1 /* SPDX-License-Identifier: GPL-2.0 */ 1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef _LINUX_RMAP_H 2 #ifndef _LINUX_RMAP_H
3 #define _LINUX_RMAP_H 3 #define _LINUX_RMAP_H
4 /* 4 /*
5 * Declarations for Reverse Mapping functions 5 * Declarations for Reverse Mapping functions in mm/rmap.c
6 */ 6 */
7 7
8 #include <linux/list.h> 8 #include <linux/list.h>
9 #include <linux/slab.h> 9 #include <linux/slab.h>
10 #include <linux/mm.h> 10 #include <linux/mm.h>
11 #include <linux/rwsem.h> 11 #include <linux/rwsem.h>
12 #include <linux/memcontrol.h> 12 #include <linux/memcontrol.h>
13 #include <linux/highmem.h> 13 #include <linux/highmem.h>
14 #include <linux/pagemap.h> <<
15 #include <linux/memremap.h> <<
16 14
17 /* 15 /*
18 * The anon_vma heads a list of private "relat 16 * The anon_vma heads a list of private "related" vmas, to scan if
19 * an anonymous page pointing to this anon_vma 17 * an anonymous page pointing to this anon_vma needs to be unmapped:
20 * the vmas on the list will be related by for 18 * the vmas on the list will be related by forking, or by splitting.
21 * 19 *
22 * Since vmas come and go as they are split an 20 * Since vmas come and go as they are split and merged (particularly
23 * in mprotect), the mapping field of an anony 21 * in mprotect), the mapping field of an anonymous page cannot point
24 * directly to a vma: instead it points to an 22 * directly to a vma: instead it points to an anon_vma, on whose list
25 * the related vmas can be easily linked or un 23 * the related vmas can be easily linked or unlinked.
26 * 24 *
27 * After unlinking the last vma on the list, w 25 * After unlinking the last vma on the list, we must garbage collect
28 * the anon_vma object itself: we're guarantee 26 * the anon_vma object itself: we're guaranteed no page can be
29 * pointing to this anon_vma once its vma list 27 * pointing to this anon_vma once its vma list is empty.
30 */ 28 */
31 struct anon_vma { 29 struct anon_vma {
32 struct anon_vma *root; /* Roo 30 struct anon_vma *root; /* Root of this anon_vma tree */
33 struct rw_semaphore rwsem; /* W: 31 struct rw_semaphore rwsem; /* W: modification, R: walking the list */
34 /* 32 /*
35 * The refcount is taken on an anon_vm 33 * The refcount is taken on an anon_vma when there is no
36 * guarantee that the vma of page tabl 34 * guarantee that the vma of page tables will exist for
37 * the duration of the operation. A ca 35 * the duration of the operation. A caller that takes
38 * the reference is responsible for cl 36 * the reference is responsible for clearing up the
39 * anon_vma if they are the last user 37 * anon_vma if they are the last user on release
40 */ 38 */
41 atomic_t refcount; 39 atomic_t refcount;
42 40
43 /* 41 /*
44 * Count of child anon_vmas. Equals to !! 42 * Count of child anon_vmas and VMAs which points to this anon_vma.
45 * have ->parent pointing to this one, <<
46 * 43 *
47 * This counter is used for making dec 44 * This counter is used for making decision about reusing anon_vma
48 * instead of forking new one. See com 45 * instead of forking new one. See comments in function anon_vma_clone.
49 */ 46 */
50 unsigned long num_children; !! 47 unsigned degree;
51 /* Count of VMAs whose ->anon_vma poin <<
52 unsigned long num_active_vmas; <<
53 48
54 struct anon_vma *parent; /* Par 49 struct anon_vma *parent; /* Parent of this anon_vma */
55 50
56 /* 51 /*
57 * NOTE: the LSB of the rb_root.rb_nod 52 * NOTE: the LSB of the rb_root.rb_node is set by
58 * mm_take_all_locks() _after_ taking 53 * mm_take_all_locks() _after_ taking the above lock. So the
59 * rb_root must only be read/written a 54 * rb_root must only be read/written after taking the above lock
60 * to be sure to see a valid next poin 55 * to be sure to see a valid next pointer. The LSB bit itself
61 * is serialized by a system wide lock 56 * is serialized by a system wide lock only visible to
62 * mm_take_all_locks() (mm_all_locks_m 57 * mm_take_all_locks() (mm_all_locks_mutex).
63 */ 58 */
64 59
65 /* Interval tree of private "related" 60 /* Interval tree of private "related" vmas */
66 struct rb_root_cached rb_root; 61 struct rb_root_cached rb_root;
67 }; 62 };
68 63
69 /* 64 /*
70 * The copy-on-write semantics of fork mean th 65 * The copy-on-write semantics of fork mean that an anon_vma
71 * can become associated with multiple process 66 * can become associated with multiple processes. Furthermore,
72 * each child process will have its own anon_v 67 * each child process will have its own anon_vma, where new
73 * pages for that process are instantiated. 68 * pages for that process are instantiated.
74 * 69 *
75 * This structure allows us to find the anon_v 70 * This structure allows us to find the anon_vmas associated
76 * with a VMA, or the VMAs associated with an 71 * with a VMA, or the VMAs associated with an anon_vma.
77 * The "same_vma" list contains the anon_vma_c 72 * The "same_vma" list contains the anon_vma_chains linking
78 * all the anon_vmas associated with this VMA. 73 * all the anon_vmas associated with this VMA.
79 * The "rb" field indexes on an interval tree 74 * The "rb" field indexes on an interval tree the anon_vma_chains
80 * which link all the VMAs associated with thi 75 * which link all the VMAs associated with this anon_vma.
81 */ 76 */
82 struct anon_vma_chain { 77 struct anon_vma_chain {
83 struct vm_area_struct *vma; 78 struct vm_area_struct *vma;
84 struct anon_vma *anon_vma; 79 struct anon_vma *anon_vma;
85 struct list_head same_vma; /* locked !! 80 struct list_head same_vma; /* locked by mmap_sem & page_table_lock */
86 struct rb_node rb; 81 struct rb_node rb; /* locked by anon_vma->rwsem */
87 unsigned long rb_subtree_last; 82 unsigned long rb_subtree_last;
88 #ifdef CONFIG_DEBUG_VM_RB 83 #ifdef CONFIG_DEBUG_VM_RB
89 unsigned long cached_vma_start, cached 84 unsigned long cached_vma_start, cached_vma_last;
90 #endif 85 #endif
91 }; 86 };
92 87
93 enum ttu_flags { 88 enum ttu_flags {
>> 89 TTU_MIGRATION = 0x1, /* migration mode */
>> 90 TTU_MUNLOCK = 0x2, /* munlock mode */
>> 91
94 TTU_SPLIT_HUGE_PMD = 0x4, /* spl 92 TTU_SPLIT_HUGE_PMD = 0x4, /* split huge PMD if any */
95 TTU_IGNORE_MLOCK = 0x8, /* ign 93 TTU_IGNORE_MLOCK = 0x8, /* ignore mlock */
96 TTU_SYNC = 0x10, /* avo !! 94 TTU_IGNORE_ACCESS = 0x10, /* don't age */
97 TTU_HWPOISON = 0x20, /* do !! 95 TTU_IGNORE_HWPOISON = 0x20, /* corrupted page is recoverable */
98 TTU_BATCH_FLUSH = 0x40, /* Bat 96 TTU_BATCH_FLUSH = 0x40, /* Batch TLB flushes where possible
99 * and 97 * and caller guarantees they will
100 * do 98 * do a final flush if necessary */
101 TTU_RMAP_LOCKED = 0x80, /* do 99 TTU_RMAP_LOCKED = 0x80, /* do not grab rmap lock:
102 * cal 100 * caller holds it */
>> 101 TTU_SPLIT_FREEZE = 0x100, /* freeze pte under splitting thp */
103 }; 102 };
104 103
105 #ifdef CONFIG_MMU 104 #ifdef CONFIG_MMU
106 static inline void get_anon_vma(struct anon_vm 105 static inline void get_anon_vma(struct anon_vma *anon_vma)
107 { 106 {
108 atomic_inc(&anon_vma->refcount); 107 atomic_inc(&anon_vma->refcount);
109 } 108 }
110 109
111 void __put_anon_vma(struct anon_vma *anon_vma) 110 void __put_anon_vma(struct anon_vma *anon_vma);
112 111
113 static inline void put_anon_vma(struct anon_vm 112 static inline void put_anon_vma(struct anon_vma *anon_vma)
114 { 113 {
115 if (atomic_dec_and_test(&anon_vma->ref 114 if (atomic_dec_and_test(&anon_vma->refcount))
116 __put_anon_vma(anon_vma); 115 __put_anon_vma(anon_vma);
117 } 116 }
118 117
119 static inline void anon_vma_lock_write(struct 118 static inline void anon_vma_lock_write(struct anon_vma *anon_vma)
120 { 119 {
121 down_write(&anon_vma->root->rwsem); 120 down_write(&anon_vma->root->rwsem);
122 } 121 }
123 122
124 static inline int anon_vma_trylock_write(struc <<
125 { <<
126 return down_write_trylock(&anon_vma->r <<
127 } <<
128 <<
129 static inline void anon_vma_unlock_write(struc 123 static inline void anon_vma_unlock_write(struct anon_vma *anon_vma)
130 { 124 {
131 up_write(&anon_vma->root->rwsem); 125 up_write(&anon_vma->root->rwsem);
132 } 126 }
133 127
134 static inline void anon_vma_lock_read(struct a 128 static inline void anon_vma_lock_read(struct anon_vma *anon_vma)
135 { 129 {
136 down_read(&anon_vma->root->rwsem); 130 down_read(&anon_vma->root->rwsem);
137 } 131 }
138 132
139 static inline int anon_vma_trylock_read(struct <<
140 { <<
141 return down_read_trylock(&anon_vma->ro <<
142 } <<
143 <<
144 static inline void anon_vma_unlock_read(struct 133 static inline void anon_vma_unlock_read(struct anon_vma *anon_vma)
145 { 134 {
146 up_read(&anon_vma->root->rwsem); 135 up_read(&anon_vma->root->rwsem);
147 } 136 }
148 137
149 138
150 /* 139 /*
151 * anon_vma helper functions. 140 * anon_vma helper functions.
152 */ 141 */
153 void anon_vma_init(void); /* create anon 142 void anon_vma_init(void); /* create anon_vma_cachep */
154 int __anon_vma_prepare(struct vm_area_struct 143 int __anon_vma_prepare(struct vm_area_struct *);
155 void unlink_anon_vmas(struct vm_area_struct *) 144 void unlink_anon_vmas(struct vm_area_struct *);
156 int anon_vma_clone(struct vm_area_struct *, st 145 int anon_vma_clone(struct vm_area_struct *, struct vm_area_struct *);
157 int anon_vma_fork(struct vm_area_struct *, str 146 int anon_vma_fork(struct vm_area_struct *, struct vm_area_struct *);
158 147
159 static inline int anon_vma_prepare(struct vm_a 148 static inline int anon_vma_prepare(struct vm_area_struct *vma)
160 { 149 {
161 if (likely(vma->anon_vma)) 150 if (likely(vma->anon_vma))
162 return 0; 151 return 0;
163 152
164 return __anon_vma_prepare(vma); 153 return __anon_vma_prepare(vma);
165 } 154 }
166 155
167 static inline void anon_vma_merge(struct vm_ar 156 static inline void anon_vma_merge(struct vm_area_struct *vma,
168 struct vm_ar 157 struct vm_area_struct *next)
169 { 158 {
170 VM_BUG_ON_VMA(vma->anon_vma != next->a 159 VM_BUG_ON_VMA(vma->anon_vma != next->anon_vma, vma);
171 unlink_anon_vmas(next); 160 unlink_anon_vmas(next);
172 } 161 }
173 162
174 struct anon_vma *folio_get_anon_vma(struct fol !! 163 struct anon_vma *page_get_anon_vma(struct page *page);
175 164
176 /* RMAP flags, currently only relevant for som !! 165 /* bitflags for do_page_add_anon_rmap() */
177 typedef int __bitwise rmap_t; !! 166 #define RMAP_EXCLUSIVE 0x01
178 !! 167 #define RMAP_COMPOUND 0x02
179 /* <<
180 * No special request: A mapped anonymous (sub <<
181 * processes. <<
182 */ <<
183 #define RMAP_NONE ((__force rmap <<
184 <<
185 /* The anonymous (sub)page is exclusive to a s <<
186 #define RMAP_EXCLUSIVE ((__force rmap <<
187 <<
188 /* <<
189 * Internally, we're using an enum to specify <<
190 * compiler emit specialized code for each gra <<
191 */ <<
192 enum rmap_level { <<
193 RMAP_LEVEL_PTE = 0, <<
194 RMAP_LEVEL_PMD, <<
195 }; <<
196 <<
197 static inline void __folio_rmap_sanity_checks( <<
198 struct page *page, int nr_page <<
199 { <<
200 /* hugetlb folios are handled separate <<
201 VM_WARN_ON_FOLIO(folio_test_hugetlb(fo <<
202 <<
203 /* When (un)mapping zeropages, we shou <<
204 VM_WARN_ON_FOLIO(is_zero_folio(folio), <<
205 <<
206 /* <<
207 * TODO: we get driver-allocated folio <<
208 * the rmap using vm_insert_page(); th <<
209 * folio_test_large_rmappable() holds <<
210 * handle any desired mapcount+stats a <<
211 * VM_MIXEDMAP VMAs separately, and th <<
212 * we really only get rmappable folios <<
213 */ <<
214 <<
215 VM_WARN_ON_ONCE(nr_pages <= 0); <<
216 VM_WARN_ON_FOLIO(page_folio(page) != f <<
217 VM_WARN_ON_FOLIO(page_folio(page + nr_ <<
218 <<
219 switch (level) { <<
220 case RMAP_LEVEL_PTE: <<
221 break; <<
222 case RMAP_LEVEL_PMD: <<
223 /* <<
224 * We don't support folios lar <<
225 * when RMAP_LEVEL_PMD is set, <<
226 * a single "entire" mapping o <<
227 */ <<
228 VM_WARN_ON_FOLIO(folio_nr_page <<
229 VM_WARN_ON_FOLIO(nr_pages != H <<
230 break; <<
231 default: <<
232 VM_WARN_ON_ONCE(true); <<
233 } <<
234 } <<
235 168
236 /* 169 /*
237 * rmap interfaces called when adding or remov 170 * rmap interfaces called when adding or removing pte of page
238 */ 171 */
239 void folio_move_anon_rmap(struct folio *, stru !! 172 void page_move_anon_rmap(struct page *, struct vm_area_struct *);
240 void folio_add_anon_rmap_ptes(struct folio *, !! 173 void page_add_anon_rmap(struct page *, struct vm_area_struct *,
241 struct vm_area_struct *, unsig !! 174 unsigned long, bool);
242 #define folio_add_anon_rmap_pte(folio, page, v !! 175 void do_page_add_anon_rmap(struct page *, struct vm_area_struct *,
243 folio_add_anon_rmap_ptes(folio, page, !! 176 unsigned long, int);
244 void folio_add_anon_rmap_pmd(struct folio *, s !! 177 void page_add_new_anon_rmap(struct page *, struct vm_area_struct *,
245 struct vm_area_struct *, unsig !! 178 unsigned long, bool);
246 void folio_add_new_anon_rmap(struct folio *, s !! 179 void page_add_file_rmap(struct page *, bool);
247 unsigned long address, rmap_t !! 180 void page_remove_rmap(struct page *, bool);
248 void folio_add_file_rmap_ptes(struct folio *, !! 181
249 struct vm_area_struct *); !! 182 void hugepage_add_anon_rmap(struct page *, struct vm_area_struct *,
250 #define folio_add_file_rmap_pte(folio, page, v !! 183 unsigned long);
251 folio_add_file_rmap_ptes(folio, page, !! 184 void hugepage_add_new_anon_rmap(struct page *, struct vm_area_struct *,
252 void folio_add_file_rmap_pmd(struct folio *, s !! 185 unsigned long);
253 struct vm_area_struct *); <<
254 void folio_remove_rmap_ptes(struct folio *, st <<
255 struct vm_area_struct *); <<
256 #define folio_remove_rmap_pte(folio, page, vma <<
257 folio_remove_rmap_ptes(folio, page, 1, <<
258 void folio_remove_rmap_pmd(struct folio *, str <<
259 struct vm_area_struct *); <<
260 <<
261 void hugetlb_add_anon_rmap(struct folio *, str <<
262 unsigned long address, rmap_t <<
263 void hugetlb_add_new_anon_rmap(struct folio *, <<
264 unsigned long address); <<
265 <<
266 /* See folio_try_dup_anon_rmap_*() */ <<
267 static inline int hugetlb_try_dup_anon_rmap(st <<
268 struct vm_area_struct *vma) <<
269 { <<
270 VM_WARN_ON_FOLIO(!folio_test_hugetlb(f <<
271 VM_WARN_ON_FOLIO(!folio_test_anon(foli <<
272 <<
273 if (PageAnonExclusive(&folio->page)) { <<
274 if (unlikely(folio_needs_cow_f <<
275 return -EBUSY; <<
276 ClearPageAnonExclusive(&folio- <<
277 } <<
278 atomic_inc(&folio->_entire_mapcount); <<
279 atomic_inc(&folio->_large_mapcount); <<
280 return 0; <<
281 } <<
282 <<
283 /* See folio_try_share_anon_rmap_*() */ <<
284 static inline int hugetlb_try_share_anon_rmap( <<
285 { <<
286 VM_WARN_ON_FOLIO(!folio_test_hugetlb(f <<
287 VM_WARN_ON_FOLIO(!folio_test_anon(foli <<
288 VM_WARN_ON_FOLIO(!PageAnonExclusive(&f <<
289 <<
290 /* Paired with the memory barrier in t <<
291 if (IS_ENABLED(CONFIG_HAVE_GUP_FAST)) <<
292 smp_mb(); <<
293 <<
294 if (unlikely(folio_maybe_dma_pinned(fo <<
295 return -EBUSY; <<
296 ClearPageAnonExclusive(&folio->page); <<
297 <<
298 /* <<
299 * This is conceptually a smp_wmb() pa <<
300 * gup_must_unshare(). <<
301 */ <<
302 if (IS_ENABLED(CONFIG_HAVE_GUP_FAST)) <<
303 smp_mb__after_atomic(); <<
304 return 0; <<
305 } <<
306 <<
307 static inline void hugetlb_add_file_rmap(struc <<
308 { <<
309 VM_WARN_ON_FOLIO(!folio_test_hugetlb(f <<
310 VM_WARN_ON_FOLIO(folio_test_anon(folio <<
311 <<
312 atomic_inc(&folio->_entire_mapcount); <<
313 atomic_inc(&folio->_large_mapcount); <<
314 } <<
315 <<
316 static inline void hugetlb_remove_rmap(struct <<
317 { <<
318 VM_WARN_ON_FOLIO(!folio_test_hugetlb(f <<
319 <<
320 atomic_dec(&folio->_entire_mapcount); <<
321 atomic_dec(&folio->_large_mapcount); <<
322 } <<
323 <<
324 static __always_inline void __folio_dup_file_r <<
325 struct page *page, int nr_page <<
326 { <<
327 const int orig_nr_pages = nr_pages; <<
328 <<
329 __folio_rmap_sanity_checks(folio, page <<
330 <<
331 switch (level) { <<
332 case RMAP_LEVEL_PTE: <<
333 if (!folio_test_large(folio)) <<
334 atomic_inc(&folio->_ma <<
335 break; <<
336 } <<
337 <<
338 do { <<
339 atomic_inc(&page->_map <<
340 } while (page++, --nr_pages > <<
341 atomic_add(orig_nr_pages, &fol <<
342 break; <<
343 case RMAP_LEVEL_PMD: <<
344 atomic_inc(&folio->_entire_map <<
345 atomic_inc(&folio->_large_mapc <<
346 break; <<
347 } <<
348 } <<
349 <<
350 /** <<
351 * folio_dup_file_rmap_ptes - duplicate PTE ma <<
352 * @folio: The folio to duplicate the map <<
353 * @page: The first page to duplicate th <<
354 * @nr_pages: The number of pages of which t <<
355 * <<
356 * The page range of the folio is defined by [ <<
357 * <<
358 * The caller needs to hold the page table loc <<
359 */ <<
360 static inline void folio_dup_file_rmap_ptes(st <<
361 struct page *page, int nr_page <<
362 { <<
363 __folio_dup_file_rmap(folio, page, nr_ <<
364 } <<
365 186
366 static __always_inline void folio_dup_file_rma !! 187 static inline void page_dup_rmap(struct page *page, bool compound)
367 struct page *page) <<
368 { 188 {
369 __folio_dup_file_rmap(folio, page, 1, !! 189 atomic_inc(compound ? compound_mapcount_ptr(page) : &page->_mapcount);
370 } <<
371 <<
372 /** <<
373 * folio_dup_file_rmap_pmd - duplicate a PMD m <<
374 * @folio: The folio to duplicate the map <<
375 * @page: The first page to duplicate th <<
376 * <<
377 * The page range of the folio is defined by [ <<
378 * <<
379 * The caller needs to hold the page table loc <<
380 */ <<
381 static inline void folio_dup_file_rmap_pmd(str <<
382 struct page *page) <<
383 { <<
384 #ifdef CONFIG_TRANSPARENT_HUGEPAGE <<
385 __folio_dup_file_rmap(folio, page, HPA <<
386 #else <<
387 WARN_ON_ONCE(true); <<
388 #endif <<
389 } <<
390 <<
391 static __always_inline int __folio_try_dup_ano <<
392 struct page *page, int nr_page <<
393 enum rmap_level level) <<
394 { <<
395 const int orig_nr_pages = nr_pages; <<
396 bool maybe_pinned; <<
397 int i; <<
398 <<
399 VM_WARN_ON_FOLIO(!folio_test_anon(foli <<
400 __folio_rmap_sanity_checks(folio, page <<
401 <<
402 /* <<
403 * If this folio may have been pinned <<
404 * don't allow to duplicate the mappin <<
405 * copy the subpage immediately for th <<
406 * guarantee the pinned folio won't be <<
407 * future on write faults. <<
408 */ <<
409 maybe_pinned = likely(!folio_is_device <<
410 unlikely(folio_needs_co <<
411 <<
412 /* <<
413 * No need to check+clear for already <<
414 * folio. But if any page is PageAnonE <<
415 * copying if the folio maybe pinned. <<
416 */ <<
417 switch (level) { <<
418 case RMAP_LEVEL_PTE: <<
419 if (unlikely(maybe_pinned)) { <<
420 for (i = 0; i < nr_pag <<
421 if (PageAnonEx <<
422 return <<
423 } <<
424 <<
425 if (!folio_test_large(folio)) <<
426 if (PageAnonExclusive( <<
427 ClearPageAnonE <<
428 atomic_inc(&folio->_ma <<
429 break; <<
430 } <<
431 <<
432 do { <<
433 if (PageAnonExclusive( <<
434 ClearPageAnonE <<
435 atomic_inc(&page->_map <<
436 } while (page++, --nr_pages > <<
437 atomic_add(orig_nr_pages, &fol <<
438 break; <<
439 case RMAP_LEVEL_PMD: <<
440 if (PageAnonExclusive(page)) { <<
441 if (unlikely(maybe_pin <<
442 return -EBUSY; <<
443 ClearPageAnonExclusive <<
444 } <<
445 atomic_inc(&folio->_entire_map <<
446 atomic_inc(&folio->_large_mapc <<
447 break; <<
448 } <<
449 return 0; <<
450 } <<
451 <<
452 /** <<
453 * folio_try_dup_anon_rmap_ptes - try duplicat <<
454 * of a folio <<
455 * @folio: The folio to duplicate the map <<
456 * @page: The first page to duplicate th <<
457 * @nr_pages: The number of pages of which t <<
458 * @src_vma: The vm area from which the map <<
459 * <<
460 * The page range of the folio is defined by [ <<
461 * <<
462 * The caller needs to hold the page table loc <<
463 * vma->vma_mm->write_protect_seq. <<
464 * <<
465 * Duplicating the mappings can only fail if t <<
466 * private folios cannot get pinned and conseq <<
467 * for them. <<
468 * <<
469 * If duplicating the mappings succeeded, the <<
470 * the parent and the child. They must *not* b <<
471 * succeeded. <<
472 * <<
473 * Returns 0 if duplicating the mappings succe <<
474 */ <<
475 static inline int folio_try_dup_anon_rmap_ptes <<
476 struct page *page, int nr_page <<
477 { <<
478 return __folio_try_dup_anon_rmap(folio <<
479 RMAP_ <<
480 } <<
481 <<
482 static __always_inline int folio_try_dup_anon_ <<
483 struct page *page, struct vm_a <<
484 { <<
485 return __folio_try_dup_anon_rmap(folio <<
486 RMAP_ <<
487 } <<
488 <<
489 /** <<
490 * folio_try_dup_anon_rmap_pmd - try duplicati <<
491 * of a folio <<
492 * @folio: The folio to duplicate the map <<
493 * @page: The first page to duplicate th <<
494 * @src_vma: The vm area from which the map <<
495 * <<
496 * The page range of the folio is defined by [ <<
497 * <<
498 * The caller needs to hold the page table loc <<
499 * vma->vma_mm->write_protect_seq. <<
500 * <<
501 * Duplicating the mapping can only fail if th <<
502 * private folios cannot get pinned and conseq <<
503 * for them. <<
504 * <<
505 * If duplicating the mapping succeeds, the du <<
506 * the parent and the child. They must *not* b <<
507 * succeeded. <<
508 * <<
509 * Returns 0 if duplicating the mapping succee <<
510 */ <<
511 static inline int folio_try_dup_anon_rmap_pmd( <<
512 struct page *page, struct vm_a <<
513 { <<
514 #ifdef CONFIG_TRANSPARENT_HUGEPAGE <<
515 return __folio_try_dup_anon_rmap(folio <<
516 RMAP_ <<
517 #else <<
518 WARN_ON_ONCE(true); <<
519 return -EBUSY; <<
520 #endif <<
521 } <<
522 <<
523 static __always_inline int __folio_try_share_a <<
524 struct page *page, int nr_page <<
525 { <<
526 VM_WARN_ON_FOLIO(!folio_test_anon(foli <<
527 VM_WARN_ON_FOLIO(!PageAnonExclusive(pa <<
528 __folio_rmap_sanity_checks(folio, page <<
529 <<
530 /* device private folios cannot get pi <<
531 if (unlikely(folio_is_device_private(f <<
532 ClearPageAnonExclusive(page); <<
533 return 0; <<
534 } <<
535 <<
536 /* <<
537 * We have to make sure that when we c <<
538 * the page is not pinned and that con <<
539 * concurrently pinning the page. <<
540 * <<
541 * Conceptually, PageAnonExclusive cle <<
542 * (A1) Clear PTE <<
543 * (A2) Check if the page is pinned; b <<
544 * (A3) Clear PageAnonExclusive <<
545 * (A4) Restore PTE (optional, but cer <<
546 * <<
547 * When clearing PageAnonExclusive, we <<
548 * writable again, because anon pages <<
549 * be writable. So in any case, if the <<
550 * be writable anymore afterwards and <<
551 * if the PTE wasn't writable, there m <<
552 * <<
553 * Conceptually, GUP-fast pinning of a <<
554 * (B1) Read the PTE <<
555 * (B2) FOLL_WRITE: check if the PTE i <<
556 * (B3) Pin the mapped page <<
557 * (B4) Check if the PTE changed by re <<
558 * (B5) If the original PTE is not wri <<
559 * PageAnonExclusive is not set; <<
560 * <<
561 * If the PTE was writable, we only ha <<
562 * observes a PTE change and properly <<
563 * <<
564 * If the PTE was not writable, we hav <<
565 * detects a (temporary) PTE change or <<
566 * and properly backs off. <<
567 * <<
568 * Consequently, when clearing PageAno <<
569 * sure that (A1), (A2)/(A3) and (A4) <<
570 * order. In GUP-fast pinning code, we <<
571 * and (B5) happen in the right memory <<
572 * <<
573 * We assume that there might not be a <<
574 * clearing/invalidating the PTE (A1) <<
575 * so we use explicit ones here. <<
576 */ <<
577 <<
578 /* Paired with the memory barrier in t <<
579 if (IS_ENABLED(CONFIG_HAVE_GUP_FAST)) <<
580 smp_mb(); <<
581 <<
582 if (unlikely(folio_maybe_dma_pinned(fo <<
583 return -EBUSY; <<
584 ClearPageAnonExclusive(page); <<
585 <<
586 /* <<
587 * This is conceptually a smp_wmb() pa <<
588 * gup_must_unshare(). <<
589 */ <<
590 if (IS_ENABLED(CONFIG_HAVE_GUP_FAST)) <<
591 smp_mb__after_atomic(); <<
592 return 0; <<
593 } <<
594 <<
595 /** <<
596 * folio_try_share_anon_rmap_pte - try marking <<
597 * mapped by a <<
598 * for KSM or <<
599 * @folio: The folio to share a mapping o <<
600 * @page: The mapped exclusive page <<
601 * <<
602 * The caller needs to hold the page table loc <<
603 * entries cleared/invalidated. <<
604 * <<
605 * This is similar to folio_try_dup_anon_rmap_ <<
606 * fork() to duplicate mappings, but instead t <<
607 * unmapping parts of a folio (swap, migration <<
608 * <<
609 * Marking the mapped page shared can only fai <<
610 * device private folios cannot get pinned and <<
611 * fail. <<
612 * <<
613 * Returns 0 if marking the mapped page possib <<
614 * -EBUSY otherwise. <<
615 */ <<
616 static inline int folio_try_share_anon_rmap_pt <<
617 struct page *page) <<
618 { <<
619 return __folio_try_share_anon_rmap(fol <<
620 } <<
621 <<
622 /** <<
623 * folio_try_share_anon_rmap_pmd - try marking <<
624 * range mappe <<
625 * prepare for <<
626 * @folio: The folio to share the mapping <<
627 * @page: The first page to share the ma <<
628 * <<
629 * The page range of the folio is defined by [ <<
630 * <<
631 * The caller needs to hold the page table loc <<
632 * entries cleared/invalidated. <<
633 * <<
634 * This is similar to folio_try_dup_anon_rmap_ <<
635 * fork() to duplicate a mapping, but instead <<
636 * unmapping parts of a folio (swap, migration <<
637 * <<
638 * Marking the mapped pages shared can only fa <<
639 * device private folios cannot get pinned and <<
640 * fail. <<
641 * <<
642 * Returns 0 if marking the mapped pages possi <<
643 * -EBUSY otherwise. <<
644 */ <<
645 static inline int folio_try_share_anon_rmap_pm <<
646 struct page *page) <<
647 { <<
648 #ifdef CONFIG_TRANSPARENT_HUGEPAGE <<
649 return __folio_try_share_anon_rmap(fol <<
650 RMA <<
651 #else <<
652 WARN_ON_ONCE(true); <<
653 return -EBUSY; <<
654 #endif <<
655 } 190 }
656 191
657 /* 192 /*
658 * Called from mm/vmscan.c to handle paging ou 193 * Called from mm/vmscan.c to handle paging out
659 */ 194 */
660 int folio_referenced(struct folio *, int is_lo !! 195 int page_referenced(struct page *, int is_locked,
661 struct mem_cgroup *mem 196 struct mem_cgroup *memcg, unsigned long *vm_flags);
662 197
663 void try_to_migrate(struct folio *folio, enum !! 198 bool try_to_unmap(struct page *, enum ttu_flags flags);
664 void try_to_unmap(struct folio *, enum ttu_fla <<
665 <<
666 int make_device_exclusive_range(struct mm_stru <<
667 unsigned long <<
668 void *arg); <<
669 199
670 /* Avoid racy checks */ 200 /* Avoid racy checks */
671 #define PVMW_SYNC (1 << 0) 201 #define PVMW_SYNC (1 << 0)
672 /* Look for migration entries rather than pres !! 202 /* Look for migarion entries rather than present PTEs */
673 #define PVMW_MIGRATION (1 << 1) 203 #define PVMW_MIGRATION (1 << 1)
674 204
675 struct page_vma_mapped_walk { 205 struct page_vma_mapped_walk {
676 unsigned long pfn; !! 206 struct page *page;
677 unsigned long nr_pages; <<
678 pgoff_t pgoff; <<
679 struct vm_area_struct *vma; 207 struct vm_area_struct *vma;
680 unsigned long address; 208 unsigned long address;
681 pmd_t *pmd; 209 pmd_t *pmd;
682 pte_t *pte; 210 pte_t *pte;
683 spinlock_t *ptl; 211 spinlock_t *ptl;
684 unsigned int flags; 212 unsigned int flags;
685 }; 213 };
686 214
687 #define DEFINE_FOLIO_VMA_WALK(name, _folio, _v <<
688 struct page_vma_mapped_walk name = { <<
689 .pfn = folio_pfn(_folio), <<
690 .nr_pages = folio_nr_pages(_fo <<
691 .pgoff = folio_pgoff(_folio), <<
692 .vma = _vma, <<
693 .address = _address, <<
694 .flags = _flags, <<
695 } <<
696 <<
697 static inline void page_vma_mapped_walk_done(s 215 static inline void page_vma_mapped_walk_done(struct page_vma_mapped_walk *pvmw)
698 { 216 {
699 /* HugeTLB pte is set to the relevant !! 217 if (pvmw->pte)
700 if (pvmw->pte && !is_vm_hugetlb_page(p <<
701 pte_unmap(pvmw->pte); 218 pte_unmap(pvmw->pte);
702 if (pvmw->ptl) 219 if (pvmw->ptl)
703 spin_unlock(pvmw->ptl); 220 spin_unlock(pvmw->ptl);
704 } 221 }
705 222
706 /** <<
707 * page_vma_mapped_walk_restart - Restart the <<
708 * @pvmw: Pointer to struct page_vma_mapped_wa <<
709 * <<
710 * It restarts the page table walk when change <<
711 * table, such as splitting a PMD. Ensures tha <<
712 * the previous walk is released and resets th <<
713 * a new walk starting at the current address <<
714 */ <<
715 static inline void <<
716 page_vma_mapped_walk_restart(struct page_vma_m <<
717 { <<
718 WARN_ON_ONCE(!pvmw->pmd && !pvmw->pte) <<
719 <<
720 if (likely(pvmw->ptl)) <<
721 spin_unlock(pvmw->ptl); <<
722 else <<
723 WARN_ON_ONCE(1); <<
724 <<
725 pvmw->ptl = NULL; <<
726 pvmw->pmd = NULL; <<
727 pvmw->pte = NULL; <<
728 } <<
729 <<
730 bool page_vma_mapped_walk(struct page_vma_mapp 223 bool page_vma_mapped_walk(struct page_vma_mapped_walk *pvmw);
731 224
732 /* 225 /*
733 * Used by swapoff to help locate where page i 226 * Used by swapoff to help locate where page is expected in vma.
734 */ 227 */
735 unsigned long page_address_in_vma(struct page 228 unsigned long page_address_in_vma(struct page *, struct vm_area_struct *);
736 229
737 /* 230 /*
738 * Cleans the PTEs of shared mappings. 231 * Cleans the PTEs of shared mappings.
739 * (and since clean PTEs should also be readon 232 * (and since clean PTEs should also be readonly, write protects them too)
740 * 233 *
741 * returns the number of cleaned PTEs. 234 * returns the number of cleaned PTEs.
742 */ 235 */
743 int folio_mkclean(struct folio *); !! 236 int page_mkclean(struct page *);
744 237
745 int pfn_mkclean_range(unsigned long pfn, unsig !! 238 /*
746 struct vm_area_struct *v !! 239 * called in munlock()/munmap() path to check for other vmas holding
>> 240 * the page mlocked.
>> 241 */
>> 242 void try_to_munlock(struct page *);
747 243
748 enum rmp_flags { !! 244 void remove_migration_ptes(struct page *old, struct page *new, bool locked);
749 RMP_LOCKED = 1 << 0, <<
750 RMP_USE_SHARED_ZEROPAGE = 1 << 1, <<
751 }; <<
752 245
753 void remove_migration_ptes(struct folio *src, !! 246 /*
>> 247 * Called by memory-failure.c to kill processes.
>> 248 */
>> 249 struct anon_vma *page_lock_anon_vma_read(struct page *page);
>> 250 void page_unlock_anon_vma_read(struct anon_vma *anon_vma);
>> 251 int page_mapped_in_vma(struct page *page, struct vm_area_struct *vma);
754 252
755 /* 253 /*
756 * rmap_walk_control: To control rmap traversi 254 * rmap_walk_control: To control rmap traversing for specific needs
757 * 255 *
758 * arg: passed to rmap_one() and invalid_vma() 256 * arg: passed to rmap_one() and invalid_vma()
759 * try_lock: bail out if the rmap lock is cont <<
760 * contended: indicate the rmap traversal bail <<
761 * rmap_one: executed on each vma where page i 257 * rmap_one: executed on each vma where page is mapped
762 * done: for checking traversing termination c 258 * done: for checking traversing termination condition
763 * anon_lock: for getting anon_lock by optimiz 259 * anon_lock: for getting anon_lock by optimized way rather than default
764 * invalid_vma: for skipping uninterested vma 260 * invalid_vma: for skipping uninterested vma
765 */ 261 */
766 struct rmap_walk_control { 262 struct rmap_walk_control {
767 void *arg; 263 void *arg;
768 bool try_lock; <<
769 bool contended; <<
770 /* 264 /*
771 * Return false if page table scanning 265 * Return false if page table scanning in rmap_walk should be stopped.
772 * Otherwise, return true. 266 * Otherwise, return true.
773 */ 267 */
774 bool (*rmap_one)(struct folio *folio, !! 268 bool (*rmap_one)(struct page *page, struct vm_area_struct *vma,
775 unsign 269 unsigned long addr, void *arg);
776 int (*done)(struct folio *folio); !! 270 int (*done)(struct page *page);
777 struct anon_vma *(*anon_lock)(struct f !! 271 struct anon_vma *(*anon_lock)(struct page *page);
778 struct r <<
779 bool (*invalid_vma)(struct vm_area_str 272 bool (*invalid_vma)(struct vm_area_struct *vma, void *arg);
780 }; 273 };
781 274
782 void rmap_walk(struct folio *folio, struct rma !! 275 void rmap_walk(struct page *page, struct rmap_walk_control *rwc);
783 void rmap_walk_locked(struct folio *folio, str !! 276 void rmap_walk_locked(struct page *page, struct rmap_walk_control *rwc);
784 struct anon_vma *folio_lock_anon_vma_read(stru <<
785 stru <<
786 277
787 #else /* !CONFIG_MMU */ 278 #else /* !CONFIG_MMU */
788 279
789 #define anon_vma_init() do {} while (0 280 #define anon_vma_init() do {} while (0)
790 #define anon_vma_prepare(vma) (0) 281 #define anon_vma_prepare(vma) (0)
>> 282 #define anon_vma_link(vma) do {} while (0)
791 283
792 static inline int folio_referenced(struct foli !! 284 static inline int page_referenced(struct page *page, int is_locked,
793 struct mem_c 285 struct mem_cgroup *memcg,
794 unsigned lon 286 unsigned long *vm_flags)
795 { 287 {
796 *vm_flags = 0; 288 *vm_flags = 0;
797 return 0; 289 return 0;
798 } 290 }
799 291
800 static inline void try_to_unmap(struct folio * !! 292 #define try_to_unmap(page, refs) false
801 { <<
802 } <<
803 293
804 static inline int folio_mkclean(struct folio * !! 294 static inline int page_mkclean(struct page *page)
805 { 295 {
806 return 0; 296 return 0;
807 } 297 }
>> 298
>> 299
808 #endif /* CONFIG_MMU */ 300 #endif /* CONFIG_MMU */
809 301
810 #endif /* _LINUX_RMAP_H */ 302 #endif /* _LINUX_RMAP_H */
811 303
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.