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 */
28static inline int folio_is_file_lru(struct folio *folio)
29{
30 return !folio_test_swapbacked(folio);
31}
32
33static inline int page_is_file_lru(struct page *page)
34{
35 return folio_is_file_lru(page_folio(page));
36}
37
38static __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, idx: NR_LRU_BASE + lru, val: nr_pages);
48 __mod_zone_page_state(&pgdat->node_zones[zid],
49 item: NR_ZONE_LRU_BASE + lru, nr_pages);
50}
51
52static __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 */
66static __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 */
87static __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
106static 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
113static 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
121static inline bool lru_gen_in_fault(void)
122{
123 return current->in_lru_fault;
124}
125
126static inline int lru_gen_from_seq(unsigned long seq)
127{
128 return seq % MAX_NR_GENS;
129}
130
131static inline int lru_hist_from_seq(unsigned long seq)
132{
133 return seq % NR_HIST_GENS;
134}
135
136static 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
144static 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
158static 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
165static 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(seq: max_seq) || gen == lru_gen_from_seq(seq: max_seq - 1);
173}
174
175static 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, gen: new_gen))
198 lru += LRU_ACTIVE;
199 __update_lru_size(lruvec, lru, zid: zone, nr_pages: delta);
200 return;
201 }
202
203 /* deletion */
204 if (new_gen < 0) {
205 if (lru_gen_is_active(lruvec, gen: old_gen))
206 lru += LRU_ACTIVE;
207 __update_lru_size(lruvec, lru, zid: zone, nr_pages: -delta);
208 return;
209 }
210
211 /* promotion */
212 if (!lru_gen_is_active(lruvec, gen: old_gen) && lru_gen_is_active(lruvec, gen: new_gen)) {
213 __update_lru_size(lruvec, lru, zid: zone, nr_pages: -delta);
214 __update_lru_size(lruvec, lru: lru + LRU_ACTIVE, zid: zone, nr_pages: 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
221static 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 three common cases for this page:
236 * 1. If it's hot, e.g., freshly faulted in or previously hot and
237 * migrated, add it to the youngest generation.
238 * 2. If it's cold but can't be evicted immediately, i.e., an anon page
239 * not in swapcache or a dirty page pending writeback, add it to the
240 * second oldest generation.
241 * 3. Everything else (clean, cold) is added to the oldest generation.
242 */
243 if (folio_test_active(folio))
244 seq = lrugen->max_seq;
245 else if ((type == LRU_GEN_ANON && !folio_test_swapcache(folio)) ||
246 (folio_test_reclaim(folio) &&
247 (folio_test_dirty(folio) || folio_test_writeback(folio))))
248 seq = lrugen->min_seq[type] + 1;
249 else
250 seq = lrugen->min_seq[type];
251
252 gen = lru_gen_from_seq(seq);
253 flags = (gen + 1UL) << LRU_GEN_PGOFF;
254 /* see the comment on MIN_NR_GENS about PG_active */
255 set_mask_bits(&folio->flags, LRU_GEN_MASK | BIT(PG_active), flags);
256
257 lru_gen_update_size(lruvec, folio, old_gen: -1, new_gen: gen);
258 /* for folio_rotate_reclaimable() */
259 if (reclaiming)
260 list_add_tail(new: &folio->lru, head: &lrugen->folios[gen][type][zone]);
261 else
262 list_add(new: &folio->lru, head: &lrugen->folios[gen][type][zone]);
263
264 return true;
265}
266
267static inline bool lru_gen_del_folio(struct lruvec *lruvec, struct folio *folio, bool reclaiming)
268{
269 unsigned long flags;
270 int gen = folio_lru_gen(folio);
271
272 if (gen < 0)
273 return false;
274
275 VM_WARN_ON_ONCE_FOLIO(folio_test_active(folio), folio);
276 VM_WARN_ON_ONCE_FOLIO(folio_test_unevictable(folio), folio);
277
278 /* for folio_migrate_flags() */
279 flags = !reclaiming && lru_gen_is_active(lruvec, gen) ? BIT(PG_active) : 0;
280 flags = set_mask_bits(&folio->flags, LRU_GEN_MASK, flags);
281 gen = ((flags & LRU_GEN_MASK) >> LRU_GEN_PGOFF) - 1;
282
283 lru_gen_update_size(lruvec, folio, old_gen: gen, new_gen: -1);
284 list_del(entry: &folio->lru);
285
286 return true;
287}
288
289#else /* !CONFIG_LRU_GEN */
290
291static inline bool lru_gen_enabled(void)
292{
293 return false;
294}
295
296static inline bool lru_gen_in_fault(void)
297{
298 return false;
299}
300
301static inline bool lru_gen_add_folio(struct lruvec *lruvec, struct folio *folio, bool reclaiming)
302{
303 return false;
304}
305
306static inline bool lru_gen_del_folio(struct lruvec *lruvec, struct folio *folio, bool reclaiming)
307{
308 return false;
309}
310
311#endif /* CONFIG_LRU_GEN */
312
313static __always_inline
314void lruvec_add_folio(struct lruvec *lruvec, struct folio *folio)
315{
316 enum lru_list lru = folio_lru_list(folio);
317
318 if (lru_gen_add_folio(lruvec, folio, reclaiming: false))
319 return;
320
321 update_lru_size(lruvec, lru, zid: folio_zonenum(folio),
322 nr_pages: folio_nr_pages(folio));
323 if (lru != LRU_UNEVICTABLE)
324 list_add(new: &folio->lru, head: &lruvec->lists[lru]);
325}
326
327static __always_inline
328void lruvec_add_folio_tail(struct lruvec *lruvec, struct folio *folio)
329{
330 enum lru_list lru = folio_lru_list(folio);
331
332 if (lru_gen_add_folio(lruvec, folio, reclaiming: true))
333 return;
334
335 update_lru_size(lruvec, lru, zid: folio_zonenum(folio),
336 nr_pages: folio_nr_pages(folio));
337 /* This is not expected to be used on LRU_UNEVICTABLE */
338 list_add_tail(new: &folio->lru, head: &lruvec->lists[lru]);
339}
340
341static __always_inline
342void lruvec_del_folio(struct lruvec *lruvec, struct folio *folio)
343{
344 enum lru_list lru = folio_lru_list(folio);
345
346 if (lru_gen_del_folio(lruvec, folio, reclaiming: false))
347 return;
348
349 if (lru != LRU_UNEVICTABLE)
350 list_del(entry: &folio->lru);
351 update_lru_size(lruvec, lru, zid: folio_zonenum(folio),
352 nr_pages: -folio_nr_pages(folio));
353}
354
355#ifdef CONFIG_ANON_VMA_NAME
356/* mmap_lock should be read-locked */
357static inline void anon_vma_name_get(struct anon_vma_name *anon_name)
358{
359 if (anon_name)
360 kref_get(kref: &anon_name->kref);
361}
362
363static inline void anon_vma_name_put(struct anon_vma_name *anon_name)
364{
365 if (anon_name)
366 kref_put(kref: &anon_name->kref, release: anon_vma_name_free);
367}
368
369static inline
370struct anon_vma_name *anon_vma_name_reuse(struct anon_vma_name *anon_name)
371{
372 /* Prevent anon_name refcount saturation early on */
373 if (kref_read(kref: &anon_name->kref) < REFCOUNT_MAX) {
374 anon_vma_name_get(anon_name);
375 return anon_name;
376
377 }
378 return anon_vma_name_alloc(name: anon_name->name);
379}
380
381static inline void dup_anon_vma_name(struct vm_area_struct *orig_vma,
382 struct vm_area_struct *new_vma)
383{
384 struct anon_vma_name *anon_name = anon_vma_name(vma: orig_vma);
385
386 if (anon_name)
387 new_vma->anon_name = anon_vma_name_reuse(anon_name);
388}
389
390static inline void free_anon_vma_name(struct vm_area_struct *vma)
391{
392 /*
393 * Not using anon_vma_name because it generates a warning if mmap_lock
394 * is not held, which might be the case here.
395 */
396 anon_vma_name_put(anon_name: vma->anon_name);
397}
398
399static inline bool anon_vma_name_eq(struct anon_vma_name *anon_name1,
400 struct anon_vma_name *anon_name2)
401{
402 if (anon_name1 == anon_name2)
403 return true;
404
405 return anon_name1 && anon_name2 &&
406 !strcmp(anon_name1->name, anon_name2->name);
407}
408
409#else /* CONFIG_ANON_VMA_NAME */
410static inline void anon_vma_name_get(struct anon_vma_name *anon_name) {}
411static inline void anon_vma_name_put(struct anon_vma_name *anon_name) {}
412static inline void dup_anon_vma_name(struct vm_area_struct *orig_vma,
413 struct vm_area_struct *new_vma) {}
414static inline void free_anon_vma_name(struct vm_area_struct *vma) {}
415
416static inline bool anon_vma_name_eq(struct anon_vma_name *anon_name1,
417 struct anon_vma_name *anon_name2)
418{
419 return true;
420}
421
422#endif /* CONFIG_ANON_VMA_NAME */
423
424static inline void init_tlb_flush_pending(struct mm_struct *mm)
425{
426 atomic_set(v: &mm->tlb_flush_pending, i: 0);
427}
428
429static inline void inc_tlb_flush_pending(struct mm_struct *mm)
430{
431 atomic_inc(v: &mm->tlb_flush_pending);
432 /*
433 * The only time this value is relevant is when there are indeed pages
434 * to flush. And we'll only flush pages after changing them, which
435 * requires the PTL.
436 *
437 * So the ordering here is:
438 *
439 * atomic_inc(&mm->tlb_flush_pending);
440 * spin_lock(&ptl);
441 * ...
442 * set_pte_at();
443 * spin_unlock(&ptl);
444 *
445 * spin_lock(&ptl)
446 * mm_tlb_flush_pending();
447 * ....
448 * spin_unlock(&ptl);
449 *
450 * flush_tlb_range();
451 * atomic_dec(&mm->tlb_flush_pending);
452 *
453 * Where the increment if constrained by the PTL unlock, it thus
454 * ensures that the increment is visible if the PTE modification is
455 * visible. After all, if there is no PTE modification, nobody cares
456 * about TLB flushes either.
457 *
458 * This very much relies on users (mm_tlb_flush_pending() and
459 * mm_tlb_flush_nested()) only caring about _specific_ PTEs (and
460 * therefore specific PTLs), because with SPLIT_PTE_PTLOCKS and RCpc
461 * locks (PPC) the unlock of one doesn't order against the lock of
462 * another PTL.
463 *
464 * The decrement is ordered by the flush_tlb_range(), such that
465 * mm_tlb_flush_pending() will not return false unless all flushes have
466 * completed.
467 */
468}
469
470static inline void dec_tlb_flush_pending(struct mm_struct *mm)
471{
472 /*
473 * See inc_tlb_flush_pending().
474 *
475 * This cannot be smp_mb__before_atomic() because smp_mb() simply does
476 * not order against TLB invalidate completion, which is what we need.
477 *
478 * Therefore we must rely on tlb_flush_*() to guarantee order.
479 */
480 atomic_dec(v: &mm->tlb_flush_pending);
481}
482
483static inline bool mm_tlb_flush_pending(struct mm_struct *mm)
484{
485 /*
486 * Must be called after having acquired the PTL; orders against that
487 * PTLs release and therefore ensures that if we observe the modified
488 * PTE we must also observe the increment from inc_tlb_flush_pending().
489 *
490 * That is, it only guarantees to return true if there is a flush
491 * pending for _this_ PTL.
492 */
493 return atomic_read(v: &mm->tlb_flush_pending);
494}
495
496static inline bool mm_tlb_flush_nested(struct mm_struct *mm)
497{
498 /*
499 * Similar to mm_tlb_flush_pending(), we must have acquired the PTL
500 * for which there is a TLB flush pending in order to guarantee
501 * we've seen both that PTE modification and the increment.
502 *
503 * (no requirement on actually still holding the PTL, that is irrelevant)
504 */
505 return atomic_read(v: &mm->tlb_flush_pending) > 1;
506}
507
508#ifdef CONFIG_MMU
509/*
510 * Computes the pte marker to copy from the given source entry into dst_vma.
511 * If no marker should be copied, returns 0.
512 * The caller should insert a new pte created with make_pte_marker().
513 */
514static inline pte_marker copy_pte_marker(
515 swp_entry_t entry, struct vm_area_struct *dst_vma)
516{
517 pte_marker srcm = pte_marker_get(entry);
518 /* Always copy error entries. */
519 pte_marker dstm = srcm & PTE_MARKER_POISONED;
520
521 /* Only copy PTE markers if UFFD register matches. */
522 if ((srcm & PTE_MARKER_UFFD_WP) && userfaultfd_wp(vma: dst_vma))
523 dstm |= PTE_MARKER_UFFD_WP;
524
525 return dstm;
526}
527#endif
528
529/*
530 * If this pte is wr-protected by uffd-wp in any form, arm the special pte to
531 * replace a none pte. NOTE! This should only be called when *pte is already
532 * cleared so we will never accidentally replace something valuable. Meanwhile
533 * none pte also means we are not demoting the pte so tlb flushed is not needed.
534 * E.g., when pte cleared the caller should have taken care of the tlb flush.
535 *
536 * Must be called with pgtable lock held so that no thread will see the none
537 * pte, and if they see it, they'll fault and serialize at the pgtable lock.
538 *
539 * This function is a no-op if PTE_MARKER_UFFD_WP is not enabled.
540 */
541static inline void
542pte_install_uffd_wp_if_needed(struct vm_area_struct *vma, unsigned long addr,
543 pte_t *pte, pte_t pteval)
544{
545#ifdef CONFIG_PTE_MARKER_UFFD_WP
546 bool arm_uffd_pte = false;
547
548 /* The current status of the pte should be "cleared" before calling */
549 WARN_ON_ONCE(!pte_none(ptep_get(pte)));
550
551 /*
552 * NOTE: userfaultfd_wp_unpopulated() doesn't need this whole
553 * thing, because when zapping either it means it's dropping the
554 * page, or in TTU where the present pte will be quickly replaced
555 * with a swap pte. There's no way of leaking the bit.
556 */
557 if (vma_is_anonymous(vma) || !userfaultfd_wp(vma))
558 return;
559
560 /* A uffd-wp wr-protected normal pte */
561 if (unlikely(pte_present(pteval) && pte_uffd_wp(pteval)))
562 arm_uffd_pte = true;
563
564 /*
565 * A uffd-wp wr-protected swap pte. Note: this should even cover an
566 * existing pte marker with uffd-wp bit set.
567 */
568 if (unlikely(pte_swp_uffd_wp_any(pteval)))
569 arm_uffd_pte = true;
570
571 if (unlikely(arm_uffd_pte))
572 set_pte_at(vma->vm_mm, addr, pte,
573 make_pte_marker(PTE_MARKER_UFFD_WP));
574#endif
575}
576
577static inline bool vma_has_recency(struct vm_area_struct *vma)
578{
579 if (vma->vm_flags & (VM_SEQ_READ | VM_RAND_READ))
580 return false;
581
582 if (vma->vm_file && (vma->vm_file->f_mode & FMODE_NOREUSE))
583 return false;
584
585 return true;
586}
587
588#endif
589

source code of linux/include/linux/mm_inline.h