1/* SPDX-License-Identifier: GPL-2.0 */
2#ifndef _LINUX_SWAPOPS_H
3#define _LINUX_SWAPOPS_H
4
5#include <linux/radix-tree.h>
6#include <linux/bug.h>
7#include <linux/mm_types.h>
8
9#ifdef CONFIG_MMU
10
11#ifdef CONFIG_SWAP
12#include <linux/swapfile.h>
13#endif /* CONFIG_SWAP */
14
15/*
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
18 * the low-order bits.
19 *
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().
24 *
25 * swp_entry_t's are *never* stored anywhere in their arch-dependent format.
26 */
27#define SWP_TYPE_SHIFT (BITS_PER_XA_VALUE - MAX_SWAPFILES_SHIFT)
28#define SWP_OFFSET_MASK ((1UL << SWP_TYPE_SHIFT) - 1)
29
30/*
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.
34 */
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, \
40 SWP_TYPE_SHIFT)
41#endif /* MAX_PHYSMEM_BITS */
42#define SWP_PFN_MASK (BIT(SWP_PFN_BITS) - 1)
43
44/**
45 * Migration swap entry specific bitfield definitions. Layout:
46 *
47 * |----------+--------------------|
48 * | swp_type | swp_offset |
49 * |----------+--------+-+-+-------|
50 * | | resv |D|A| PFN |
51 * |----------+--------+-+-+-------|
52 *
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)
55 *
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.
61 */
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)
65
66#define SWP_MIG_YOUNG BIT(SWP_MIG_YOUNG_BIT)
67#define SWP_MIG_DIRTY BIT(SWP_MIG_DIRTY_BIT)
68
69static inline bool is_pfn_swap_entry(swp_entry_t entry);
70
71/* Clear all flags but only keep swp_entry_t related information */
72static inline pte_t pte_swp_clear_flags(pte_t pte)
73{
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);
80 return pte;
81}
82
83/*
84 * Store a type+offset into a swp_entry_t in an arch-independent format
85 */
86static inline swp_entry_t swp_entry(unsigned long type, pgoff_t offset)
87{
88 swp_entry_t ret;
89
90 ret.val = (type << SWP_TYPE_SHIFT) | (offset & SWP_OFFSET_MASK);
91 return ret;
92}
93
94/*
95 * Extract the `type' field from a swp_entry_t. The swp_entry_t is in
96 * arch-independent format
97 */
98static inline unsigned swp_type(swp_entry_t entry)
99{
100 return (entry.val >> SWP_TYPE_SHIFT);
101}
102
103/*
104 * Extract the `offset' field from a swp_entry_t. The swp_entry_t is in
105 * arch-independent format
106 */
107static inline pgoff_t swp_offset(swp_entry_t entry)
108{
109 return entry.val & SWP_OFFSET_MASK;
110}
111
112/*
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
115 * of pfn swap entry.
116 */
117static inline unsigned long swp_offset_pfn(swp_entry_t entry)
118{
119 VM_BUG_ON(!is_pfn_swap_entry(entry));
120 return swp_offset(entry) & SWP_PFN_MASK;
121}
122
123/* check whether a pte points to a swap entry */
124static inline int is_swap_pte(pte_t pte)
125{
126 return !pte_none(pte) && !pte_present(a: pte);
127}
128
129/*
130 * Convert the arch-dependent pte representation of a swp_entry_t into an
131 * arch-independent swp_entry_t.
132 */
133static inline swp_entry_t pte_to_swp_entry(pte_t pte)
134{
135 swp_entry_t arch_entry;
136
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));
140}
141
142/*
143 * Convert the arch-independent representation of a swp_entry_t into the
144 * arch-dependent pte representation.
145 */
146static inline pte_t swp_entry_to_pte(swp_entry_t entry)
147{
148 swp_entry_t arch_entry;
149
150 arch_entry = __swp_entry(swp_type(entry), swp_offset(entry));
151 return __swp_entry_to_pte(arch_entry);
152}
153
154static inline swp_entry_t radix_to_swp_entry(void *arg)
155{
156 swp_entry_t entry;
157
158 entry.val = xa_to_value(entry: arg);
159 return entry;
160}
161
162static inline void *swp_to_radix_entry(swp_entry_t entry)
163{
164 return xa_mk_value(v: entry.val);
165}
166
167#if IS_ENABLED(CONFIG_DEVICE_PRIVATE)
168static inline swp_entry_t make_readable_device_private_entry(pgoff_t offset)
169{
170 return swp_entry(SWP_DEVICE_READ, offset);
171}
172
173static inline swp_entry_t make_writable_device_private_entry(pgoff_t offset)
174{
175 return swp_entry(SWP_DEVICE_WRITE, offset);
176}
177
178static inline bool is_device_private_entry(swp_entry_t entry)
179{
180 int type = swp_type(entry);
181 return type == SWP_DEVICE_READ || type == SWP_DEVICE_WRITE;
182}
183
184static inline bool is_writable_device_private_entry(swp_entry_t entry)
185{
186 return unlikely(swp_type(entry) == SWP_DEVICE_WRITE);
187}
188
189static inline swp_entry_t make_device_exclusive_entry(pgoff_t offset)
190{
191 return swp_entry(SWP_DEVICE_EXCLUSIVE, offset);
192}
193
194static inline bool is_device_exclusive_entry(swp_entry_t entry)
195{
196 return swp_type(entry) == SWP_DEVICE_EXCLUSIVE;
197}
198
199#else /* CONFIG_DEVICE_PRIVATE */
200static inline swp_entry_t make_readable_device_private_entry(pgoff_t offset)
201{
202 return swp_entry(type: 0, offset: 0);
203}
204
205static inline swp_entry_t make_writable_device_private_entry(pgoff_t offset)
206{
207 return swp_entry(type: 0, offset: 0);
208}
209
210static inline bool is_device_private_entry(swp_entry_t entry)
211{
212 return false;
213}
214
215static inline bool is_writable_device_private_entry(swp_entry_t entry)
216{
217 return false;
218}
219
220static inline swp_entry_t make_device_exclusive_entry(pgoff_t offset)
221{
222 return swp_entry(type: 0, offset: 0);
223}
224
225static inline bool is_device_exclusive_entry(swp_entry_t entry)
226{
227 return false;
228}
229
230#endif /* CONFIG_DEVICE_PRIVATE */
231
232#ifdef CONFIG_MIGRATION
233static inline int is_migration_entry(swp_entry_t entry)
234{
235 return unlikely(swp_type(entry) == SWP_MIGRATION_READ ||
236 swp_type(entry) == SWP_MIGRATION_READ_EXCLUSIVE ||
237 swp_type(entry) == SWP_MIGRATION_WRITE);
238}
239
240static inline int is_writable_migration_entry(swp_entry_t entry)
241{
242 return unlikely(swp_type(entry) == SWP_MIGRATION_WRITE);
243}
244
245static inline int is_readable_migration_entry(swp_entry_t entry)
246{
247 return unlikely(swp_type(entry) == SWP_MIGRATION_READ);
248}
249
250static inline int is_readable_exclusive_migration_entry(swp_entry_t entry)
251{
252 return unlikely(swp_type(entry) == SWP_MIGRATION_READ_EXCLUSIVE);
253}
254
255static inline swp_entry_t make_readable_migration_entry(pgoff_t offset)
256{
257 return swp_entry(SWP_MIGRATION_READ, offset);
258}
259
260static inline swp_entry_t make_readable_exclusive_migration_entry(pgoff_t offset)
261{
262 return swp_entry(SWP_MIGRATION_READ_EXCLUSIVE, offset);
263}
264
265static inline swp_entry_t make_writable_migration_entry(pgoff_t offset)
266{
267 return swp_entry(SWP_MIGRATION_WRITE, offset);
268}
269
270/*
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.
274 */
275static inline bool migration_entry_supports_ad(void)
276{
277#ifdef CONFIG_SWAP
278 return swap_migration_ad_supported;
279#else /* CONFIG_SWAP */
280 return false;
281#endif /* CONFIG_SWAP */
282}
283
284static inline swp_entry_t make_migration_entry_young(swp_entry_t entry)
285{
286 if (migration_entry_supports_ad())
287 return swp_entry(type: swp_type(entry),
288 offset: swp_offset(entry) | SWP_MIG_YOUNG);
289 return entry;
290}
291
292static inline bool is_migration_entry_young(swp_entry_t entry)
293{
294 if (migration_entry_supports_ad())
295 return swp_offset(entry) & SWP_MIG_YOUNG;
296 /* Keep the old behavior of aging page after migration */
297 return false;
298}
299
300static inline swp_entry_t make_migration_entry_dirty(swp_entry_t entry)
301{
302 if (migration_entry_supports_ad())
303 return swp_entry(type: swp_type(entry),
304 offset: swp_offset(entry) | SWP_MIG_DIRTY);
305 return entry;
306}
307
308static inline bool is_migration_entry_dirty(swp_entry_t entry)
309{
310 if (migration_entry_supports_ad())
311 return swp_offset(entry) & SWP_MIG_DIRTY;
312 /* Keep the old behavior of clean page after migration */
313 return false;
314}
315
316extern void migration_entry_wait(struct mm_struct *mm, pmd_t *pmd,
317 unsigned long address);
318extern void migration_entry_wait_huge(struct vm_area_struct *vma, unsigned long addr, pte_t *pte);
319#else /* CONFIG_MIGRATION */
320static inline swp_entry_t make_readable_migration_entry(pgoff_t offset)
321{
322 return swp_entry(0, 0);
323}
324
325static inline swp_entry_t make_readable_exclusive_migration_entry(pgoff_t offset)
326{
327 return swp_entry(0, 0);
328}
329
330static inline swp_entry_t make_writable_migration_entry(pgoff_t offset)
331{
332 return swp_entry(0, 0);
333}
334
335static inline int is_migration_entry(swp_entry_t swp)
336{
337 return 0;
338}
339
340static inline void migration_entry_wait(struct mm_struct *mm, pmd_t *pmd,
341 unsigned long address) { }
342static inline void migration_entry_wait_huge(struct vm_area_struct *vma,
343 unsigned long addr, pte_t *pte) { }
344static inline int is_writable_migration_entry(swp_entry_t entry)
345{
346 return 0;
347}
348static inline int is_readable_migration_entry(swp_entry_t entry)
349{
350 return 0;
351}
352
353static inline swp_entry_t make_migration_entry_young(swp_entry_t entry)
354{
355 return entry;
356}
357
358static inline bool is_migration_entry_young(swp_entry_t entry)
359{
360 return false;
361}
362
363static inline swp_entry_t make_migration_entry_dirty(swp_entry_t entry)
364{
365 return entry;
366}
367
368static inline bool is_migration_entry_dirty(swp_entry_t entry)
369{
370 return false;
371}
372#endif /* CONFIG_MIGRATION */
373
374#ifdef CONFIG_MEMORY_FAILURE
375
376/*
377 * Support for hardware poisoned pages
378 */
379static inline swp_entry_t make_hwpoison_entry(struct page *page)
380{
381 BUG_ON(!PageLocked(page));
382 return swp_entry(SWP_HWPOISON, page_to_pfn(page));
383}
384
385static inline int is_hwpoison_entry(swp_entry_t entry)
386{
387 return swp_type(entry) == SWP_HWPOISON;
388}
389
390#else
391
392static inline swp_entry_t make_hwpoison_entry(struct page *page)
393{
394 return swp_entry(type: 0, offset: 0);
395}
396
397static inline int is_hwpoison_entry(swp_entry_t swp)
398{
399 return 0;
400}
401#endif
402
403typedef unsigned long pte_marker;
404
405#define PTE_MARKER_UFFD_WP BIT(0)
406/*
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.
410 *
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
413 * logic.
414 */
415#define PTE_MARKER_POISONED BIT(1)
416/*
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.
420 */
421#define PTE_MARKER_GUARD BIT(2)
422#define PTE_MARKER_MASK (BIT(3) - 1)
423
424static inline swp_entry_t make_pte_marker_entry(pte_marker marker)
425{
426 return swp_entry(SWP_PTE_MARKER, offset: marker);
427}
428
429static inline bool is_pte_marker_entry(swp_entry_t entry)
430{
431 return swp_type(entry) == SWP_PTE_MARKER;
432}
433
434static inline pte_marker pte_marker_get(swp_entry_t entry)
435{
436 return swp_offset(entry) & PTE_MARKER_MASK;
437}
438
439static inline bool is_pte_marker(pte_t pte)
440{
441 return is_swap_pte(pte) && is_pte_marker_entry(entry: pte_to_swp_entry(pte));
442}
443
444static inline pte_t make_pte_marker(pte_marker marker)
445{
446 return swp_entry_to_pte(entry: make_pte_marker_entry(marker));
447}
448
449static inline swp_entry_t make_poisoned_swp_entry(void)
450{
451 return make_pte_marker_entry(PTE_MARKER_POISONED);
452}
453
454static inline int is_poisoned_swp_entry(swp_entry_t entry)
455{
456 return is_pte_marker_entry(entry) &&
457 (pte_marker_get(entry) & PTE_MARKER_POISONED);
458
459}
460
461static inline swp_entry_t make_guard_swp_entry(void)
462{
463 return make_pte_marker_entry(PTE_MARKER_GUARD);
464}
465
466static inline int is_guard_swp_entry(swp_entry_t entry)
467{
468 return is_pte_marker_entry(entry) &&
469 (pte_marker_get(entry) & PTE_MARKER_GUARD);
470}
471
472/*
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.
477 *
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.
484 */
485static inline int pte_none_mostly(pte_t pte)
486{
487 return pte_none(pte) || is_pte_marker(pte);
488}
489
490static inline struct page *pfn_swap_entry_to_page(swp_entry_t entry)
491{
492 struct page *p = pfn_to_page(swp_offset_pfn(entry));
493
494 /*
495 * Any use of migration entries may only occur while the
496 * corresponding page is locked
497 */
498 BUG_ON(is_migration_entry(entry) && !PageLocked(p));
499
500 return p;
501}
502
503static inline struct folio *pfn_swap_entry_folio(swp_entry_t entry)
504{
505 struct folio *folio = pfn_folio(pfn: swp_offset_pfn(entry));
506
507 /*
508 * Any use of migration entries may only occur while the
509 * corresponding folio is locked
510 */
511 BUG_ON(is_migration_entry(entry) && !folio_test_locked(folio));
512
513 return folio;
514}
515
516/*
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.
521 */
522static inline bool is_pfn_swap_entry(swp_entry_t entry)
523{
524 /* Make sure the swp offset can always store the needed fields */
525 BUILD_BUG_ON(SWP_TYPE_SHIFT < SWP_PFN_BITS);
526
527 return is_migration_entry(entry) || is_device_private_entry(entry) ||
528 is_device_exclusive_entry(entry) || is_hwpoison_entry(swp: entry);
529}
530
531struct page_vma_mapped_walk;
532
533#ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION
534extern int set_pmd_migration_entry(struct page_vma_mapped_walk *pvmw,
535 struct page *page);
536
537extern void remove_migration_pmd(struct page_vma_mapped_walk *pvmw,
538 struct page *new);
539
540extern void pmd_migration_entry_wait(struct mm_struct *mm, pmd_t *pmd);
541
542static inline swp_entry_t pmd_to_swp_entry(pmd_t pmd)
543{
544 swp_entry_t arch_entry;
545
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));
552}
553
554static inline pmd_t swp_entry_to_pmd(swp_entry_t entry)
555{
556 swp_entry_t arch_entry;
557
558 arch_entry = __swp_entry(swp_type(entry), swp_offset(entry));
559 return __swp_entry_to_pmd(arch_entry);
560}
561
562static inline int is_pmd_migration_entry(pmd_t pmd)
563{
564 return is_swap_pmd(pmd) && is_migration_entry(pmd_to_swp_entry(pmd));
565}
566#else /* CONFIG_ARCH_ENABLE_THP_MIGRATION */
567static inline int set_pmd_migration_entry(struct page_vma_mapped_walk *pvmw,
568 struct page *page)
569{
570 BUILD_BUG();
571}
572
573static inline void remove_migration_pmd(struct page_vma_mapped_walk *pvmw,
574 struct page *new)
575{
576 BUILD_BUG();
577}
578
579static inline void pmd_migration_entry_wait(struct mm_struct *m, pmd_t *p) { }
580
581static inline swp_entry_t pmd_to_swp_entry(pmd_t pmd)
582{
583 return swp_entry(type: 0, offset: 0);
584}
585
586static inline pmd_t swp_entry_to_pmd(swp_entry_t entry)
587{
588 return __pmd(0);
589}
590
591static inline int is_pmd_migration_entry(pmd_t pmd)
592{
593 return 0;
594}
595#endif /* CONFIG_ARCH_ENABLE_THP_MIGRATION */
596
597static inline int non_swap_entry(swp_entry_t entry)
598{
599 return swp_type(entry) >= MAX_SWAPFILES;
600}
601
602#endif /* CONFIG_MMU */
603#endif /* _LINUX_SWAPOPS_H */
604