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