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/swap.h>
8 #include <linux/string.h>
11 * folio_is_file_lru - Should the folio be on a file LRU or anon LRU?
12 * @folio: The folio to test.
14 * We would like to get this info without a page flag, but the state
15 * needs to survive until the folio is last deleted from the LRU, which
16 * could be as far down as __page_cache_release.
18 * Return: An integer (not a boolean!) used to sort a folio onto the
19 * right LRU list and to account folios correctly.
20 * 1 if @folio is a regular filesystem backed page cache folio
21 * or a lazily freed anonymous folio (e.g. via MADV_FREE).
22 * 0 if @folio is a normal anonymous folio, a tmpfs folio or otherwise
23 * ram or swap backed folio.
25 static inline int folio_is_file_lru(struct folio *folio)
27 return !folio_test_swapbacked(folio);
30 static inline int page_is_file_lru(struct page *page)
32 return folio_is_file_lru(page_folio(page));
35 static __always_inline void update_lru_size(struct lruvec *lruvec,
36 enum lru_list lru, enum zone_type zid,
39 struct pglist_data *pgdat = lruvec_pgdat(lruvec);
41 __mod_lruvec_state(lruvec, NR_LRU_BASE + lru, nr_pages);
42 __mod_zone_page_state(&pgdat->node_zones[zid],
43 NR_ZONE_LRU_BASE + lru, nr_pages);
45 mem_cgroup_update_lru_size(lruvec, lru, zid, nr_pages);
50 * __folio_clear_lru_flags - Clear page lru flags before releasing a page.
51 * @folio: The folio that was on lru and now has a zero reference.
53 static __always_inline void __folio_clear_lru_flags(struct folio *folio)
55 VM_BUG_ON_FOLIO(!folio_test_lru(folio), folio);
57 __folio_clear_lru(folio);
59 /* this shouldn't happen, so leave the flags to bad_page() */
60 if (folio_test_active(folio) && folio_test_unevictable(folio))
63 __folio_clear_active(folio);
64 __folio_clear_unevictable(folio);
67 static __always_inline void __clear_page_lru_flags(struct page *page)
69 __folio_clear_lru_flags(page_folio(page));
73 * folio_lru_list - Which LRU list should a folio be on?
74 * @folio: The folio to test.
76 * Return: The LRU list a folio should be on, as an index
77 * into the array of LRU lists.
79 static __always_inline enum lru_list folio_lru_list(struct folio *folio)
83 VM_BUG_ON_FOLIO(folio_test_active(folio) && folio_test_unevictable(folio), folio);
85 if (folio_test_unevictable(folio))
86 return LRU_UNEVICTABLE;
88 lru = folio_is_file_lru(folio) ? LRU_INACTIVE_FILE : LRU_INACTIVE_ANON;
89 if (folio_test_active(folio))
95 static __always_inline
96 void lruvec_add_folio(struct lruvec *lruvec, struct folio *folio)
98 enum lru_list lru = folio_lru_list(folio);
100 update_lru_size(lruvec, lru, folio_zonenum(folio),
101 folio_nr_pages(folio));
102 list_add(&folio->lru, &lruvec->lists[lru]);
105 static __always_inline void add_page_to_lru_list(struct page *page,
106 struct lruvec *lruvec)
108 lruvec_add_folio(lruvec, page_folio(page));
111 static __always_inline
112 void lruvec_add_folio_tail(struct lruvec *lruvec, struct folio *folio)
114 enum lru_list lru = folio_lru_list(folio);
116 update_lru_size(lruvec, lru, folio_zonenum(folio),
117 folio_nr_pages(folio));
118 list_add_tail(&folio->lru, &lruvec->lists[lru]);
121 static __always_inline void add_page_to_lru_list_tail(struct page *page,
122 struct lruvec *lruvec)
124 lruvec_add_folio_tail(lruvec, page_folio(page));
127 static __always_inline
128 void lruvec_del_folio(struct lruvec *lruvec, struct folio *folio)
130 list_del(&folio->lru);
131 update_lru_size(lruvec, folio_lru_list(folio), folio_zonenum(folio),
132 -folio_nr_pages(folio));
135 static __always_inline void del_page_from_lru_list(struct page *page,
136 struct lruvec *lruvec)
138 lruvec_del_folio(lruvec, page_folio(page));
141 #ifdef CONFIG_ANON_VMA_NAME
143 * mmap_lock should be read-locked when calling anon_vma_name(). Caller should
144 * either keep holding the lock while using the returned pointer or it should
145 * raise anon_vma_name refcount before releasing the lock.
147 extern struct anon_vma_name *anon_vma_name(struct vm_area_struct *vma);
148 extern struct anon_vma_name *anon_vma_name_alloc(const char *name);
149 extern void anon_vma_name_free(struct kref *kref);
151 /* mmap_lock should be read-locked */
152 static inline void anon_vma_name_get(struct anon_vma_name *anon_name)
155 kref_get(&anon_name->kref);
158 static inline void anon_vma_name_put(struct anon_vma_name *anon_name)
161 kref_put(&anon_name->kref, anon_vma_name_free);
165 struct anon_vma_name *anon_vma_name_reuse(struct anon_vma_name *anon_name)
167 /* Prevent anon_name refcount saturation early on */
168 if (kref_read(&anon_name->kref) < REFCOUNT_MAX) {
169 anon_vma_name_get(anon_name);
173 return anon_vma_name_alloc(anon_name->name);
176 static inline void dup_anon_vma_name(struct vm_area_struct *orig_vma,
177 struct vm_area_struct *new_vma)
179 struct anon_vma_name *anon_name = anon_vma_name(orig_vma);
182 new_vma->anon_name = anon_vma_name_reuse(anon_name);
185 static inline void free_anon_vma_name(struct vm_area_struct *vma)
188 * Not using anon_vma_name because it generates a warning if mmap_lock
189 * is not held, which might be the case here.
192 anon_vma_name_put(vma->anon_name);
195 static inline bool anon_vma_name_eq(struct anon_vma_name *anon_name1,
196 struct anon_vma_name *anon_name2)
198 if (anon_name1 == anon_name2)
201 return anon_name1 && anon_name2 &&
202 !strcmp(anon_name1->name, anon_name2->name);
205 #else /* CONFIG_ANON_VMA_NAME */
206 static inline struct anon_vma_name *anon_vma_name(struct vm_area_struct *vma)
211 static inline struct anon_vma_name *anon_vma_name_alloc(const char *name)
216 static inline void anon_vma_name_get(struct anon_vma_name *anon_name) {}
217 static inline void anon_vma_name_put(struct anon_vma_name *anon_name) {}
218 static inline void dup_anon_vma_name(struct vm_area_struct *orig_vma,
219 struct vm_area_struct *new_vma) {}
220 static inline void free_anon_vma_name(struct vm_area_struct *vma) {}
222 static inline bool anon_vma_name_eq(struct anon_vma_name *anon_name1,
223 struct anon_vma_name *anon_name2)
228 #endif /* CONFIG_ANON_VMA_NAME */
230 static inline void init_tlb_flush_pending(struct mm_struct *mm)
232 atomic_set(&mm->tlb_flush_pending, 0);
235 static inline void inc_tlb_flush_pending(struct mm_struct *mm)
237 atomic_inc(&mm->tlb_flush_pending);
239 * The only time this value is relevant is when there are indeed pages
240 * to flush. And we'll only flush pages after changing them, which
243 * So the ordering here is:
245 * atomic_inc(&mm->tlb_flush_pending);
252 * mm_tlb_flush_pending();
257 * atomic_dec(&mm->tlb_flush_pending);
259 * Where the increment if constrained by the PTL unlock, it thus
260 * ensures that the increment is visible if the PTE modification is
261 * visible. After all, if there is no PTE modification, nobody cares
262 * about TLB flushes either.
264 * This very much relies on users (mm_tlb_flush_pending() and
265 * mm_tlb_flush_nested()) only caring about _specific_ PTEs (and
266 * therefore specific PTLs), because with SPLIT_PTE_PTLOCKS and RCpc
267 * locks (PPC) the unlock of one doesn't order against the lock of
270 * The decrement is ordered by the flush_tlb_range(), such that
271 * mm_tlb_flush_pending() will not return false unless all flushes have
276 static inline void dec_tlb_flush_pending(struct mm_struct *mm)
279 * See inc_tlb_flush_pending().
281 * This cannot be smp_mb__before_atomic() because smp_mb() simply does
282 * not order against TLB invalidate completion, which is what we need.
284 * Therefore we must rely on tlb_flush_*() to guarantee order.
286 atomic_dec(&mm->tlb_flush_pending);
289 static inline bool mm_tlb_flush_pending(struct mm_struct *mm)
292 * Must be called after having acquired the PTL; orders against that
293 * PTLs release and therefore ensures that if we observe the modified
294 * PTE we must also observe the increment from inc_tlb_flush_pending().
296 * That is, it only guarantees to return true if there is a flush
297 * pending for _this_ PTL.
299 return atomic_read(&mm->tlb_flush_pending);
302 static inline bool mm_tlb_flush_nested(struct mm_struct *mm)
305 * Similar to mm_tlb_flush_pending(), we must have acquired the PTL
306 * for which there is a TLB flush pending in order to guarantee
307 * we've seen both that PTE modification and the increment.
309 * (no requirement on actually still holding the PTL, that is irrelevant)
311 return atomic_read(&mm->tlb_flush_pending) > 1;