Commit | Line | Data |
---|---|---|
b2441318 | 1 | /* SPDX-License-Identifier: GPL-2.0 */ |
b2e18538 RR |
2 | #ifndef LINUX_MM_INLINE_H |
3 | #define LINUX_MM_INLINE_H | |
4 | ||
36090def | 5 | #include <linux/atomic.h> |
2c888cfb | 6 | #include <linux/huge_mm.h> |
6e543d57 | 7 | #include <linux/swap.h> |
17fca131 | 8 | #include <linux/string.h> |
999dad82 PX |
9 | #include <linux/userfaultfd_k.h> |
10 | #include <linux/swapops.h> | |
2c888cfb | 11 | |
b2e18538 | 12 | /** |
889a3747 MWO |
13 | * folio_is_file_lru - Should the folio be on a file LRU or anon LRU? |
14 | * @folio: The folio to test. | |
b2e18538 RR |
15 | * |
16 | * We would like to get this info without a page flag, but the state | |
889a3747 | 17 | * needs to survive until the folio is last deleted from the LRU, which |
b2e18538 | 18 | * could be as far down as __page_cache_release. |
889a3747 MWO |
19 | * |
20 | * Return: An integer (not a boolean!) used to sort a folio onto the | |
21 | * right LRU list and to account folios correctly. | |
22 | * 1 if @folio is a regular filesystem backed page cache folio | |
23 | * or a lazily freed anonymous folio (e.g. via MADV_FREE). | |
24 | * 0 if @folio is a normal anonymous folio, a tmpfs folio or otherwise | |
25 | * ram or swap backed folio. | |
b2e18538 | 26 | */ |
889a3747 MWO |
27 | static inline int folio_is_file_lru(struct folio *folio) |
28 | { | |
29 | return !folio_test_swapbacked(folio); | |
30 | } | |
31 | ||
9de4f22a | 32 | static inline int page_is_file_lru(struct page *page) |
b2e18538 | 33 | { |
889a3747 | 34 | return folio_is_file_lru(page_folio(page)); |
b2e18538 RR |
35 | } |
36 | ||
aa1b6790 | 37 | static __always_inline void __update_lru_size(struct lruvec *lruvec, |
599d0c95 | 38 | enum lru_list lru, enum zone_type zid, |
889a3747 | 39 | long nr_pages) |
9d5e6a9f | 40 | { |
599d0c95 MG |
41 | struct pglist_data *pgdat = lruvec_pgdat(lruvec); |
42 | ||
ec1c86b2 YZ |
43 | lockdep_assert_held(&lruvec->lru_lock); |
44 | WARN_ON_ONCE(nr_pages != (int)nr_pages); | |
45 | ||
e0ee0e71 | 46 | __mod_lruvec_state(lruvec, NR_LRU_BASE + lru, nr_pages); |
71c799f4 MK |
47 | __mod_zone_page_state(&pgdat->node_zones[zid], |
48 | NR_ZONE_LRU_BASE + lru, nr_pages); | |
aa1b6790 YZ |
49 | } |
50 | ||
51 | static __always_inline void update_lru_size(struct lruvec *lruvec, | |
52 | enum lru_list lru, enum zone_type zid, | |
53 | long nr_pages) | |
54 | { | |
55 | __update_lru_size(lruvec, lru, zid, nr_pages); | |
7ee36a14 | 56 | #ifdef CONFIG_MEMCG |
b4536f0c | 57 | mem_cgroup_update_lru_size(lruvec, lru, zid, nr_pages); |
9d5e6a9f HD |
58 | #endif |
59 | } | |
60 | ||
1c1c53d4 | 61 | /** |
889a3747 MWO |
62 | * __folio_clear_lru_flags - Clear page lru flags before releasing a page. |
63 | * @folio: The folio that was on lru and now has a zero reference. | |
1c1c53d4 | 64 | */ |
889a3747 | 65 | static __always_inline void __folio_clear_lru_flags(struct folio *folio) |
1da177e4 | 66 | { |
889a3747 | 67 | VM_BUG_ON_FOLIO(!folio_test_lru(folio), folio); |
bc711271 | 68 | |
889a3747 | 69 | __folio_clear_lru(folio); |
87560179 YZ |
70 | |
71 | /* this shouldn't happen, so leave the flags to bad_page() */ | |
889a3747 | 72 | if (folio_test_active(folio) && folio_test_unevictable(folio)) |
87560179 | 73 | return; |
b69408e8 | 74 | |
889a3747 MWO |
75 | __folio_clear_active(folio); |
76 | __folio_clear_unevictable(folio); | |
77 | } | |
78 | ||
b69408e8 | 79 | /** |
889a3747 MWO |
80 | * folio_lru_list - Which LRU list should a folio be on? |
81 | * @folio: The folio to test. | |
b69408e8 | 82 | * |
889a3747 | 83 | * Return: The LRU list a folio should be on, as an index |
b69408e8 CL |
84 | * into the array of LRU lists. |
85 | */ | |
889a3747 | 86 | static __always_inline enum lru_list folio_lru_list(struct folio *folio) |
b69408e8 | 87 | { |
401a8e1c | 88 | enum lru_list lru; |
b69408e8 | 89 | |
889a3747 | 90 | VM_BUG_ON_FOLIO(folio_test_active(folio) && folio_test_unevictable(folio), folio); |
bc711271 | 91 | |
889a3747 | 92 | if (folio_test_unevictable(folio)) |
c1770e34 YZ |
93 | return LRU_UNEVICTABLE; |
94 | ||
889a3747 MWO |
95 | lru = folio_is_file_lru(folio) ? LRU_INACTIVE_FILE : LRU_INACTIVE_ANON; |
96 | if (folio_test_active(folio)) | |
c1770e34 YZ |
97 | lru += LRU_ACTIVE; |
98 | ||
b69408e8 CL |
99 | return lru; |
100 | } | |
f90d8191 | 101 | |
ec1c86b2 YZ |
102 | #ifdef CONFIG_LRU_GEN |
103 | ||
354ed597 | 104 | #ifdef CONFIG_LRU_GEN_ENABLED |
ec1c86b2 YZ |
105 | static inline bool lru_gen_enabled(void) |
106 | { | |
354ed597 YZ |
107 | DECLARE_STATIC_KEY_TRUE(lru_gen_caps[NR_LRU_GEN_CAPS]); |
108 | ||
109 | return static_branch_likely(&lru_gen_caps[LRU_GEN_CORE]); | |
ec1c86b2 | 110 | } |
354ed597 YZ |
111 | #else |
112 | static inline bool lru_gen_enabled(void) | |
113 | { | |
114 | DECLARE_STATIC_KEY_FALSE(lru_gen_caps[NR_LRU_GEN_CAPS]); | |
115 | ||
116 | return static_branch_unlikely(&lru_gen_caps[LRU_GEN_CORE]); | |
117 | } | |
118 | #endif | |
ec1c86b2 YZ |
119 | |
120 | static inline bool lru_gen_in_fault(void) | |
121 | { | |
122 | return current->in_lru_fault; | |
123 | } | |
124 | ||
125 | static inline int lru_gen_from_seq(unsigned long seq) | |
126 | { | |
127 | return seq % MAX_NR_GENS; | |
128 | } | |
129 | ||
ac35a490 YZ |
130 | static inline int lru_hist_from_seq(unsigned long seq) |
131 | { | |
132 | return seq % NR_HIST_GENS; | |
133 | } | |
134 | ||
135 | static inline int lru_tier_from_refs(int refs) | |
136 | { | |
137 | VM_WARN_ON_ONCE(refs > BIT(LRU_REFS_WIDTH)); | |
138 | ||
139 | /* see the comment in folio_lru_refs() */ | |
140 | return order_base_2(refs + 1); | |
141 | } | |
142 | ||
143 | static inline int folio_lru_refs(struct folio *folio) | |
144 | { | |
145 | unsigned long flags = READ_ONCE(folio->flags); | |
146 | bool workingset = flags & BIT(PG_workingset); | |
147 | ||
148 | /* | |
149 | * Return the number of accesses beyond PG_referenced, i.e., N-1 if the | |
150 | * total number of accesses is N>1, since N=0,1 both map to the first | |
151 | * tier. lru_tier_from_refs() will account for this off-by-one. Also see | |
152 | * the comment on MAX_NR_TIERS. | |
153 | */ | |
154 | return ((flags & LRU_REFS_MASK) >> LRU_REFS_PGOFF) + workingset; | |
155 | } | |
156 | ||
ec1c86b2 YZ |
157 | static inline int folio_lru_gen(struct folio *folio) |
158 | { | |
159 | unsigned long flags = READ_ONCE(folio->flags); | |
160 | ||
161 | return ((flags & LRU_GEN_MASK) >> LRU_GEN_PGOFF) - 1; | |
162 | } | |
163 | ||
164 | static inline bool lru_gen_is_active(struct lruvec *lruvec, int gen) | |
165 | { | |
166 | unsigned long max_seq = lruvec->lrugen.max_seq; | |
167 | ||
168 | VM_WARN_ON_ONCE(gen >= MAX_NR_GENS); | |
169 | ||
170 | /* see the comment on MIN_NR_GENS */ | |
171 | return gen == lru_gen_from_seq(max_seq) || gen == lru_gen_from_seq(max_seq - 1); | |
172 | } | |
173 | ||
174 | static inline void lru_gen_update_size(struct lruvec *lruvec, struct folio *folio, | |
175 | int old_gen, int new_gen) | |
176 | { | |
177 | int type = folio_is_file_lru(folio); | |
178 | int zone = folio_zonenum(folio); | |
179 | int delta = folio_nr_pages(folio); | |
180 | enum lru_list lru = type * LRU_INACTIVE_FILE; | |
391655fe | 181 | struct lru_gen_folio *lrugen = &lruvec->lrugen; |
ec1c86b2 YZ |
182 | |
183 | VM_WARN_ON_ONCE(old_gen != -1 && old_gen >= MAX_NR_GENS); | |
184 | VM_WARN_ON_ONCE(new_gen != -1 && new_gen >= MAX_NR_GENS); | |
185 | VM_WARN_ON_ONCE(old_gen == -1 && new_gen == -1); | |
186 | ||
187 | if (old_gen >= 0) | |
188 | WRITE_ONCE(lrugen->nr_pages[old_gen][type][zone], | |
189 | lrugen->nr_pages[old_gen][type][zone] - delta); | |
190 | if (new_gen >= 0) | |
191 | WRITE_ONCE(lrugen->nr_pages[new_gen][type][zone], | |
192 | lrugen->nr_pages[new_gen][type][zone] + delta); | |
193 | ||
194 | /* addition */ | |
195 | if (old_gen < 0) { | |
196 | if (lru_gen_is_active(lruvec, new_gen)) | |
197 | lru += LRU_ACTIVE; | |
198 | __update_lru_size(lruvec, lru, zone, delta); | |
199 | return; | |
200 | } | |
201 | ||
202 | /* deletion */ | |
203 | if (new_gen < 0) { | |
204 | if (lru_gen_is_active(lruvec, old_gen)) | |
205 | lru += LRU_ACTIVE; | |
206 | __update_lru_size(lruvec, lru, zone, -delta); | |
207 | return; | |
208 | } | |
ac35a490 YZ |
209 | |
210 | /* promotion */ | |
211 | if (!lru_gen_is_active(lruvec, old_gen) && lru_gen_is_active(lruvec, new_gen)) { | |
212 | __update_lru_size(lruvec, lru, zone, -delta); | |
213 | __update_lru_size(lruvec, lru + LRU_ACTIVE, zone, delta); | |
214 | } | |
215 | ||
216 | /* demotion requires isolation, e.g., lru_deactivate_fn() */ | |
217 | VM_WARN_ON_ONCE(lru_gen_is_active(lruvec, old_gen) && !lru_gen_is_active(lruvec, new_gen)); | |
ec1c86b2 YZ |
218 | } |
219 | ||
220 | static inline bool lru_gen_add_folio(struct lruvec *lruvec, struct folio *folio, bool reclaiming) | |
221 | { | |
222 | unsigned long seq; | |
223 | unsigned long flags; | |
224 | int gen = folio_lru_gen(folio); | |
225 | int type = folio_is_file_lru(folio); | |
226 | int zone = folio_zonenum(folio); | |
391655fe | 227 | struct lru_gen_folio *lrugen = &lruvec->lrugen; |
ec1c86b2 YZ |
228 | |
229 | VM_WARN_ON_ONCE_FOLIO(gen != -1, folio); | |
230 | ||
354ed597 | 231 | if (folio_test_unevictable(folio) || !lrugen->enabled) |
ec1c86b2 YZ |
232 | return false; |
233 | /* | |
234 | * There are three common cases for this page: | |
235 | * 1. If it's hot, e.g., freshly faulted in or previously hot and | |
236 | * migrated, add it to the youngest generation. | |
237 | * 2. If it's cold but can't be evicted immediately, i.e., an anon page | |
238 | * not in swapcache or a dirty page pending writeback, add it to the | |
239 | * second oldest generation. | |
240 | * 3. Everything else (clean, cold) is added to the oldest generation. | |
241 | */ | |
242 | if (folio_test_active(folio)) | |
243 | seq = lrugen->max_seq; | |
244 | else if ((type == LRU_GEN_ANON && !folio_test_swapcache(folio)) || | |
245 | (folio_test_reclaim(folio) && | |
246 | (folio_test_dirty(folio) || folio_test_writeback(folio)))) | |
247 | seq = lrugen->min_seq[type] + 1; | |
248 | else | |
249 | seq = lrugen->min_seq[type]; | |
250 | ||
251 | gen = lru_gen_from_seq(seq); | |
252 | flags = (gen + 1UL) << LRU_GEN_PGOFF; | |
253 | /* see the comment on MIN_NR_GENS about PG_active */ | |
254 | set_mask_bits(&folio->flags, LRU_GEN_MASK | BIT(PG_active), flags); | |
255 | ||
256 | lru_gen_update_size(lruvec, folio, -1, gen); | |
257 | /* for folio_rotate_reclaimable() */ | |
258 | if (reclaiming) | |
6df1b221 | 259 | list_add_tail(&folio->lru, &lrugen->folios[gen][type][zone]); |
ec1c86b2 | 260 | else |
6df1b221 | 261 | list_add(&folio->lru, &lrugen->folios[gen][type][zone]); |
ec1c86b2 YZ |
262 | |
263 | return true; | |
264 | } | |
265 | ||
266 | static inline bool lru_gen_del_folio(struct lruvec *lruvec, struct folio *folio, bool reclaiming) | |
267 | { | |
268 | unsigned long flags; | |
269 | int gen = folio_lru_gen(folio); | |
270 | ||
271 | if (gen < 0) | |
272 | return false; | |
273 | ||
274 | VM_WARN_ON_ONCE_FOLIO(folio_test_active(folio), folio); | |
275 | VM_WARN_ON_ONCE_FOLIO(folio_test_unevictable(folio), folio); | |
276 | ||
277 | /* for folio_migrate_flags() */ | |
278 | flags = !reclaiming && lru_gen_is_active(lruvec, gen) ? BIT(PG_active) : 0; | |
279 | flags = set_mask_bits(&folio->flags, LRU_GEN_MASK, flags); | |
280 | gen = ((flags & LRU_GEN_MASK) >> LRU_GEN_PGOFF) - 1; | |
281 | ||
282 | lru_gen_update_size(lruvec, folio, gen, -1); | |
283 | list_del(&folio->lru); | |
284 | ||
285 | return true; | |
286 | } | |
287 | ||
288 | #else /* !CONFIG_LRU_GEN */ | |
289 | ||
290 | static inline bool lru_gen_enabled(void) | |
291 | { | |
292 | return false; | |
293 | } | |
294 | ||
295 | static inline bool lru_gen_in_fault(void) | |
296 | { | |
297 | return false; | |
298 | } | |
299 | ||
300 | static inline bool lru_gen_add_folio(struct lruvec *lruvec, struct folio *folio, bool reclaiming) | |
301 | { | |
302 | return false; | |
303 | } | |
304 | ||
305 | static inline bool lru_gen_del_folio(struct lruvec *lruvec, struct folio *folio, bool reclaiming) | |
306 | { | |
307 | return false; | |
308 | } | |
309 | ||
310 | #endif /* CONFIG_LRU_GEN */ | |
311 | ||
889a3747 MWO |
312 | static __always_inline |
313 | void lruvec_add_folio(struct lruvec *lruvec, struct folio *folio) | |
314 | { | |
315 | enum lru_list lru = folio_lru_list(folio); | |
316 | ||
ec1c86b2 YZ |
317 | if (lru_gen_add_folio(lruvec, folio, false)) |
318 | return; | |
319 | ||
889a3747 MWO |
320 | update_lru_size(lruvec, lru, folio_zonenum(folio), |
321 | folio_nr_pages(folio)); | |
07ca7606 HD |
322 | if (lru != LRU_UNEVICTABLE) |
323 | list_add(&folio->lru, &lruvec->lists[lru]); | |
889a3747 MWO |
324 | } |
325 | ||
889a3747 MWO |
326 | static __always_inline |
327 | void lruvec_add_folio_tail(struct lruvec *lruvec, struct folio *folio) | |
328 | { | |
329 | enum lru_list lru = folio_lru_list(folio); | |
330 | ||
ec1c86b2 YZ |
331 | if (lru_gen_add_folio(lruvec, folio, true)) |
332 | return; | |
333 | ||
889a3747 MWO |
334 | update_lru_size(lruvec, lru, folio_zonenum(folio), |
335 | folio_nr_pages(folio)); | |
07ca7606 | 336 | /* This is not expected to be used on LRU_UNEVICTABLE */ |
889a3747 | 337 | list_add_tail(&folio->lru, &lruvec->lists[lru]); |
f90d8191 YZ |
338 | } |
339 | ||
889a3747 MWO |
340 | static __always_inline |
341 | void lruvec_del_folio(struct lruvec *lruvec, struct folio *folio) | |
342 | { | |
07ca7606 HD |
343 | enum lru_list lru = folio_lru_list(folio); |
344 | ||
ec1c86b2 YZ |
345 | if (lru_gen_del_folio(lruvec, folio, false)) |
346 | return; | |
347 | ||
07ca7606 HD |
348 | if (lru != LRU_UNEVICTABLE) |
349 | list_del(&folio->lru); | |
350 | update_lru_size(lruvec, lru, folio_zonenum(folio), | |
889a3747 | 351 | -folio_nr_pages(folio)); |
f90d8191 YZ |
352 | } |
353 | ||
17fca131 AB |
354 | #ifdef CONFIG_ANON_VMA_NAME |
355 | /* | |
5c26f6ac SB |
356 | * mmap_lock should be read-locked when calling anon_vma_name(). Caller should |
357 | * either keep holding the lock while using the returned pointer or it should | |
358 | * raise anon_vma_name refcount before releasing the lock. | |
17fca131 | 359 | */ |
5c26f6ac SB |
360 | extern struct anon_vma_name *anon_vma_name(struct vm_area_struct *vma); |
361 | extern struct anon_vma_name *anon_vma_name_alloc(const char *name); | |
362 | extern void anon_vma_name_free(struct kref *kref); | |
17fca131 | 363 | |
5c26f6ac SB |
364 | /* mmap_lock should be read-locked */ |
365 | static inline void anon_vma_name_get(struct anon_vma_name *anon_name) | |
366 | { | |
367 | if (anon_name) | |
368 | kref_get(&anon_name->kref); | |
369 | } | |
17fca131 | 370 | |
5c26f6ac SB |
371 | static inline void anon_vma_name_put(struct anon_vma_name *anon_name) |
372 | { | |
373 | if (anon_name) | |
374 | kref_put(&anon_name->kref, anon_vma_name_free); | |
375 | } | |
17fca131 | 376 | |
96403e11 SB |
377 | static inline |
378 | struct anon_vma_name *anon_vma_name_reuse(struct anon_vma_name *anon_name) | |
379 | { | |
380 | /* Prevent anon_name refcount saturation early on */ | |
381 | if (kref_read(&anon_name->kref) < REFCOUNT_MAX) { | |
382 | anon_vma_name_get(anon_name); | |
383 | return anon_name; | |
384 | ||
385 | } | |
386 | return anon_vma_name_alloc(anon_name->name); | |
387 | } | |
388 | ||
5c26f6ac SB |
389 | static inline void dup_anon_vma_name(struct vm_area_struct *orig_vma, |
390 | struct vm_area_struct *new_vma) | |
17fca131 | 391 | { |
5c26f6ac | 392 | struct anon_vma_name *anon_name = anon_vma_name(orig_vma); |
17fca131 | 393 | |
96403e11 SB |
394 | if (anon_name) |
395 | new_vma->anon_name = anon_vma_name_reuse(anon_name); | |
5c26f6ac SB |
396 | } |
397 | ||
398 | static inline void free_anon_vma_name(struct vm_area_struct *vma) | |
399 | { | |
400 | /* | |
401 | * Not using anon_vma_name because it generates a warning if mmap_lock | |
402 | * is not held, which might be the case here. | |
403 | */ | |
a1193de5 | 404 | anon_vma_name_put(vma->anon_name); |
5c26f6ac SB |
405 | } |
406 | ||
407 | static inline bool anon_vma_name_eq(struct anon_vma_name *anon_name1, | |
408 | struct anon_vma_name *anon_name2) | |
409 | { | |
410 | if (anon_name1 == anon_name2) | |
17fca131 AB |
411 | return true; |
412 | ||
5c26f6ac SB |
413 | return anon_name1 && anon_name2 && |
414 | !strcmp(anon_name1->name, anon_name2->name); | |
17fca131 | 415 | } |
5c26f6ac | 416 | |
17fca131 | 417 | #else /* CONFIG_ANON_VMA_NAME */ |
5c26f6ac SB |
418 | static inline struct anon_vma_name *anon_vma_name(struct vm_area_struct *vma) |
419 | { | |
420 | return NULL; | |
421 | } | |
422 | ||
423 | static inline struct anon_vma_name *anon_vma_name_alloc(const char *name) | |
17fca131 AB |
424 | { |
425 | return NULL; | |
426 | } | |
5c26f6ac SB |
427 | |
428 | static inline void anon_vma_name_get(struct anon_vma_name *anon_name) {} | |
429 | static inline void anon_vma_name_put(struct anon_vma_name *anon_name) {} | |
430 | static inline void dup_anon_vma_name(struct vm_area_struct *orig_vma, | |
431 | struct vm_area_struct *new_vma) {} | |
432 | static inline void free_anon_vma_name(struct vm_area_struct *vma) {} | |
433 | ||
434 | static inline bool anon_vma_name_eq(struct anon_vma_name *anon_name1, | |
435 | struct anon_vma_name *anon_name2) | |
17fca131 AB |
436 | { |
437 | return true; | |
438 | } | |
5c26f6ac | 439 | |
17fca131 AB |
440 | #endif /* CONFIG_ANON_VMA_NAME */ |
441 | ||
36090def AB |
442 | static inline void init_tlb_flush_pending(struct mm_struct *mm) |
443 | { | |
444 | atomic_set(&mm->tlb_flush_pending, 0); | |
445 | } | |
446 | ||
447 | static inline void inc_tlb_flush_pending(struct mm_struct *mm) | |
448 | { | |
449 | atomic_inc(&mm->tlb_flush_pending); | |
450 | /* | |
451 | * The only time this value is relevant is when there are indeed pages | |
452 | * to flush. And we'll only flush pages after changing them, which | |
453 | * requires the PTL. | |
454 | * | |
455 | * So the ordering here is: | |
456 | * | |
457 | * atomic_inc(&mm->tlb_flush_pending); | |
458 | * spin_lock(&ptl); | |
459 | * ... | |
460 | * set_pte_at(); | |
461 | * spin_unlock(&ptl); | |
462 | * | |
463 | * spin_lock(&ptl) | |
464 | * mm_tlb_flush_pending(); | |
465 | * .... | |
466 | * spin_unlock(&ptl); | |
467 | * | |
468 | * flush_tlb_range(); | |
469 | * atomic_dec(&mm->tlb_flush_pending); | |
470 | * | |
471 | * Where the increment if constrained by the PTL unlock, it thus | |
472 | * ensures that the increment is visible if the PTE modification is | |
473 | * visible. After all, if there is no PTE modification, nobody cares | |
474 | * about TLB flushes either. | |
475 | * | |
476 | * This very much relies on users (mm_tlb_flush_pending() and | |
477 | * mm_tlb_flush_nested()) only caring about _specific_ PTEs (and | |
478 | * therefore specific PTLs), because with SPLIT_PTE_PTLOCKS and RCpc | |
479 | * locks (PPC) the unlock of one doesn't order against the lock of | |
480 | * another PTL. | |
481 | * | |
482 | * The decrement is ordered by the flush_tlb_range(), such that | |
483 | * mm_tlb_flush_pending() will not return false unless all flushes have | |
484 | * completed. | |
485 | */ | |
486 | } | |
487 | ||
488 | static inline void dec_tlb_flush_pending(struct mm_struct *mm) | |
489 | { | |
490 | /* | |
491 | * See inc_tlb_flush_pending(). | |
492 | * | |
493 | * This cannot be smp_mb__before_atomic() because smp_mb() simply does | |
494 | * not order against TLB invalidate completion, which is what we need. | |
495 | * | |
496 | * Therefore we must rely on tlb_flush_*() to guarantee order. | |
497 | */ | |
498 | atomic_dec(&mm->tlb_flush_pending); | |
499 | } | |
500 | ||
501 | static inline bool mm_tlb_flush_pending(struct mm_struct *mm) | |
502 | { | |
503 | /* | |
504 | * Must be called after having acquired the PTL; orders against that | |
505 | * PTLs release and therefore ensures that if we observe the modified | |
506 | * PTE we must also observe the increment from inc_tlb_flush_pending(). | |
507 | * | |
508 | * That is, it only guarantees to return true if there is a flush | |
509 | * pending for _this_ PTL. | |
510 | */ | |
511 | return atomic_read(&mm->tlb_flush_pending); | |
512 | } | |
513 | ||
514 | static inline bool mm_tlb_flush_nested(struct mm_struct *mm) | |
515 | { | |
516 | /* | |
517 | * Similar to mm_tlb_flush_pending(), we must have acquired the PTL | |
518 | * for which there is a TLB flush pending in order to guarantee | |
519 | * we've seen both that PTE modification and the increment. | |
520 | * | |
521 | * (no requirement on actually still holding the PTL, that is irrelevant) | |
522 | */ | |
523 | return atomic_read(&mm->tlb_flush_pending) > 1; | |
524 | } | |
525 | ||
f92cedfa | 526 | #ifdef CONFIG_MMU |
af19487f AR |
527 | /* |
528 | * Computes the pte marker to copy from the given source entry into dst_vma. | |
529 | * If no marker should be copied, returns 0. | |
530 | * The caller should insert a new pte created with make_pte_marker(). | |
531 | */ | |
532 | static inline pte_marker copy_pte_marker( | |
533 | swp_entry_t entry, struct vm_area_struct *dst_vma) | |
534 | { | |
535 | pte_marker srcm = pte_marker_get(entry); | |
536 | /* Always copy error entries. */ | |
537 | pte_marker dstm = srcm & PTE_MARKER_POISONED; | |
538 | ||
539 | /* Only copy PTE markers if UFFD register matches. */ | |
540 | if ((srcm & PTE_MARKER_UFFD_WP) && userfaultfd_wp(dst_vma)) | |
541 | dstm |= PTE_MARKER_UFFD_WP; | |
542 | ||
543 | return dstm; | |
544 | } | |
f92cedfa | 545 | #endif |
af19487f | 546 | |
999dad82 PX |
547 | /* |
548 | * If this pte is wr-protected by uffd-wp in any form, arm the special pte to | |
549 | * replace a none pte. NOTE! This should only be called when *pte is already | |
550 | * cleared so we will never accidentally replace something valuable. Meanwhile | |
551 | * none pte also means we are not demoting the pte so tlb flushed is not needed. | |
552 | * E.g., when pte cleared the caller should have taken care of the tlb flush. | |
553 | * | |
554 | * Must be called with pgtable lock held so that no thread will see the none | |
555 | * pte, and if they see it, they'll fault and serialize at the pgtable lock. | |
556 | * | |
557 | * This function is a no-op if PTE_MARKER_UFFD_WP is not enabled. | |
558 | */ | |
559 | static inline void | |
560 | pte_install_uffd_wp_if_needed(struct vm_area_struct *vma, unsigned long addr, | |
561 | pte_t *pte, pte_t pteval) | |
562 | { | |
563 | #ifdef CONFIG_PTE_MARKER_UFFD_WP | |
564 | bool arm_uffd_pte = false; | |
565 | ||
566 | /* The current status of the pte should be "cleared" before calling */ | |
c33c7948 | 567 | WARN_ON_ONCE(!pte_none(ptep_get(pte))); |
999dad82 | 568 | |
2bad466c PX |
569 | /* |
570 | * NOTE: userfaultfd_wp_unpopulated() doesn't need this whole | |
571 | * thing, because when zapping either it means it's dropping the | |
572 | * page, or in TTU where the present pte will be quickly replaced | |
573 | * with a swap pte. There's no way of leaking the bit. | |
574 | */ | |
999dad82 PX |
575 | if (vma_is_anonymous(vma) || !userfaultfd_wp(vma)) |
576 | return; | |
577 | ||
578 | /* A uffd-wp wr-protected normal pte */ | |
579 | if (unlikely(pte_present(pteval) && pte_uffd_wp(pteval))) | |
580 | arm_uffd_pte = true; | |
581 | ||
582 | /* | |
583 | * A uffd-wp wr-protected swap pte. Note: this should even cover an | |
584 | * existing pte marker with uffd-wp bit set. | |
585 | */ | |
586 | if (unlikely(pte_swp_uffd_wp_any(pteval))) | |
587 | arm_uffd_pte = true; | |
588 | ||
589 | if (unlikely(arm_uffd_pte)) | |
590 | set_pte_at(vma->vm_mm, addr, pte, | |
591 | make_pte_marker(PTE_MARKER_UFFD_WP)); | |
592 | #endif | |
593 | } | |
36090def | 594 | |
8788f678 YZ |
595 | static inline bool vma_has_recency(struct vm_area_struct *vma) |
596 | { | |
597 | if (vma->vm_flags & (VM_SEQ_READ | VM_RAND_READ)) | |
598 | return false; | |
599 | ||
17e81022 YZ |
600 | if (vma->vm_file && (vma->vm_file->f_mode & FMODE_NOREUSE)) |
601 | return false; | |
602 | ||
8788f678 YZ |
603 | return true; |
604 | } | |
605 | ||
b2e18538 | 606 | #endif |