entry(swp_entry_t swp)
403 typedef unsigned long pte_marker;
405 #define PTE_MARKER_UFFD_WP BIT(0)
407 * "Poisoned" here is meant in the very general sense of "future accesses are
408 * invalid", instead of referring very specifically to hardware memory errors.
409 * This marker is meant to represent any of various different causes of this.
411 * Note that, when encountered by the faulting logic, PTEs with this marker will
412 * result in VM_FAULT_HWPOISON and thus regardless trigger hardware memory error
415 #define PTE_MARKER_POISONED BIT(1)
417 * Indicates that, on fault, this PTE will case a SIGSEGV signal to be
418 * sent. This means guard markers behave in effect as if the region were mapped
419 * PROT_NONE, rather than if they were a memory hole or equivalent.
421 #define PTE_MARKER_GUARD BIT(2)
422 #define PTE_MARKER_MASK (BIT(3) - 1)
424 static inline swp_entry_t make_pte_marker_entry(pte_marker marker)
426 return swp_entry(SWP_PTE_MARKER, marker);
429 static inline bool is_pte_marker_entry(swp_entry_t entry)
431 return swp_type(entry) == SWP_PTE_MARKER;
434 static inline pte_marker pte_marker_get(swp_entry_t entry)
436 return swp_offset(entry) & PTE_MARKER_MASK;
439 static inline bool is_pte_marker(pte_t pte)
441 return is_swap_pte(pte) && is_pte_marker_entry(pte_to_swp_entry(pte));
444 static inline pte_t make_pte_marker(pte_marker marker)
446 return swp_entry_to_pte(make_pte_marker_entry(marker));
449 static inline swp_entry_t make_poisoned_swp_entry(void)
451 return make_pte_marker_entry(PTE_MARKER_POISONED);
454 static inline int is_poisoned_swp_entry(swp_entry_t entry)
456 return is_pte_marker_entry(entry) &&
457 (pte_marker_get(entry) & PTE_MARKER_POISONED);
461 static inline swp_entry_t make_guard_swp_entry(void)
463 return make_pte_marker_entry(PTE_MARKER_GUARD);
466 static inline int is_guard_swp_entry(swp_entry_t entry)
468 return is_pte_marker_entry(entry) &&
469 (pte_marker_get(entry) & PTE_MARKER_GUARD);
473 * This is a special version to check pte_none() just to cover the case when
474 * the pte is a pte marker. It existed because in many cases the pte marker
475 * should be seen as a none pte; it's just that we have stored some information
476 * onto the none pte so it becomes not-none any more.
478 * It should be used when the pte is file-backed, ram-based and backing
479 * userspace pages, like shmem. It is not needed upon pgtables that do not
480 * support pte markers at all. For example, it's not needed on anonymous
481 * memory, kernel-only memory (including when the system is during-boot),
482 * non-ram based generic file-system. It's fine to be used even there, but the
483 * extra pte marker check will be pure overhead.
485 static inline int pte_none_mostly(pte_t pte)
487 return pte_none(pte) || is_pte_marker(pte);
490 static inline struct page *pfn_swap_entry_to_page(swp_entry_t entry)
492 struct page *p = pfn_to_page(swp_offset_pfn(entry));
495 * Any use of migration entries may only occur while the
496 * corresponding page is locked
498 BUG_ON(is_migration_entry(entry) && !PageLocked(p));
503 static inline struct folio *pfn_swap_entry_folio(swp_entry_t entry)
505 struct folio *folio = pfn_folio(swp_offset_pfn(entry));
508 * Any use of migration entries may only occur while the
509 * corresponding folio is locked
511 BUG_ON(is_migration_entry(entry) && !folio_test_locked(folio));
517 * A pfn swap entry is a special type of swap entry that always has a pfn stored
518 * in the swap offset. They can either be used to represent unaddressable device
519 * memory, to restrict access to a page undergoing migration or to represent a
520 * pfn which has been hwpoisoned and unmapped.
522 static inline bool is_pfn_swap_entry(swp_entry_t entry)
524 /* Make sure the swp offset can always store the needed fields */
525 BUILD_BUG_ON(SWP_TYPE_SHIFT < SWP_PFN_BITS);
527 return is_migration_entry(entry) || is_device_private_entry(entry) ||
528 is_device_exclusive_entry(entry) || is_hwpoison_entry(entry);
531 struct page_vma_mapped_walk;
533 #ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION
534 extern int set_pmd_migration_entry(struct page_vma_mapped_walk *pvmw,
537 extern void remove_migration_pmd(struct page_vma_mapped_walk *pvmw,
540 extern void pmd_migration_entry_wait(struct mm_struct *mm, pmd_t *pmd);
542 static inline swp_entry_t pmd_to_swp_entry(pmd_t pmd)
544 swp_entry_t arch_entry;
546 if (pmd_swp_soft_dirty(pmd))
547 pmd = pmd_swp_clear_soft_dirty(pmd);
548 if (pmd_swp_uffd_wp(pmd))
549 pmd = pmd_swp_clear_uffd_wp(pmd);
550 arch_entry = __pmd_to_swp_entry(pmd);
551 return swp_entry(__swp_type(arch_entry), __swp_offset(arch_entry));
554 static inline pmd_t swp_entry_to_pmd(swp_entry_t entry)
556 swp_entry_t arch_entry;
558 arch_entry = __swp_entry(swp_type(entry), swp_offset(entry));
559 return __swp_entry_to_pmd(arch_entry);
562 static inline int is_pmd_migration_entry(pmd_t pmd)
564 return is_swap_pmd(pmd) && is_migration_entry(pmd_to_swp_entry(pmd));
566 #else /* CONFIG_ARCH_ENABLE_THP_MIGRATION */
567 static inline int set_pmd_migration_entry(struct page_vma_mapped_walk *pvmw,
573 static inline void remove_migration_pmd(struct page_vma_mapped_walk *pvmw,
579 static inline void pmd_migration_entry_wait(struct mm_struct *m, pmd_t *p) { }
581 static inline swp_entry_t pmd_to_swp_entry(pmd_t pmd)
583 return swp_entry(0, 0);
586 static inline pmd_t swp_entry_to_pmd(swp_entry_t entry)
591 static inline int is_pmd_migration_entry(pmd_t pmd)
595 #endif /* CONFIG_ARCH_ENABLE_THP_MIGRATION */
597 static inline int non_swap_entry(swp_entry_t entry)
599 return swp_type(entry) >= MAX_SWAPFILES;
602 #endif /* CONFIG_MMU */
603 #endif /* _LINUX_SWAPOPS_H */