xref: /linux-6.15/include/linux/swapops.h (revision bb7e5ce7)
1 #ifndef _LINUX_SWAPOPS_H
2 #define _LINUX_SWAPOPS_H
3 
4 #include <linux/radix-tree.h>
5 #include <linux/bug.h>
6 
7 /*
8  * swapcache pages are stored in the swapper_space radix tree.  We want to
9  * get good packing density in that tree, so the index should be dense in
10  * the low-order bits.
11  *
12  * We arrange the `type' and `offset' fields so that `type' is at the seven
13  * high-order bits of the swp_entry_t and `offset' is right-aligned in the
14  * remaining bits.  Although `type' itself needs only five bits, we allow for
15  * shmem/tmpfs to shift it all up a further two bits: see swp_to_radix_entry().
16  *
17  * swp_entry_t's are *never* stored anywhere in their arch-dependent format.
18  */
19 #define SWP_TYPE_SHIFT(e)	((sizeof(e.val) * 8) - \
20 			(MAX_SWAPFILES_SHIFT + RADIX_TREE_EXCEPTIONAL_SHIFT))
21 #define SWP_OFFSET_MASK(e)	((1UL << SWP_TYPE_SHIFT(e)) - 1)
22 
23 /*
24  * Store a type+offset into a swp_entry_t in an arch-independent format
25  */
26 static inline swp_entry_t swp_entry(unsigned long type, pgoff_t offset)
27 {
28 	swp_entry_t ret;
29 
30 	ret.val = (type << SWP_TYPE_SHIFT(ret)) |
31 			(offset & SWP_OFFSET_MASK(ret));
32 	return ret;
33 }
34 
35 /*
36  * Extract the `type' field from a swp_entry_t.  The swp_entry_t is in
37  * arch-independent format
38  */
39 static inline unsigned swp_type(swp_entry_t entry)
40 {
41 	return (entry.val >> SWP_TYPE_SHIFT(entry));
42 }
43 
44 /*
45  * Extract the `offset' field from a swp_entry_t.  The swp_entry_t is in
46  * arch-independent format
47  */
48 static inline pgoff_t swp_offset(swp_entry_t entry)
49 {
50 	return entry.val & SWP_OFFSET_MASK(entry);
51 }
52 
53 #ifdef CONFIG_MMU
54 /* check whether a pte points to a swap entry */
55 static inline int is_swap_pte(pte_t pte)
56 {
57 	return !pte_none(pte) && !pte_present(pte);
58 }
59 #endif
60 
61 /*
62  * Convert the arch-dependent pte representation of a swp_entry_t into an
63  * arch-independent swp_entry_t.
64  */
65 static inline swp_entry_t pte_to_swp_entry(pte_t pte)
66 {
67 	swp_entry_t arch_entry;
68 
69 	if (pte_swp_soft_dirty(pte))
70 		pte = pte_swp_clear_soft_dirty(pte);
71 	arch_entry = __pte_to_swp_entry(pte);
72 	return swp_entry(__swp_type(arch_entry), __swp_offset(arch_entry));
73 }
74 
75 /*
76  * Convert the arch-independent representation of a swp_entry_t into the
77  * arch-dependent pte representation.
78  */
79 static inline pte_t swp_entry_to_pte(swp_entry_t entry)
80 {
81 	swp_entry_t arch_entry;
82 
83 	arch_entry = __swp_entry(swp_type(entry), swp_offset(entry));
84 	return __swp_entry_to_pte(arch_entry);
85 }
86 
87 static inline swp_entry_t radix_to_swp_entry(void *arg)
88 {
89 	swp_entry_t entry;
90 
91 	entry.val = (unsigned long)arg >> RADIX_TREE_EXCEPTIONAL_SHIFT;
92 	return entry;
93 }
94 
95 static inline void *swp_to_radix_entry(swp_entry_t entry)
96 {
97 	unsigned long value;
98 
99 	value = entry.val << RADIX_TREE_EXCEPTIONAL_SHIFT;
100 	return (void *)(value | RADIX_TREE_EXCEPTIONAL_ENTRY);
101 }
102 
103 #if IS_ENABLED(CONFIG_DEVICE_PRIVATE)
104 static inline swp_entry_t make_device_private_entry(struct page *page, bool write)
105 {
106 	return swp_entry(write ? SWP_DEVICE_WRITE : SWP_DEVICE_READ,
107 			 page_to_pfn(page));
108 }
109 
110 static inline bool is_device_private_entry(swp_entry_t entry)
111 {
112 	int type = swp_type(entry);
113 	return type == SWP_DEVICE_READ || type == SWP_DEVICE_WRITE;
114 }
115 
116 static inline void make_device_private_entry_read(swp_entry_t *entry)
117 {
118 	*entry = swp_entry(SWP_DEVICE_READ, swp_offset(*entry));
119 }
120 
121 static inline bool is_write_device_private_entry(swp_entry_t entry)
122 {
123 	return unlikely(swp_type(entry) == SWP_DEVICE_WRITE);
124 }
125 
126 static inline struct page *device_private_entry_to_page(swp_entry_t entry)
127 {
128 	return pfn_to_page(swp_offset(entry));
129 }
130 
131 int device_private_entry_fault(struct vm_area_struct *vma,
132 		       unsigned long addr,
133 		       swp_entry_t entry,
134 		       unsigned int flags,
135 		       pmd_t *pmdp);
136 #else /* CONFIG_DEVICE_PRIVATE */
137 static inline swp_entry_t make_device_private_entry(struct page *page, bool write)
138 {
139 	return swp_entry(0, 0);
140 }
141 
142 static inline void make_device_private_entry_read(swp_entry_t *entry)
143 {
144 }
145 
146 static inline bool is_device_private_entry(swp_entry_t entry)
147 {
148 	return false;
149 }
150 
151 static inline bool is_write_device_private_entry(swp_entry_t entry)
152 {
153 	return false;
154 }
155 
156 static inline struct page *device_private_entry_to_page(swp_entry_t entry)
157 {
158 	return NULL;
159 }
160 
161 static inline int device_private_entry_fault(struct vm_area_struct *vma,
162 				     unsigned long addr,
163 				     swp_entry_t entry,
164 				     unsigned int flags,
165 				     pmd_t *pmdp)
166 {
167 	return VM_FAULT_SIGBUS;
168 }
169 #endif /* CONFIG_DEVICE_PRIVATE */
170 
171 #ifdef CONFIG_MIGRATION
172 static inline swp_entry_t make_migration_entry(struct page *page, int write)
173 {
174 	BUG_ON(!PageLocked(compound_head(page)));
175 
176 	return swp_entry(write ? SWP_MIGRATION_WRITE : SWP_MIGRATION_READ,
177 			page_to_pfn(page));
178 }
179 
180 static inline int is_migration_entry(swp_entry_t entry)
181 {
182 	return unlikely(swp_type(entry) == SWP_MIGRATION_READ ||
183 			swp_type(entry) == SWP_MIGRATION_WRITE);
184 }
185 
186 static inline int is_write_migration_entry(swp_entry_t entry)
187 {
188 	return unlikely(swp_type(entry) == SWP_MIGRATION_WRITE);
189 }
190 
191 static inline struct page *migration_entry_to_page(swp_entry_t entry)
192 {
193 	struct page *p = pfn_to_page(swp_offset(entry));
194 	/*
195 	 * Any use of migration entries may only occur while the
196 	 * corresponding page is locked
197 	 */
198 	BUG_ON(!PageLocked(compound_head(p)));
199 	return p;
200 }
201 
202 static inline void make_migration_entry_read(swp_entry_t *entry)
203 {
204 	*entry = swp_entry(SWP_MIGRATION_READ, swp_offset(*entry));
205 }
206 
207 extern void __migration_entry_wait(struct mm_struct *mm, pte_t *ptep,
208 					spinlock_t *ptl);
209 extern void migration_entry_wait(struct mm_struct *mm, pmd_t *pmd,
210 					unsigned long address);
211 extern void migration_entry_wait_huge(struct vm_area_struct *vma,
212 		struct mm_struct *mm, pte_t *pte);
213 #else
214 
215 #define make_migration_entry(page, write) swp_entry(0, 0)
216 static inline int is_migration_entry(swp_entry_t swp)
217 {
218 	return 0;
219 }
220 static inline struct page *migration_entry_to_page(swp_entry_t entry)
221 {
222 	return NULL;
223 }
224 
225 static inline void make_migration_entry_read(swp_entry_t *entryp) { }
226 static inline void __migration_entry_wait(struct mm_struct *mm, pte_t *ptep,
227 					spinlock_t *ptl) { }
228 static inline void migration_entry_wait(struct mm_struct *mm, pmd_t *pmd,
229 					 unsigned long address) { }
230 static inline void migration_entry_wait_huge(struct vm_area_struct *vma,
231 		struct mm_struct *mm, pte_t *pte) { }
232 static inline int is_write_migration_entry(swp_entry_t entry)
233 {
234 	return 0;
235 }
236 
237 #endif
238 
239 struct page_vma_mapped_walk;
240 
241 #ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION
242 extern void set_pmd_migration_entry(struct page_vma_mapped_walk *pvmw,
243 		struct page *page);
244 
245 extern void remove_migration_pmd(struct page_vma_mapped_walk *pvmw,
246 		struct page *new);
247 
248 extern void pmd_migration_entry_wait(struct mm_struct *mm, pmd_t *pmd);
249 
250 static inline swp_entry_t pmd_to_swp_entry(pmd_t pmd)
251 {
252 	swp_entry_t arch_entry;
253 
254 	if (pmd_swp_soft_dirty(pmd))
255 		pmd = pmd_swp_clear_soft_dirty(pmd);
256 	arch_entry = __pmd_to_swp_entry(pmd);
257 	return swp_entry(__swp_type(arch_entry), __swp_offset(arch_entry));
258 }
259 
260 static inline pmd_t swp_entry_to_pmd(swp_entry_t entry)
261 {
262 	swp_entry_t arch_entry;
263 
264 	arch_entry = __swp_entry(swp_type(entry), swp_offset(entry));
265 	return __swp_entry_to_pmd(arch_entry);
266 }
267 
268 static inline int is_pmd_migration_entry(pmd_t pmd)
269 {
270 	return !pmd_present(pmd) && is_migration_entry(pmd_to_swp_entry(pmd));
271 }
272 #else
273 static inline void set_pmd_migration_entry(struct page_vma_mapped_walk *pvmw,
274 		struct page *page)
275 {
276 	BUILD_BUG();
277 }
278 
279 static inline void remove_migration_pmd(struct page_vma_mapped_walk *pvmw,
280 		struct page *new)
281 {
282 	BUILD_BUG();
283 }
284 
285 static inline void pmd_migration_entry_wait(struct mm_struct *m, pmd_t *p) { }
286 
287 static inline swp_entry_t pmd_to_swp_entry(pmd_t pmd)
288 {
289 	return swp_entry(0, 0);
290 }
291 
292 static inline pmd_t swp_entry_to_pmd(swp_entry_t entry)
293 {
294 	return __pmd(0);
295 }
296 
297 static inline int is_pmd_migration_entry(pmd_t pmd)
298 {
299 	return 0;
300 }
301 #endif
302 
303 #ifdef CONFIG_MEMORY_FAILURE
304 
305 extern atomic_long_t num_poisoned_pages __read_mostly;
306 
307 /*
308  * Support for hardware poisoned pages
309  */
310 static inline swp_entry_t make_hwpoison_entry(struct page *page)
311 {
312 	BUG_ON(!PageLocked(page));
313 	return swp_entry(SWP_HWPOISON, page_to_pfn(page));
314 }
315 
316 static inline int is_hwpoison_entry(swp_entry_t entry)
317 {
318 	return swp_type(entry) == SWP_HWPOISON;
319 }
320 
321 static inline bool test_set_page_hwpoison(struct page *page)
322 {
323 	return TestSetPageHWPoison(page);
324 }
325 
326 static inline void num_poisoned_pages_inc(void)
327 {
328 	atomic_long_inc(&num_poisoned_pages);
329 }
330 
331 static inline void num_poisoned_pages_dec(void)
332 {
333 	atomic_long_dec(&num_poisoned_pages);
334 }
335 
336 #else
337 
338 static inline swp_entry_t make_hwpoison_entry(struct page *page)
339 {
340 	return swp_entry(0, 0);
341 }
342 
343 static inline int is_hwpoison_entry(swp_entry_t swp)
344 {
345 	return 0;
346 }
347 
348 static inline bool test_set_page_hwpoison(struct page *page)
349 {
350 	return false;
351 }
352 
353 static inline void num_poisoned_pages_inc(void)
354 {
355 }
356 #endif
357 
358 #if defined(CONFIG_MEMORY_FAILURE) || defined(CONFIG_MIGRATION)
359 static inline int non_swap_entry(swp_entry_t entry)
360 {
361 	return swp_type(entry) >= MAX_SWAPFILES;
362 }
363 #else
364 static inline int non_swap_entry(swp_entry_t entry)
365 {
366 	return 0;
367 }
368 #endif
369 
370 #endif /* _LINUX_SWAPOPS_H */
371