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(const 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(const 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, bool workingset)
137{
138 VM_WARN_ON_ONCE(refs > BIT(LRU_REFS_WIDTH));
139
140 /* see the comment on MAX_NR_TIERS */
141 return workingset ? MAX_NR_TIERS - 1 : order_base_2(refs);
142}
143
144static inline int folio_lru_refs(const struct folio *folio)
145{
146 unsigned long flags = READ_ONCE(folio->flags.f);
147
148 if (!(flags & BIT(PG_referenced)))
149 return 0;
150 /*
151 * Return the total number of accesses including PG_referenced. Also see
152 * the comment on LRU_REFS_FLAGS.
153 */
154 return ((flags & LRU_REFS_MASK) >> LRU_REFS_PGOFF) + 1;
155}
156
157static inline int folio_lru_gen(const struct folio *folio)
158{
159 unsigned long flags = READ_ONCE(folio->flags.f);
160
161 return ((flags & LRU_GEN_MASK) >> LRU_GEN_PGOFF) - 1;
162}
163
164static inline bool lru_gen_is_active(const 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
174static 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;
181 struct lru_gen_folio *lrugen = &lruvec->lrugen;
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 }
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));
218}
219
220static inline unsigned long lru_gen_folio_seq(const struct lruvec *lruvec,
221 const struct folio *folio,
222 bool reclaiming)
223{
224 int gen;
225 int type = folio_is_file_lru(folio);
226 const struct lru_gen_folio *lrugen = &lruvec->lrugen;
227
228 /*
229 * +-----------------------------------+-----------------------------------+
230 * | Accessed through page tables and | Accessed through file descriptors |
231 * | promoted by folio_update_gen() | and protected by folio_inc_gen() |
232 * +-----------------------------------+-----------------------------------+
233 * | PG_active (set while isolated) | |
234 * +-----------------+-----------------+-----------------+-----------------+
235 * | PG_workingset | PG_referenced | PG_workingset | LRU_REFS_FLAGS |
236 * +-----------------------------------+-----------------------------------+
237 * |<---------- MIN_NR_GENS ---------->| |
238 * |<---------------------------- MAX_NR_GENS ---------------------------->|
239 */
240 if (folio_test_active(folio))
241 gen = MIN_NR_GENS - folio_test_workingset(folio);
242 else if (reclaiming)
243 gen = MAX_NR_GENS;
244 else if ((!folio_is_file_lru(folio) && !folio_test_swapcache(folio)) ||
245 (folio_test_reclaim(folio) &&
246 (folio_test_dirty(folio) || folio_test_writeback(folio))))
247 gen = MIN_NR_GENS;
248 else
249 gen = MAX_NR_GENS - folio_test_workingset(folio);
250
251 return max(READ_ONCE(lrugen->max_seq) - gen + 1, READ_ONCE(lrugen->min_seq[type]));
252}
253
254static inline bool lru_gen_add_folio(struct lruvec *lruvec, struct folio *folio, bool reclaiming)
255{
256 unsigned long seq;
257 unsigned long flags;
258 int gen = folio_lru_gen(folio);
259 int type = folio_is_file_lru(folio);
260 int zone = folio_zonenum(folio);
261 struct lru_gen_folio *lrugen = &lruvec->lrugen;
262
263 VM_WARN_ON_ONCE_FOLIO(gen != -1, folio);
264
265 if (folio_test_unevictable(folio) || !lrugen->enabled)
266 return false;
267
268 seq = lru_gen_folio_seq(lruvec, folio, reclaiming);
269 gen = lru_gen_from_seq(seq);
270 flags = (gen + 1UL) << LRU_GEN_PGOFF;
271 /* see the comment on MIN_NR_GENS about PG_active */
272 set_mask_bits(&folio->flags.f, LRU_GEN_MASK | BIT(PG_active), flags);
273
274 lru_gen_update_size(lruvec, folio, -1, gen);
275 /* for folio_rotate_reclaimable() */
276 if (reclaiming)
277 list_add_tail(&folio->lru, &lrugen->folios[gen][type][zone]);
278 else
279 list_add(&folio->lru, &lrugen->folios[gen][type][zone]);
280
281 return true;
282}
283
284static inline bool lru_gen_del_folio(struct lruvec *lruvec, struct folio *folio, bool reclaiming)
285{
286 unsigned long flags;
287 int gen = folio_lru_gen(folio);
288
289 if (gen < 0)
290 return false;
291
292 VM_WARN_ON_ONCE_FOLIO(folio_test_active(folio), folio);
293 VM_WARN_ON_ONCE_FOLIO(folio_test_unevictable(folio), folio);
294
295 /* for folio_migrate_flags() */
296 flags = !reclaiming && lru_gen_is_active(lruvec, gen) ? BIT(PG_active) : 0;
297 flags = set_mask_bits(&folio->flags.f, LRU_GEN_MASK, flags);
298 gen = ((flags & LRU_GEN_MASK) >> LRU_GEN_PGOFF) - 1;
299
300 lru_gen_update_size(lruvec, folio, gen, -1);
301 list_del(&folio->lru);
302
303 return true;
304}
305
306static inline void folio_migrate_refs(struct folio *new, const struct folio *old)
307{
308 unsigned long refs = READ_ONCE(old->flags.f) & LRU_REFS_MASK;
309
310 set_mask_bits(&new->flags.f, LRU_REFS_MASK, refs);
311}
312#else /* !CONFIG_LRU_GEN */
313
314static inline bool lru_gen_enabled(void)
315{
316 return false;
317}
318
319static inline bool lru_gen_in_fault(void)
320{
321 return false;
322}
323
324static inline bool lru_gen_add_folio(struct lruvec *lruvec, struct folio *folio, bool reclaiming)
325{
326 return false;
327}
328
329static inline bool lru_gen_del_folio(struct lruvec *lruvec, struct folio *folio, bool reclaiming)
330{
331 return false;
332}
333
334static inline void folio_migrate_refs(struct folio *new, const struct folio *old)
335{
336
337}
338#endif /* CONFIG_LRU_GEN */
339
340static __always_inline
341void lruvec_add_folio(struct lruvec *lruvec, struct folio *folio)
342{
343 enum lru_list lru = folio_lru_list(folio);
344
345 if (lru_gen_add_folio(lruvec, folio, reclaiming: false))
346 return;
347
348 update_lru_size(lruvec, lru, zid: folio_zonenum(folio),
349 nr_pages: folio_nr_pages(folio));
350 if (lru != LRU_UNEVICTABLE)
351 list_add(new: &folio->lru, head: &lruvec->lists[lru]);
352}
353
354static __always_inline
355void lruvec_add_folio_tail(struct lruvec *lruvec, struct folio *folio)
356{
357 enum lru_list lru = folio_lru_list(folio);
358
359 if (lru_gen_add_folio(lruvec, folio, reclaiming: true))
360 return;
361
362 update_lru_size(lruvec, lru, zid: folio_zonenum(folio),
363 nr_pages: folio_nr_pages(folio));
364 /* This is not expected to be used on LRU_UNEVICTABLE */
365 list_add_tail(new: &folio->lru, head: &lruvec->lists[lru]);
366}
367
368static __always_inline
369void lruvec_del_folio(struct lruvec *lruvec, struct folio *folio)
370{
371 enum lru_list lru = folio_lru_list(folio);
372
373 if (lru_gen_del_folio(lruvec, folio, reclaiming: false))
374 return;
375
376 if (lru != LRU_UNEVICTABLE)
377 list_del(entry: &folio->lru);
378 update_lru_size(lruvec, lru, zid: folio_zonenum(folio),
379 nr_pages: -folio_nr_pages(folio));
380}
381
382#ifdef CONFIG_ANON_VMA_NAME
383/* mmap_lock should be read-locked */
384static inline void anon_vma_name_get(struct anon_vma_name *anon_name)
385{
386 if (anon_name)
387 kref_get(&anon_name->kref);
388}
389
390static inline void anon_vma_name_put(struct anon_vma_name *anon_name)
391{
392 if (anon_name)
393 kref_put(&anon_name->kref, anon_vma_name_free);
394}
395
396static inline
397struct anon_vma_name *anon_vma_name_reuse(struct anon_vma_name *anon_name)
398{
399 /* Prevent anon_name refcount saturation early on */
400 if (kref_read(&anon_name->kref) < REFCOUNT_MAX) {
401 anon_vma_name_get(anon_name);
402 return anon_name;
403
404 }
405 return anon_vma_name_alloc(anon_name->name);
406}
407
408static inline void dup_anon_vma_name(struct vm_area_struct *orig_vma,
409 struct vm_area_struct *new_vma)
410{
411 struct anon_vma_name *anon_name = anon_vma_name(orig_vma);
412
413 if (anon_name)
414 new_vma->anon_name = anon_vma_name_reuse(anon_name);
415}
416
417static inline void free_anon_vma_name(struct vm_area_struct *vma)
418{
419 /*
420 * Not using anon_vma_name because it generates a warning if mmap_lock
421 * is not held, which might be the case here.
422 */
423 anon_vma_name_put(vma->anon_name);
424}
425
426static inline bool anon_vma_name_eq(struct anon_vma_name *anon_name1,
427 struct anon_vma_name *anon_name2)
428{
429 if (anon_name1 == anon_name2)
430 return true;
431
432 return anon_name1 && anon_name2 &&
433 !strcmp(anon_name1->name, anon_name2->name);
434}
435
436#else /* CONFIG_ANON_VMA_NAME */
437static inline void anon_vma_name_get(struct anon_vma_name *anon_name) {}
438static inline void anon_vma_name_put(struct anon_vma_name *anon_name) {}
439static inline void dup_anon_vma_name(struct vm_area_struct *orig_vma,
440 struct vm_area_struct *new_vma) {}
441static inline void free_anon_vma_name(struct vm_area_struct *vma) {}
442
443static inline bool anon_vma_name_eq(struct anon_vma_name *anon_name1,
444 struct anon_vma_name *anon_name2)
445{
446 return true;
447}
448
449#endif /* CONFIG_ANON_VMA_NAME */
450
451void pfnmap_track_ctx_release(struct kref *ref);
452
453static inline void init_tlb_flush_pending(struct mm_struct *mm)
454{
455 atomic_set(v: &mm->tlb_flush_pending, i: 0);
456}
457
458static inline void inc_tlb_flush_pending(struct mm_struct *mm)
459{
460 atomic_inc(v: &mm->tlb_flush_pending);
461 /*
462 * The only time this value is relevant is when there are indeed pages
463 * to flush. And we'll only flush pages after changing them, which
464 * requires the PTL.
465 *
466 * So the ordering here is:
467 *
468 * atomic_inc(&mm->tlb_flush_pending);
469 * spin_lock(&ptl);
470 * ...
471 * set_pte_at();
472 * spin_unlock(&ptl);
473 *
474 * spin_lock(&ptl)
475 * mm_tlb_flush_pending();
476 * ....
477 * spin_unlock(&ptl);
478 *
479 * flush_tlb_range();
480 * atomic_dec(&mm->tlb_flush_pending);
481 *
482 * Where the increment if constrained by the PTL unlock, it thus
483 * ensures that the increment is visible if the PTE modification is
484 * visible. After all, if there is no PTE modification, nobody cares
485 * about TLB flushes either.
486 *
487 * This very much relies on users (mm_tlb_flush_pending() and
488 * mm_tlb_flush_nested()) only caring about _specific_ PTEs (and
489 * therefore specific PTLs), because with SPLIT_PTE_PTLOCKS and RCpc
490 * locks (PPC) the unlock of one doesn't order against the lock of
491 * another PTL.
492 *
493 * The decrement is ordered by the flush_tlb_range(), such that
494 * mm_tlb_flush_pending() will not return false unless all flushes have
495 * completed.
496 */
497}
498
499static inline void dec_tlb_flush_pending(struct mm_struct *mm)
500{
501 /*
502 * See inc_tlb_flush_pending().
503 *
504 * This cannot be smp_mb__before_atomic() because smp_mb() simply does
505 * not order against TLB invalidate completion, which is what we need.
506 *
507 * Therefore we must rely on tlb_flush_*() to guarantee order.
508 */
509 atomic_dec(v: &mm->tlb_flush_pending);
510}
511
512static inline bool mm_tlb_flush_pending(const struct mm_struct *mm)
513{
514 /*
515 * Must be called after having acquired the PTL; orders against that
516 * PTLs release and therefore ensures that if we observe the modified
517 * PTE we must also observe the increment from inc_tlb_flush_pending().
518 *
519 * That is, it only guarantees to return true if there is a flush
520 * pending for _this_ PTL.
521 */
522 return atomic_read(v: &mm->tlb_flush_pending);
523}
524
525static inline bool mm_tlb_flush_nested(const struct mm_struct *mm)
526{
527 /*
528 * Similar to mm_tlb_flush_pending(), we must have acquired the PTL
529 * for which there is a TLB flush pending in order to guarantee
530 * we've seen both that PTE modification and the increment.
531 *
532 * (no requirement on actually still holding the PTL, that is irrelevant)
533 */
534 return atomic_read(v: &mm->tlb_flush_pending) > 1;
535}
536
537#ifdef CONFIG_MMU
538/*
539 * Computes the pte marker to copy from the given source entry into dst_vma.
540 * If no marker should be copied, returns 0.
541 * The caller should insert a new pte created with make_pte_marker().
542 */
543static inline pte_marker copy_pte_marker(
544 swp_entry_t entry, struct vm_area_struct *dst_vma)
545{
546 pte_marker srcm = pte_marker_get(entry);
547 /* Always copy error entries. */
548 pte_marker dstm = srcm & (PTE_MARKER_POISONED | PTE_MARKER_GUARD);
549
550 /* Only copy PTE markers if UFFD register matches. */
551 if ((srcm & PTE_MARKER_UFFD_WP) && userfaultfd_wp(vma: dst_vma))
552 dstm |= PTE_MARKER_UFFD_WP;
553
554 return dstm;
555}
556#endif
557
558/*
559 * If this pte is wr-protected by uffd-wp in any form, arm the special pte to
560 * replace a none pte. NOTE! This should only be called when *pte is already
561 * cleared so we will never accidentally replace something valuable. Meanwhile
562 * none pte also means we are not demoting the pte so tlb flushed is not needed.
563 * E.g., when pte cleared the caller should have taken care of the tlb flush.
564 *
565 * Must be called with pgtable lock held so that no thread will see the none
566 * pte, and if they see it, they'll fault and serialize at the pgtable lock.
567 *
568 * Returns true if an uffd-wp pte was installed, false otherwise.
569 */
570static inline bool
571pte_install_uffd_wp_if_needed(struct vm_area_struct *vma, unsigned long addr,
572 pte_t *pte, pte_t pteval)
573{
574#ifdef CONFIG_PTE_MARKER_UFFD_WP
575 bool arm_uffd_pte = false;
576
577 /* The current status of the pte should be "cleared" before calling */
578 WARN_ON_ONCE(!pte_none(ptep_get(pte)));
579
580 /*
581 * NOTE: userfaultfd_wp_unpopulated() doesn't need this whole
582 * thing, because when zapping either it means it's dropping the
583 * page, or in TTU where the present pte will be quickly replaced
584 * with a swap pte. There's no way of leaking the bit.
585 */
586 if (vma_is_anonymous(vma) || !userfaultfd_wp(vma))
587 return false;
588
589 /* A uffd-wp wr-protected normal pte */
590 if (unlikely(pte_present(pteval) && pte_uffd_wp(pteval)))
591 arm_uffd_pte = true;
592
593 /*
594 * A uffd-wp wr-protected swap pte. Note: this should even cover an
595 * existing pte marker with uffd-wp bit set.
596 */
597 if (unlikely(pte_swp_uffd_wp_any(pteval)))
598 arm_uffd_pte = true;
599
600 if (unlikely(arm_uffd_pte)) {
601 set_pte_at(vma->vm_mm, addr, pte,
602 make_pte_marker(PTE_MARKER_UFFD_WP));
603 return true;
604 }
605#endif
606 return false;
607}
608
609static inline bool vma_has_recency(const struct vm_area_struct *vma)
610{
611 if (vma->vm_flags & (VM_SEQ_READ | VM_RAND_READ))
612 return false;
613
614 if (vma->vm_file && (vma->vm_file->f_mode & FMODE_NOREUSE))
615 return false;
616
617 return true;
618}
619
620/**
621 * num_pages_contiguous() - determine the number of contiguous pages
622 * that represent contiguous PFNs
623 * @pages: an array of page pointers
624 * @nr_pages: length of the array, at least 1
625 *
626 * Determine the number of contiguous pages that represent contiguous PFNs
627 * in @pages, starting from the first page.
628 *
629 * In some kernel configs contiguous PFNs will not have contiguous struct
630 * pages. In these configurations num_pages_contiguous() will return a num
631 * smaller than ideal number. The caller should continue to check for pfn
632 * contiguity after each call to num_pages_contiguous().
633 *
634 * Returns the number of contiguous pages.
635 */
636static inline size_t num_pages_contiguous(struct page **pages, size_t nr_pages)
637{
638 struct page *cur_page = pages[0];
639 unsigned long section = memdesc_section(mdf: cur_page->flags);
640 size_t i;
641
642 for (i = 1; i < nr_pages; i++) {
643 if (++cur_page != pages[i])
644 break;
645 /*
646 * In unproblematic kernel configs, page_to_section() == 0 and
647 * the whole check will get optimized out.
648 */
649 if (memdesc_section(mdf: cur_page->flags) != section)
650 break;
651 }
652
653 return i;
654}
655
656#endif
657