1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef LINUX_MM_INLINE_H
3 #define LINUX_MM_INLINE_H
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>
14 * folio_is_file_lru - Should the folio be on a file LRU or anon LRU?
15 * @folio: The folio to test.
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.
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.
28 static inline int folio_is_file_lru(struct folio *folio)
30 return !folio_test_swapbacked(folio);
33 static inline int page_is_file_lru(struct page *page)
35 return folio_is_file_lru(page_folio(page));
38 static __always_inline void __update_lru_size(struct lruvec *lruvec,
39 enum lru_list lru, enum zone_type zid,
42 struct pglist_data *pgdat = lruvec_pgdat(lruvec);
44 lockdep_assert_held(&lruvec->lru_lock);
45 WARN_ON_ONCE(nr_pages != (int)nr_pages);
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);
52 static __always_inline void update_lru_size(struct lruvec *lruvec,
53 enum lru_list lru, enum zone_type zid,
56 __update_lru_size(lruvec, lru, zid, nr_pages);
58 mem_cgroup_update_lru_size(lruvec, lru, zid, nr_pages);
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.
66 static __always_inline void __folio_clear_lru_flags(struct folio *folio)
68 VM_BUG_ON_FOLIO(!folio_test_lru(folio), folio);
70 __folio_clear_lru(folio);
72 /* this shouldn't happen, so leave the flags to bad_page() */
73 if (folio_test_active(folio) && folio_test_unevictable(folio))
76 __folio_clear_active(folio);
77 __folio_clear_unevictable(folio);
81 * folio_lru_list - Which LRU list should a folio be on?
82 * @folio: The folio to test.
84 * Return: The LRU list a folio should be on, as an index
85 * into the array of LRU lists.
87 static __always_inline enum lru_list folio_lru_list(struct folio *folio)
91 VM_BUG_ON_FOLIO(folio_test_active(folio) && folio_test_unevictable(folio), folio);
93 if (folio_test_unevictable(folio))
94 return LRU_UNEVICTABLE;
96 lru = folio_is_file_lru(folio) ? LRU_INACTIVE_FILE : LRU_INACTIVE_ANON;
97 if (folio_test_active(folio))
103 #ifdef CONFIG_LRU_GEN
105 #ifdef CONFIG_LRU_GEN_ENABLED
106 static inline bool lru_gen_enabled(void)
108 DECLARE_STATIC_KEY_TRUE(lru_gen_caps[NR_LRU_GEN_CAPS]);
110 return static_branch_likely(&lru_gen_caps[LRU_GEN_CORE]);
113 static inline bool lru_gen_enabled(void)
115 DECLARE_STATIC_KEY_FALSE(lru_gen_caps[NR_LRU_GEN_CAPS]);
117 return static_branch_unlikely(&lru_gen_caps[LRU_GEN_CORE]);
121 static inline bool lru_gen_in_fault(void)
123 return current->in_lru_fault;
126 static inline int lru_gen_from_seq(unsigned long seq)
128 return seq % MAX_NR_GENS;
131 static inline int lru_hist_from_seq(unsigned long seq)
133 return seq % NR_HIST_GENS;
136 static inline int lru_tier_from_refs(int refs)
138 VM_WARN_ON_ONCE(refs > BIT(LRU_REFS_WIDTH));
140 /* see the comment in folio_lru_refs() */
141 return order_base_2(refs + 1);
144 static inline int folio_lru_refs(struct folio *folio)
146 unsigned long flags = READ_ONCE(folio->flags);
147 bool workingset = flags & BIT(PG_workingset);
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.
155 return ((flags & LRU_REFS_MASK) >> LRU_REFS_PGOFF) + workingset;
158 static inline int folio_lru_gen(struct folio *folio)
160 unsigned long flags = READ_ONCE(folio->flags);
162 return ((flags & LRU_GEN_MASK) >> LRU_GEN_PGOFF) - 1;
165 static inline bool lru_gen_is_active(struct lruvec *lruvec, int gen)
167 unsigned long max_seq = lruvec->lrugen.max_seq;
169 VM_WARN_ON_ONCE(gen >= MAX_NR_GENS);
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);
175 static inline void lru_gen_update_size(struct lruvec *lruvec, struct folio *folio,
176 int old_gen, int new_gen)
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;
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);
189 WRITE_ONCE(lrugen->nr_pages[old_gen][type][zone],
190 lrugen->nr_pages[old_gen][type][zone] - delta);
192 WRITE_ONCE(lrugen->nr_pages[new_gen][type][zone],
193 lrugen->nr_pages[new_gen][type][zone] + delta);
197 if (lru_gen_is_active(lruvec, new_gen))
199 __update_lru_size(lruvec, lru, zone, delta);
205 if (lru_gen_is_active(lruvec, old_gen))
207 __update_lru_size(lruvec, lru, zone, -delta);
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);
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));
221 static inline bool lru_gen_add_folio(struct lruvec *lruvec, struct folio *folio, bool reclaiming)
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;
230 VM_WARN_ON_ONCE_FOLIO(gen != -1, folio);
232 if (folio_test_unevictable(folio) || !lrugen->enabled)
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.
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;
250 seq = lrugen->min_seq[type];
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);
257 lru_gen_update_size(lruvec, folio, -1, gen);
258 /* for folio_rotate_reclaimable() */
260 list_add_tail(&folio->lru, &lrugen->folios[gen][type][zone]);
262 list_add(&folio->lru, &lrugen->folios[gen][type][zone]);
267 static inline bool lru_gen_del_folio(struct lruvec *lruvec, struct folio *folio, bool reclaiming)
270 int gen = folio_lru_gen(folio);
275 VM_WARN_ON_ONCE_FOLIO(folio_test_active(folio), folio);
276 VM_WARN_ON_ONCE_FOLIO(folio_test_unevictable(folio), folio);
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;
283 lru_gen_update_size(lruvec, folio, gen, -1);
284 list_del(&folio->lru);
289 #else /* !CONFIG_LRU_GEN */
291 static inline bool lru_gen_enabled(void)
296 static inline bool lru_gen_in_fault(void)
301 static inline bool lru_gen_add_folio(struct lruvec *lruvec, struct folio *folio, bool reclaiming)
306 static inline bool lru_gen_del_folio(struct lruvec *lruvec, struct folio *folio, bool reclaiming)
311 #endif /* CONFIG_LRU_GEN */
313 static __always_inline
314 void lruvec_add_folio(struct lruvec *lruvec, struct folio *folio)
316 enum lru_list lru = folio_lru_list(folio);
318 if (lru_gen_add_folio(lruvec, folio, false))
321 update_lru_size(lruvec, lru, folio_zonenum(folio),
322 folio_nr_pages(folio));
323 if (lru != LRU_UNEVICTABLE)
324 list_add(&folio->lru, &lruvec->lists[lru]);
327 static __always_inline
328 void lruvec_add_folio_tail(struct lruvec *lruvec, struct folio *folio)
330 enum lru_list lru = folio_lru_list(folio);
332 if (lru_gen_add_folio(lruvec, folio, true))
335 update_lru_size(lruvec, lru, folio_zonenum(folio),
336 folio_nr_pages(folio));
337 /* This is not expected to be used on LRU_UNEVICTABLE */
338 list_add_tail(&folio->lru, &lruvec->lists[lru]);
341 static __always_inline
342 void lruvec_del_folio(struct lruvec *lruvec, struct folio *folio)
344 enum lru_list lru = folio_lru_list(folio);
346 if (lru_gen_del_folio(lruvec, folio, false))
349 if (lru != LRU_UNEVICTABLE)
350 list_del(&folio->lru);
351 update_lru_size(lruvec, lru, folio_zonenum(folio),
352 -folio_nr_pages(folio));
355 #ifdef CONFIG_ANON_VMA_NAME
356 /* mmap_lock should be read-locked */
357 static inline void anon_vma_name_get(struct anon_vma_name *anon_name)
360 kref_get(&anon_name->kref);
363 static inline void anon_vma_name_put(struct anon_vma_name *anon_name)
366 kref_put(&anon_name->kref, anon_vma_name_free);
370 struct anon_vma_name *anon_vma_name_reuse(struct anon_vma_name *anon_name)
372 /* Prevent anon_name refcount saturation early on */
373 if (kref_read(&anon_name->kref) < REFCOUNT_MAX) {
374 anon_vma_name_get(anon_name);
378 return anon_vma_name_alloc(anon_name->name);
381 static inline void dup_anon_vma_name(struct vm_area_struct *orig_vma,
382 struct vm_area_struct *new_vma)
384 struct anon_vma_name *anon_name = anon_vma_name(orig_vma);
387 new_vma->anon_name = anon_vma_name_reuse(anon_name);
390 static inline void free_anon_vma_name(struct vm_area_struct *vma)
393 * Not using anon_vma_name because it generates a warning if mmap_lock
394 * is not held, which might be the case here.
396 anon_vma_name_put(vma->anon_name);
399 static inline bool anon_vma_name_eq(struct anon_vma_name *anon_name1,
400 struct anon_vma_name *anon_name2)
402 if (anon_name1 == anon_name2)
405 return anon_name1 && anon_name2 &&
406 !strcmp(anon_name1->name, anon_name2->name);
409 #else /* CONFIG_ANON_VMA_NAME */
410 static inline void anon_vma_name_get(struct anon_vma_name *anon_name) {}
411 static inline void anon_vma_name_put(struct anon_vma_name *anon_name) {}
412 static inline void dup_anon_vma_name(struct vm_area_struct *orig_vma,
413 struct vm_area_struct *new_vma) {}
414 static inline void free_anon_vma_name(struct vm_area_struct *vma) {}
416 static inline bool anon_vma_name_eq(struct anon_vma_name *anon_name1,
417 struct anon_vma_name *anon_name2)
422 #endif /* CONFIG_ANON_VMA_NAME */
424 static inline void init_tlb_flush_pending(struct mm_struct *mm)
426 atomic_set(&mm->tlb_flush_pending, 0);
429 static inline void inc_tlb_flush_pending(struct mm_struct *mm)
431 atomic_inc(&mm->tlb_flush_pending);
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
437 * So the ordering here is:
439 * atomic_inc(&mm->tlb_flush_pending);
446 * mm_tlb_flush_pending();
451 * atomic_dec(&mm->tlb_flush_pending);
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.
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
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
470 static inline void dec_tlb_flush_pending(struct mm_struct *mm)
473 * See inc_tlb_flush_pending().
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.
478 * Therefore we must rely on tlb_flush_*() to guarantee order.
480 atomic_dec(&mm->tlb_flush_pending);
483 static inline bool mm_tlb_flush_pending(struct mm_struct *mm)
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().
490 * That is, it only guarantees to return true if there is a flush
491 * pending for _this_ PTL.
493 return atomic_read(&mm->tlb_flush_pending);
496 static inline bool mm_tlb_flush_nested(struct mm_struct *mm)
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.
503 * (no requirement on actually still holding the PTL, that is irrelevant)
505 return atomic_read(&mm->tlb_flush_pending) > 1;
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().
514 static inline pte_marker copy_pte_marker(
515 swp_entry_t entry, struct vm_area_struct *dst_vma)
517 pte_marker srcm = pte_marker_get(entry);
518 /* Always copy error entries. */
519 pte_marker dstm = srcm & PTE_MARKER_POISONED;
521 /* Only copy PTE markers if UFFD register matches. */
522 if ((srcm & PTE_MARKER_UFFD_WP) && userfaultfd_wp(dst_vma))
523 dstm |= PTE_MARKER_UFFD_WP;
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.
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.
539 * This function is a no-op if PTE_MARKER_UFFD_WP is not enabled.
542 pte_install_uffd_wp_if_needed(struct vm_area_struct *vma, unsigned long addr,
543 pte_t *pte, pte_t pteval)
545 #ifdef CONFIG_PTE_MARKER_UFFD_WP
546 bool arm_uffd_pte = false;
548 /* The current status of the pte should be "cleared" before calling */
549 WARN_ON_ONCE(!pte_none(ptep_get(pte)));
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.
557 if (vma_is_anonymous(vma) || !userfaultfd_wp(vma))
560 /* A uffd-wp wr-protected normal pte */
561 if (unlikely(pte_present(pteval) && pte_uffd_wp(pteval)))
565 * A uffd-wp wr-protected swap pte. Note: this should even cover an
566 * existing pte marker with uffd-wp bit set.
568 if (unlikely(pte_swp_uffd_wp_any(pteval)))
571 if (unlikely(arm_uffd_pte))
572 set_pte_at(vma->vm_mm, addr, pte,
573 make_pte_marker(PTE_MARKER_UFFD_WP));
577 static inline bool vma_has_recency(struct vm_area_struct *vma)
579 if (vma->vm_flags & (VM_SEQ_READ | VM_RAND_READ))
582 if (vma->vm_file && (vma->vm_file->f_mode & FMODE_NOREUSE))