Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
tjh.dev
kernel
os
linux
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/*
12 * swapcache pages are stored in the swapper_space radix tree. We want to
13 * get good packing density in that tree, so the index should be dense in
14 * the low-order bits.
15 *
16 * We arrange the `type' and `offset' fields so that `type' is at the six
17 * high-order bits of the swp_entry_t and `offset' is right-aligned in the
18 * remaining bits. Although `type' itself needs only five bits, we allow for
19 * shmem/tmpfs to shift it all up a further one bit: see swp_to_radix_entry().
20 *
21 * swp_entry_t's are *never* stored anywhere in their arch-dependent format.
22 */
23#define SWP_TYPE_SHIFT (BITS_PER_XA_VALUE - MAX_SWAPFILES_SHIFT)
24#define SWP_OFFSET_MASK ((1UL << SWP_TYPE_SHIFT) - 1)
25
26/* Clear all flags but only keep swp_entry_t related information */
27static inline pte_t pte_swp_clear_flags(pte_t pte)
28{
29 if (pte_swp_exclusive(pte))
30 pte = pte_swp_clear_exclusive(pte);
31 if (pte_swp_soft_dirty(pte))
32 pte = pte_swp_clear_soft_dirty(pte);
33 if (pte_swp_uffd_wp(pte))
34 pte = pte_swp_clear_uffd_wp(pte);
35 return pte;
36}
37
38/*
39 * Store a type+offset into a swp_entry_t in an arch-independent format
40 */
41static inline swp_entry_t swp_entry(unsigned long type, pgoff_t offset)
42{
43 swp_entry_t ret;
44
45 ret.val = (type << SWP_TYPE_SHIFT) | (offset & SWP_OFFSET_MASK);
46 return ret;
47}
48
49/*
50 * Extract the `type' field from a swp_entry_t. The swp_entry_t is in
51 * arch-independent format
52 */
53static inline unsigned swp_type(swp_entry_t entry)
54{
55 return (entry.val >> SWP_TYPE_SHIFT);
56}
57
58/*
59 * Extract the `offset' field from a swp_entry_t. The swp_entry_t is in
60 * arch-independent format
61 */
62static inline pgoff_t swp_offset(swp_entry_t entry)
63{
64 return entry.val & SWP_OFFSET_MASK;
65}
66
67/* check whether a pte points to a swap entry */
68static inline int is_swap_pte(pte_t pte)
69{
70 return !pte_none(pte) && !pte_present(pte);
71}
72
73/*
74 * Convert the arch-dependent pte representation of a swp_entry_t into an
75 * arch-independent swp_entry_t.
76 */
77static inline swp_entry_t pte_to_swp_entry(pte_t pte)
78{
79 swp_entry_t arch_entry;
80
81 pte = pte_swp_clear_flags(pte);
82 arch_entry = __pte_to_swp_entry(pte);
83 return swp_entry(__swp_type(arch_entry), __swp_offset(arch_entry));
84}
85
86/*
87 * Convert the arch-independent representation of a swp_entry_t into the
88 * arch-dependent pte representation.
89 */
90static inline pte_t swp_entry_to_pte(swp_entry_t entry)
91{
92 swp_entry_t arch_entry;
93
94 arch_entry = __swp_entry(swp_type(entry), swp_offset(entry));
95 return __swp_entry_to_pte(arch_entry);
96}
97
98static inline swp_entry_t radix_to_swp_entry(void *arg)
99{
100 swp_entry_t entry;
101
102 entry.val = xa_to_value(arg);
103 return entry;
104}
105
106static inline void *swp_to_radix_entry(swp_entry_t entry)
107{
108 return xa_mk_value(entry.val);
109}
110
111static inline swp_entry_t make_swapin_error_entry(struct page *page)
112{
113 return swp_entry(SWP_SWAPIN_ERROR, page_to_pfn(page));
114}
115
116static inline int is_swapin_error_entry(swp_entry_t entry)
117{
118 return swp_type(entry) == SWP_SWAPIN_ERROR;
119}
120
121#if IS_ENABLED(CONFIG_DEVICE_PRIVATE)
122static inline swp_entry_t make_readable_device_private_entry(pgoff_t offset)
123{
124 return swp_entry(SWP_DEVICE_READ, offset);
125}
126
127static inline swp_entry_t make_writable_device_private_entry(pgoff_t offset)
128{
129 return swp_entry(SWP_DEVICE_WRITE, offset);
130}
131
132static inline bool is_device_private_entry(swp_entry_t entry)
133{
134 int type = swp_type(entry);
135 return type == SWP_DEVICE_READ || type == SWP_DEVICE_WRITE;
136}
137
138static inline bool is_writable_device_private_entry(swp_entry_t entry)
139{
140 return unlikely(swp_type(entry) == SWP_DEVICE_WRITE);
141}
142
143static inline swp_entry_t make_readable_device_exclusive_entry(pgoff_t offset)
144{
145 return swp_entry(SWP_DEVICE_EXCLUSIVE_READ, offset);
146}
147
148static inline swp_entry_t make_writable_device_exclusive_entry(pgoff_t offset)
149{
150 return swp_entry(SWP_DEVICE_EXCLUSIVE_WRITE, offset);
151}
152
153static inline bool is_device_exclusive_entry(swp_entry_t entry)
154{
155 return swp_type(entry) == SWP_DEVICE_EXCLUSIVE_READ ||
156 swp_type(entry) == SWP_DEVICE_EXCLUSIVE_WRITE;
157}
158
159static inline bool is_writable_device_exclusive_entry(swp_entry_t entry)
160{
161 return unlikely(swp_type(entry) == SWP_DEVICE_EXCLUSIVE_WRITE);
162}
163#else /* CONFIG_DEVICE_PRIVATE */
164static inline swp_entry_t make_readable_device_private_entry(pgoff_t offset)
165{
166 return swp_entry(0, 0);
167}
168
169static inline swp_entry_t make_writable_device_private_entry(pgoff_t offset)
170{
171 return swp_entry(0, 0);
172}
173
174static inline bool is_device_private_entry(swp_entry_t entry)
175{
176 return false;
177}
178
179static inline bool is_writable_device_private_entry(swp_entry_t entry)
180{
181 return false;
182}
183
184static inline swp_entry_t make_readable_device_exclusive_entry(pgoff_t offset)
185{
186 return swp_entry(0, 0);
187}
188
189static inline swp_entry_t make_writable_device_exclusive_entry(pgoff_t offset)
190{
191 return swp_entry(0, 0);
192}
193
194static inline bool is_device_exclusive_entry(swp_entry_t entry)
195{
196 return false;
197}
198
199static inline bool is_writable_device_exclusive_entry(swp_entry_t entry)
200{
201 return false;
202}
203#endif /* CONFIG_DEVICE_PRIVATE */
204
205#ifdef CONFIG_MIGRATION
206static inline int is_migration_entry(swp_entry_t entry)
207{
208 return unlikely(swp_type(entry) == SWP_MIGRATION_READ ||
209 swp_type(entry) == SWP_MIGRATION_READ_EXCLUSIVE ||
210 swp_type(entry) == SWP_MIGRATION_WRITE);
211}
212
213static inline int is_writable_migration_entry(swp_entry_t entry)
214{
215 return unlikely(swp_type(entry) == SWP_MIGRATION_WRITE);
216}
217
218static inline int is_readable_migration_entry(swp_entry_t entry)
219{
220 return unlikely(swp_type(entry) == SWP_MIGRATION_READ);
221}
222
223static inline int is_readable_exclusive_migration_entry(swp_entry_t entry)
224{
225 return unlikely(swp_type(entry) == SWP_MIGRATION_READ_EXCLUSIVE);
226}
227
228static inline swp_entry_t make_readable_migration_entry(pgoff_t offset)
229{
230 return swp_entry(SWP_MIGRATION_READ, offset);
231}
232
233static inline swp_entry_t make_readable_exclusive_migration_entry(pgoff_t offset)
234{
235 return swp_entry(SWP_MIGRATION_READ_EXCLUSIVE, offset);
236}
237
238static inline swp_entry_t make_writable_migration_entry(pgoff_t offset)
239{
240 return swp_entry(SWP_MIGRATION_WRITE, offset);
241}
242
243extern void __migration_entry_wait(struct mm_struct *mm, pte_t *ptep,
244 spinlock_t *ptl);
245extern void migration_entry_wait(struct mm_struct *mm, pmd_t *pmd,
246 unsigned long address);
247extern void migration_entry_wait_huge(struct vm_area_struct *vma,
248 struct mm_struct *mm, pte_t *pte);
249#else
250static inline swp_entry_t make_readable_migration_entry(pgoff_t offset)
251{
252 return swp_entry(0, 0);
253}
254
255static inline swp_entry_t make_readable_exclusive_migration_entry(pgoff_t offset)
256{
257 return swp_entry(0, 0);
258}
259
260static inline swp_entry_t make_writable_migration_entry(pgoff_t offset)
261{
262 return swp_entry(0, 0);
263}
264
265static inline int is_migration_entry(swp_entry_t swp)
266{
267 return 0;
268}
269
270static inline void __migration_entry_wait(struct mm_struct *mm, pte_t *ptep,
271 spinlock_t *ptl) { }
272static inline void migration_entry_wait(struct mm_struct *mm, pmd_t *pmd,
273 unsigned long address) { }
274static inline void migration_entry_wait_huge(struct vm_area_struct *vma,
275 struct mm_struct *mm, pte_t *pte) { }
276static inline int is_writable_migration_entry(swp_entry_t entry)
277{
278 return 0;
279}
280static inline int is_readable_migration_entry(swp_entry_t entry)
281{
282 return 0;
283}
284
285#endif
286
287typedef unsigned long pte_marker;
288
289#define PTE_MARKER_UFFD_WP BIT(0)
290#define PTE_MARKER_MASK (PTE_MARKER_UFFD_WP)
291
292#ifdef CONFIG_PTE_MARKER
293
294static inline swp_entry_t make_pte_marker_entry(pte_marker marker)
295{
296 return swp_entry(SWP_PTE_MARKER, marker);
297}
298
299static inline bool is_pte_marker_entry(swp_entry_t entry)
300{
301 return swp_type(entry) == SWP_PTE_MARKER;
302}
303
304static inline pte_marker pte_marker_get(swp_entry_t entry)
305{
306 return swp_offset(entry) & PTE_MARKER_MASK;
307}
308
309static inline bool is_pte_marker(pte_t pte)
310{
311 return is_swap_pte(pte) && is_pte_marker_entry(pte_to_swp_entry(pte));
312}
313
314#else /* CONFIG_PTE_MARKER */
315
316static inline swp_entry_t make_pte_marker_entry(pte_marker marker)
317{
318 /* This should never be called if !CONFIG_PTE_MARKER */
319 WARN_ON_ONCE(1);
320 return swp_entry(0, 0);
321}
322
323static inline bool is_pte_marker_entry(swp_entry_t entry)
324{
325 return false;
326}
327
328static inline pte_marker pte_marker_get(swp_entry_t entry)
329{
330 return 0;
331}
332
333static inline bool is_pte_marker(pte_t pte)
334{
335 return false;
336}
337
338#endif /* CONFIG_PTE_MARKER */
339
340static inline pte_t make_pte_marker(pte_marker marker)
341{
342 return swp_entry_to_pte(make_pte_marker_entry(marker));
343}
344
345/*
346 * This is a special version to check pte_none() just to cover the case when
347 * the pte is a pte marker. It existed because in many cases the pte marker
348 * should be seen as a none pte; it's just that we have stored some information
349 * onto the none pte so it becomes not-none any more.
350 *
351 * It should be used when the pte is file-backed, ram-based and backing
352 * userspace pages, like shmem. It is not needed upon pgtables that do not
353 * support pte markers at all. For example, it's not needed on anonymous
354 * memory, kernel-only memory (including when the system is during-boot),
355 * non-ram based generic file-system. It's fine to be used even there, but the
356 * extra pte marker check will be pure overhead.
357 *
358 * For systems configured with !CONFIG_PTE_MARKER this will be automatically
359 * optimized to pte_none().
360 */
361static inline int pte_none_mostly(pte_t pte)
362{
363 return pte_none(pte) || is_pte_marker(pte);
364}
365
366static inline struct page *pfn_swap_entry_to_page(swp_entry_t entry)
367{
368 struct page *p = pfn_to_page(swp_offset(entry));
369
370 /*
371 * Any use of migration entries may only occur while the
372 * corresponding page is locked
373 */
374 BUG_ON(is_migration_entry(entry) && !PageLocked(p));
375
376 return p;
377}
378
379/*
380 * A pfn swap entry is a special type of swap entry that always has a pfn stored
381 * in the swap offset. They are used to represent unaddressable device memory
382 * and to restrict access to a page undergoing migration.
383 */
384static inline bool is_pfn_swap_entry(swp_entry_t entry)
385{
386 return is_migration_entry(entry) || is_device_private_entry(entry) ||
387 is_device_exclusive_entry(entry);
388}
389
390struct page_vma_mapped_walk;
391
392#ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION
393extern int set_pmd_migration_entry(struct page_vma_mapped_walk *pvmw,
394 struct page *page);
395
396extern void remove_migration_pmd(struct page_vma_mapped_walk *pvmw,
397 struct page *new);
398
399extern void pmd_migration_entry_wait(struct mm_struct *mm, pmd_t *pmd);
400
401static inline swp_entry_t pmd_to_swp_entry(pmd_t pmd)
402{
403 swp_entry_t arch_entry;
404
405 if (pmd_swp_soft_dirty(pmd))
406 pmd = pmd_swp_clear_soft_dirty(pmd);
407 if (pmd_swp_uffd_wp(pmd))
408 pmd = pmd_swp_clear_uffd_wp(pmd);
409 arch_entry = __pmd_to_swp_entry(pmd);
410 return swp_entry(__swp_type(arch_entry), __swp_offset(arch_entry));
411}
412
413static inline pmd_t swp_entry_to_pmd(swp_entry_t entry)
414{
415 swp_entry_t arch_entry;
416
417 arch_entry = __swp_entry(swp_type(entry), swp_offset(entry));
418 return __swp_entry_to_pmd(arch_entry);
419}
420
421static inline int is_pmd_migration_entry(pmd_t pmd)
422{
423 return is_swap_pmd(pmd) && is_migration_entry(pmd_to_swp_entry(pmd));
424}
425#else
426static inline int set_pmd_migration_entry(struct page_vma_mapped_walk *pvmw,
427 struct page *page)
428{
429 BUILD_BUG();
430}
431
432static inline void remove_migration_pmd(struct page_vma_mapped_walk *pvmw,
433 struct page *new)
434{
435 BUILD_BUG();
436}
437
438static inline void pmd_migration_entry_wait(struct mm_struct *m, pmd_t *p) { }
439
440static inline swp_entry_t pmd_to_swp_entry(pmd_t pmd)
441{
442 return swp_entry(0, 0);
443}
444
445static inline pmd_t swp_entry_to_pmd(swp_entry_t entry)
446{
447 return __pmd(0);
448}
449
450static inline int is_pmd_migration_entry(pmd_t pmd)
451{
452 return 0;
453}
454#endif
455
456#ifdef CONFIG_MEMORY_FAILURE
457
458extern atomic_long_t num_poisoned_pages __read_mostly;
459
460/*
461 * Support for hardware poisoned pages
462 */
463static inline swp_entry_t make_hwpoison_entry(struct page *page)
464{
465 BUG_ON(!PageLocked(page));
466 return swp_entry(SWP_HWPOISON, page_to_pfn(page));
467}
468
469static inline int is_hwpoison_entry(swp_entry_t entry)
470{
471 return swp_type(entry) == SWP_HWPOISON;
472}
473
474static inline unsigned long hwpoison_entry_to_pfn(swp_entry_t entry)
475{
476 return swp_offset(entry);
477}
478
479static inline void num_poisoned_pages_inc(void)
480{
481 atomic_long_inc(&num_poisoned_pages);
482}
483
484static inline void num_poisoned_pages_dec(void)
485{
486 atomic_long_dec(&num_poisoned_pages);
487}
488
489#else
490
491static inline swp_entry_t make_hwpoison_entry(struct page *page)
492{
493 return swp_entry(0, 0);
494}
495
496static inline int is_hwpoison_entry(swp_entry_t swp)
497{
498 return 0;
499}
500
501static inline void num_poisoned_pages_inc(void)
502{
503}
504#endif
505
506static inline int non_swap_entry(swp_entry_t entry)
507{
508 return swp_type(entry) >= MAX_SWAPFILES;
509}
510
511#endif /* CONFIG_MMU */
512#endif /* _LINUX_SWAPOPS_H */