Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel
os
linux
1/* SPDX-License-Identifier: GPL-2.0 */
2#ifndef _LINUX_HUGETLB_H
3#define _LINUX_HUGETLB_H
4
5#include <linux/mm.h>
6#include <linux/mm_types.h>
7#include <linux/mmdebug.h>
8#include <linux/fs.h>
9#include <linux/hugetlb_inline.h>
10#include <linux/cgroup.h>
11#include <linux/page_ref.h>
12#include <linux/list.h>
13#include <linux/kref.h>
14#include <linux/pgtable.h>
15#include <linux/gfp.h>
16#include <linux/userfaultfd_k.h>
17#include <linux/nodemask.h>
18
19struct mmu_gather;
20struct node;
21
22void free_huge_folio(struct folio *folio);
23
24#ifdef CONFIG_HUGETLB_PAGE
25
26#include <linux/pagemap.h>
27#include <linux/shm.h>
28#include <asm/tlbflush.h>
29
30/*
31 * For HugeTLB page, there are more metadata to save in the struct page. But
32 * the head struct page cannot meet our needs, so we have to abuse other tail
33 * struct page to store the metadata.
34 */
35#define __NR_USED_SUBPAGE 3
36
37struct hugepage_subpool {
38 spinlock_t lock;
39 long count;
40 long max_hpages; /* Maximum huge pages or -1 if no maximum. */
41 long used_hpages; /* Used count against maximum, includes */
42 /* both allocated and reserved pages. */
43 struct hstate *hstate;
44 long min_hpages; /* Minimum huge pages or -1 if no minimum. */
45 long rsv_hpages; /* Pages reserved against global pool to */
46 /* satisfy minimum size. */
47};
48
49struct resv_map {
50 struct kref refs;
51 spinlock_t lock;
52 struct list_head regions;
53 long adds_in_progress;
54 struct list_head region_cache;
55 long region_cache_count;
56 struct rw_semaphore rw_sema;
57#ifdef CONFIG_CGROUP_HUGETLB
58 /*
59 * On private mappings, the counter to uncharge reservations is stored
60 * here. If these fields are 0, then either the mapping is shared, or
61 * cgroup accounting is disabled for this resv_map.
62 */
63 struct page_counter *reservation_counter;
64 unsigned long pages_per_hpage;
65 struct cgroup_subsys_state *css;
66#endif
67};
68
69/*
70 * Region tracking -- allows tracking of reservations and instantiated pages
71 * across the pages in a mapping.
72 *
73 * The region data structures are embedded into a resv_map and protected
74 * by a resv_map's lock. The set of regions within the resv_map represent
75 * reservations for huge pages, or huge pages that have already been
76 * instantiated within the map. The from and to elements are huge page
77 * indices into the associated mapping. from indicates the starting index
78 * of the region. to represents the first index past the end of the region.
79 *
80 * For example, a file region structure with from == 0 and to == 4 represents
81 * four huge pages in a mapping. It is important to note that the to element
82 * represents the first element past the end of the region. This is used in
83 * arithmetic as 4(to) - 0(from) = 4 huge pages in the region.
84 *
85 * Interval notation of the form [from, to) will be used to indicate that
86 * the endpoint from is inclusive and to is exclusive.
87 */
88struct file_region {
89 struct list_head link;
90 long from;
91 long to;
92#ifdef CONFIG_CGROUP_HUGETLB
93 /*
94 * On shared mappings, each reserved region appears as a struct
95 * file_region in resv_map. These fields hold the info needed to
96 * uncharge each reservation.
97 */
98 struct page_counter *reservation_counter;
99 struct cgroup_subsys_state *css;
100#endif
101};
102
103struct hugetlb_vma_lock {
104 struct kref refs;
105 struct rw_semaphore rw_sema;
106 struct vm_area_struct *vma;
107};
108
109extern struct resv_map *resv_map_alloc(void);
110void resv_map_release(struct kref *ref);
111
112extern spinlock_t hugetlb_lock;
113extern int hugetlb_max_hstate __read_mostly;
114#define for_each_hstate(h) \
115 for ((h) = hstates; (h) < &hstates[hugetlb_max_hstate]; (h)++)
116
117struct hugepage_subpool *hugepage_new_subpool(struct hstate *h, long max_hpages,
118 long min_hpages);
119void hugepage_put_subpool(struct hugepage_subpool *spool);
120
121void hugetlb_dup_vma_private(struct vm_area_struct *vma);
122void clear_vma_resv_huge_pages(struct vm_area_struct *vma);
123int move_hugetlb_page_tables(struct vm_area_struct *vma,
124 struct vm_area_struct *new_vma,
125 unsigned long old_addr, unsigned long new_addr,
126 unsigned long len);
127int copy_hugetlb_page_range(struct mm_struct *, struct mm_struct *,
128 struct vm_area_struct *, struct vm_area_struct *);
129void unmap_hugepage_range(struct vm_area_struct *,
130 unsigned long start, unsigned long end,
131 struct folio *, zap_flags_t);
132void __unmap_hugepage_range(struct mmu_gather *tlb,
133 struct vm_area_struct *vma,
134 unsigned long start, unsigned long end,
135 struct folio *, zap_flags_t zap_flags);
136void hugetlb_report_meminfo(struct seq_file *);
137int hugetlb_report_node_meminfo(char *buf, int len, int nid);
138void hugetlb_show_meminfo_node(int nid);
139unsigned long hugetlb_total_pages(void);
140vm_fault_t hugetlb_fault(struct mm_struct *mm, struct vm_area_struct *vma,
141 unsigned long address, unsigned int flags);
142#ifdef CONFIG_USERFAULTFD
143int hugetlb_mfill_atomic_pte(pte_t *dst_pte,
144 struct vm_area_struct *dst_vma,
145 unsigned long dst_addr,
146 unsigned long src_addr,
147 uffd_flags_t flags,
148 struct folio **foliop);
149#endif /* CONFIG_USERFAULTFD */
150long hugetlb_reserve_pages(struct inode *inode, long from, long to,
151 struct vm_area_desc *desc, vma_flags_t vma_flags);
152long hugetlb_unreserve_pages(struct inode *inode, long start, long end,
153 long freed);
154bool folio_isolate_hugetlb(struct folio *folio, struct list_head *list);
155int get_hwpoison_hugetlb_folio(struct folio *folio, bool *hugetlb, bool unpoison);
156int get_huge_page_for_hwpoison(unsigned long pfn, int flags,
157 bool *migratable_cleared);
158void folio_putback_hugetlb(struct folio *folio);
159void move_hugetlb_state(struct folio *old_folio, struct folio *new_folio, int reason);
160void hugetlb_fix_reserve_counts(struct inode *inode);
161extern struct mutex *hugetlb_fault_mutex_table;
162u32 hugetlb_fault_mutex_hash(struct address_space *mapping, pgoff_t idx);
163
164pte_t *huge_pmd_share(struct mm_struct *mm, struct vm_area_struct *vma,
165 unsigned long addr, pud_t *pud);
166bool hugetlbfs_pagecache_present(struct hstate *h,
167 struct vm_area_struct *vma,
168 unsigned long address);
169
170struct address_space *hugetlb_folio_mapping_lock_write(struct folio *folio);
171
172extern int movable_gigantic_pages __read_mostly;
173extern int sysctl_hugetlb_shm_group __read_mostly;
174extern struct list_head huge_boot_pages[MAX_NUMNODES];
175
176void hugetlb_bootmem_alloc(void);
177extern nodemask_t hugetlb_bootmem_nodes;
178void hugetlb_bootmem_set_nodes(void);
179
180/* arch callbacks */
181
182#ifndef CONFIG_HIGHPTE
183/*
184 * pte_offset_huge() and pte_alloc_huge() are helpers for those architectures
185 * which may go down to the lowest PTE level in their huge_pte_offset() and
186 * huge_pte_alloc(): to avoid reliance on pte_offset_map() without pte_unmap().
187 */
188static inline pte_t *pte_offset_huge(pmd_t *pmd, unsigned long address)
189{
190 return pte_offset_kernel(pmd, address);
191}
192static inline pte_t *pte_alloc_huge(struct mm_struct *mm, pmd_t *pmd,
193 unsigned long address)
194{
195 return pte_alloc(mm, pmd) ? NULL : pte_offset_huge(pmd, address);
196}
197#endif
198
199pte_t *huge_pte_alloc(struct mm_struct *mm, struct vm_area_struct *vma,
200 unsigned long addr, unsigned long sz);
201/*
202 * huge_pte_offset(): Walk the hugetlb pgtable until the last level PTE.
203 * Returns the pte_t* if found, or NULL if the address is not mapped.
204 *
205 * IMPORTANT: we should normally not directly call this function, instead
206 * this is only a common interface to implement arch-specific
207 * walker. Please use hugetlb_walk() instead, because that will attempt to
208 * verify the locking for you.
209 *
210 * Since this function will walk all the pgtable pages (including not only
211 * high-level pgtable page, but also PUD entry that can be unshared
212 * concurrently for VM_SHARED), the caller of this function should be
213 * responsible of its thread safety. One can follow this rule:
214 *
215 * (1) For private mappings: pmd unsharing is not possible, so holding the
216 * mmap_lock for either read or write is sufficient. Most callers
217 * already hold the mmap_lock, so normally, no special action is
218 * required.
219 *
220 * (2) For shared mappings: pmd unsharing is possible (so the PUD-ranged
221 * pgtable page can go away from under us! It can be done by a pmd
222 * unshare with a follow up munmap() on the other process), then we
223 * need either:
224 *
225 * (2.1) hugetlb vma lock read or write held, to make sure pmd unshare
226 * won't happen upon the range (it also makes sure the pte_t we
227 * read is the right and stable one), or,
228 *
229 * (2.2) hugetlb mapping i_mmap_rwsem lock held read or write, to make
230 * sure even if unshare happened the racy unmap() will wait until
231 * i_mmap_rwsem is released.
232 *
233 * Option (2.1) is the safest, which guarantees pte stability from pmd
234 * sharing pov, until the vma lock released. Option (2.2) doesn't protect
235 * a concurrent pmd unshare, but it makes sure the pgtable page is safe to
236 * access.
237 */
238pte_t *huge_pte_offset(struct mm_struct *mm,
239 unsigned long addr, unsigned long sz);
240unsigned long hugetlb_mask_last_page(struct hstate *h);
241int huge_pmd_unshare(struct mmu_gather *tlb, struct vm_area_struct *vma,
242 unsigned long addr, pte_t *ptep);
243void huge_pmd_unshare_flush(struct mmu_gather *tlb, struct vm_area_struct *vma);
244void adjust_range_if_pmd_sharing_possible(struct vm_area_struct *vma,
245 unsigned long *start, unsigned long *end);
246
247extern void __hugetlb_zap_begin(struct vm_area_struct *vma,
248 unsigned long *begin, unsigned long *end);
249extern void __hugetlb_zap_end(struct vm_area_struct *vma,
250 struct zap_details *details);
251
252static inline void hugetlb_zap_begin(struct vm_area_struct *vma,
253 unsigned long *start, unsigned long *end)
254{
255 if (is_vm_hugetlb_page(vma))
256 __hugetlb_zap_begin(vma, start, end);
257}
258
259static inline void hugetlb_zap_end(struct vm_area_struct *vma,
260 struct zap_details *details)
261{
262 if (is_vm_hugetlb_page(vma))
263 __hugetlb_zap_end(vma, details);
264}
265
266void hugetlb_vma_lock_read(struct vm_area_struct *vma);
267void hugetlb_vma_unlock_read(struct vm_area_struct *vma);
268void hugetlb_vma_lock_write(struct vm_area_struct *vma);
269void hugetlb_vma_unlock_write(struct vm_area_struct *vma);
270int hugetlb_vma_trylock_write(struct vm_area_struct *vma);
271void hugetlb_vma_assert_locked(struct vm_area_struct *vma);
272void hugetlb_vma_lock_release(struct kref *kref);
273long hugetlb_change_protection(struct vm_area_struct *vma,
274 unsigned long address, unsigned long end, pgprot_t newprot,
275 unsigned long cp_flags);
276void hugetlb_unshare_all_pmds(struct vm_area_struct *vma);
277void fixup_hugetlb_reservations(struct vm_area_struct *vma);
278void hugetlb_split(struct vm_area_struct *vma, unsigned long addr);
279int hugetlb_vma_lock_alloc(struct vm_area_struct *vma);
280
281unsigned int arch_hugetlb_cma_order(void);
282
283#else /* !CONFIG_HUGETLB_PAGE */
284
285static inline void hugetlb_dup_vma_private(struct vm_area_struct *vma)
286{
287}
288
289static inline void clear_vma_resv_huge_pages(struct vm_area_struct *vma)
290{
291}
292
293static inline unsigned long hugetlb_total_pages(void)
294{
295 return 0;
296}
297
298static inline struct address_space *hugetlb_folio_mapping_lock_write(
299 struct folio *folio)
300{
301 return NULL;
302}
303
304static inline int huge_pmd_unshare(struct mmu_gather *tlb,
305 struct vm_area_struct *vma, unsigned long addr, pte_t *ptep)
306{
307 return 0;
308}
309
310static inline void huge_pmd_unshare_flush(struct mmu_gather *tlb,
311 struct vm_area_struct *vma)
312{
313}
314
315static inline void adjust_range_if_pmd_sharing_possible(
316 struct vm_area_struct *vma,
317 unsigned long *start, unsigned long *end)
318{
319}
320
321static inline void hugetlb_zap_begin(
322 struct vm_area_struct *vma,
323 unsigned long *start, unsigned long *end)
324{
325}
326
327static inline void hugetlb_zap_end(
328 struct vm_area_struct *vma,
329 struct zap_details *details)
330{
331}
332
333static inline int copy_hugetlb_page_range(struct mm_struct *dst,
334 struct mm_struct *src,
335 struct vm_area_struct *dst_vma,
336 struct vm_area_struct *src_vma)
337{
338 BUG();
339 return 0;
340}
341
342static inline int move_hugetlb_page_tables(struct vm_area_struct *vma,
343 struct vm_area_struct *new_vma,
344 unsigned long old_addr,
345 unsigned long new_addr,
346 unsigned long len)
347{
348 BUG();
349 return 0;
350}
351
352static inline void hugetlb_report_meminfo(struct seq_file *m)
353{
354}
355
356static inline int hugetlb_report_node_meminfo(char *buf, int len, int nid)
357{
358 return 0;
359}
360
361static inline void hugetlb_show_meminfo_node(int nid)
362{
363}
364
365static inline void hugetlb_vma_lock_read(struct vm_area_struct *vma)
366{
367}
368
369static inline void hugetlb_vma_unlock_read(struct vm_area_struct *vma)
370{
371}
372
373static inline void hugetlb_vma_lock_write(struct vm_area_struct *vma)
374{
375}
376
377static inline void hugetlb_vma_unlock_write(struct vm_area_struct *vma)
378{
379}
380
381static inline int hugetlb_vma_trylock_write(struct vm_area_struct *vma)
382{
383 return 1;
384}
385
386static inline void hugetlb_vma_assert_locked(struct vm_area_struct *vma)
387{
388}
389
390static inline int is_hugepage_only_range(struct mm_struct *mm,
391 unsigned long addr, unsigned long len)
392{
393 return 0;
394}
395
396#ifdef CONFIG_USERFAULTFD
397static inline int hugetlb_mfill_atomic_pte(pte_t *dst_pte,
398 struct vm_area_struct *dst_vma,
399 unsigned long dst_addr,
400 unsigned long src_addr,
401 uffd_flags_t flags,
402 struct folio **foliop)
403{
404 BUG();
405 return 0;
406}
407#endif /* CONFIG_USERFAULTFD */
408
409static inline pte_t *huge_pte_offset(struct mm_struct *mm, unsigned long addr,
410 unsigned long sz)
411{
412 return NULL;
413}
414
415static inline bool folio_isolate_hugetlb(struct folio *folio, struct list_head *list)
416{
417 return false;
418}
419
420static inline int get_hwpoison_hugetlb_folio(struct folio *folio, bool *hugetlb, bool unpoison)
421{
422 return 0;
423}
424
425static inline int get_huge_page_for_hwpoison(unsigned long pfn, int flags,
426 bool *migratable_cleared)
427{
428 return 0;
429}
430
431static inline void folio_putback_hugetlb(struct folio *folio)
432{
433}
434
435static inline void move_hugetlb_state(struct folio *old_folio,
436 struct folio *new_folio, int reason)
437{
438}
439
440static inline long hugetlb_change_protection(
441 struct vm_area_struct *vma, unsigned long address,
442 unsigned long end, pgprot_t newprot,
443 unsigned long cp_flags)
444{
445 return 0;
446}
447
448static inline void __unmap_hugepage_range(struct mmu_gather *tlb,
449 struct vm_area_struct *vma, unsigned long start,
450 unsigned long end, struct folio *folio,
451 zap_flags_t zap_flags)
452{
453 BUG();
454}
455
456static inline vm_fault_t hugetlb_fault(struct mm_struct *mm,
457 struct vm_area_struct *vma, unsigned long address,
458 unsigned int flags)
459{
460 BUG();
461 return 0;
462}
463
464static inline void hugetlb_unshare_all_pmds(struct vm_area_struct *vma) { }
465
466static inline void fixup_hugetlb_reservations(struct vm_area_struct *vma)
467{
468}
469
470static inline void hugetlb_split(struct vm_area_struct *vma, unsigned long addr) {}
471
472static inline int hugetlb_vma_lock_alloc(struct vm_area_struct *vma)
473{
474 return 0;
475}
476
477#endif /* !CONFIG_HUGETLB_PAGE */
478
479#ifndef pgd_write
480static inline int pgd_write(pgd_t pgd)
481{
482 BUG();
483 return 0;
484}
485#endif
486
487#define HUGETLB_ANON_FILE "anon_hugepage"
488
489enum {
490 /*
491 * The file will be used as an shm file so shmfs accounting rules
492 * apply
493 */
494 HUGETLB_SHMFS_INODE = 1,
495 /*
496 * The file is being created on the internal vfs mount and shmfs
497 * accounting rules do not apply
498 */
499 HUGETLB_ANONHUGE_INODE = 2,
500};
501
502#ifdef CONFIG_HUGETLBFS
503struct hugetlbfs_sb_info {
504 long max_inodes; /* inodes allowed */
505 long free_inodes; /* inodes free */
506 spinlock_t stat_lock;
507 struct hstate *hstate;
508 struct hugepage_subpool *spool;
509 kuid_t uid;
510 kgid_t gid;
511 umode_t mode;
512};
513
514static inline struct hugetlbfs_sb_info *HUGETLBFS_SB(struct super_block *sb)
515{
516 return sb->s_fs_info;
517}
518
519struct hugetlbfs_inode_info {
520 struct inode vfs_inode;
521 struct resv_map *resv_map;
522 unsigned int seals;
523};
524
525static inline struct hugetlbfs_inode_info *HUGETLBFS_I(struct inode *inode)
526{
527 return container_of(inode, struct hugetlbfs_inode_info, vfs_inode);
528}
529
530extern const struct vm_operations_struct hugetlb_vm_ops;
531struct file *hugetlb_file_setup(const char *name, size_t size, vma_flags_t acct,
532 int creat_flags, int page_size_log);
533
534static inline bool is_file_hugepages(const struct file *file)
535{
536 return file->f_op->fop_flags & FOP_HUGE_PAGES;
537}
538
539static inline struct hstate *hstate_inode(struct inode *i)
540{
541 return HUGETLBFS_SB(i->i_sb)->hstate;
542}
543#else /* !CONFIG_HUGETLBFS */
544
545#define is_file_hugepages(file) false
546static inline struct file *
547hugetlb_file_setup(const char *name, size_t size, vma_flags_t acctflag,
548 int creat_flags, int page_size_log)
549{
550 return ERR_PTR(-ENOSYS);
551}
552
553static inline struct hstate *hstate_inode(struct inode *i)
554{
555 return NULL;
556}
557#endif /* !CONFIG_HUGETLBFS */
558
559unsigned long
560hugetlb_get_unmapped_area(struct file *file, unsigned long addr,
561 unsigned long len, unsigned long pgoff,
562 unsigned long flags);
563
564/*
565 * huegtlb page specific state flags. These flags are located in page.private
566 * of the hugetlb head page. Functions created via the below macros should be
567 * used to manipulate these flags.
568 *
569 * HPG_restore_reserve - Set when a hugetlb page consumes a reservation at
570 * allocation time. Cleared when page is fully instantiated. Free
571 * routine checks flag to restore a reservation on error paths.
572 * Synchronization: Examined or modified by code that knows it has
573 * the only reference to page. i.e. After allocation but before use
574 * or when the page is being freed.
575 * HPG_migratable - Set after a newly allocated page is added to the page
576 * cache and/or page tables. Indicates the page is a candidate for
577 * migration.
578 * Synchronization: Initially set after new page allocation with no
579 * locking. When examined and modified during migration processing
580 * (isolate, migrate, putback) the hugetlb_lock is held.
581 * HPG_temporary - Set on a page that is temporarily allocated from the buddy
582 * allocator. Typically used for migration target pages when no pages
583 * are available in the pool. The hugetlb free page path will
584 * immediately free pages with this flag set to the buddy allocator.
585 * Synchronization: Can be set after huge page allocation from buddy when
586 * code knows it has only reference. All other examinations and
587 * modifications require hugetlb_lock.
588 * HPG_freed - Set when page is on the free lists.
589 * Synchronization: hugetlb_lock held for examination and modification.
590 * HPG_vmemmap_optimized - Set when the vmemmap pages of the page are freed.
591 * HPG_raw_hwp_unreliable - Set when the hugetlb page has a hwpoison sub-page
592 * that is not tracked by raw_hwp_page list.
593 */
594enum hugetlb_page_flags {
595 HPG_restore_reserve = 0,
596 HPG_migratable,
597 HPG_temporary,
598 HPG_freed,
599 HPG_vmemmap_optimized,
600 HPG_raw_hwp_unreliable,
601 HPG_cma,
602 __NR_HPAGEFLAGS,
603};
604
605/*
606 * Macros to create test, set and clear function definitions for
607 * hugetlb specific page flags.
608 */
609#ifdef CONFIG_HUGETLB_PAGE
610#define TESTHPAGEFLAG(uname, flname) \
611static __always_inline \
612bool folio_test_hugetlb_##flname(struct folio *folio) \
613 { void *private = &folio->private; \
614 return test_bit(HPG_##flname, private); \
615 }
616
617#define SETHPAGEFLAG(uname, flname) \
618static __always_inline \
619void folio_set_hugetlb_##flname(struct folio *folio) \
620 { void *private = &folio->private; \
621 set_bit(HPG_##flname, private); \
622 }
623
624#define CLEARHPAGEFLAG(uname, flname) \
625static __always_inline \
626void folio_clear_hugetlb_##flname(struct folio *folio) \
627 { void *private = &folio->private; \
628 clear_bit(HPG_##flname, private); \
629 }
630#else
631#define TESTHPAGEFLAG(uname, flname) \
632static inline bool \
633folio_test_hugetlb_##flname(struct folio *folio) \
634 { return 0; }
635
636#define SETHPAGEFLAG(uname, flname) \
637static inline void \
638folio_set_hugetlb_##flname(struct folio *folio) \
639 { }
640
641#define CLEARHPAGEFLAG(uname, flname) \
642static inline void \
643folio_clear_hugetlb_##flname(struct folio *folio) \
644 { }
645#endif
646
647#define HPAGEFLAG(uname, flname) \
648 TESTHPAGEFLAG(uname, flname) \
649 SETHPAGEFLAG(uname, flname) \
650 CLEARHPAGEFLAG(uname, flname) \
651
652/*
653 * Create functions associated with hugetlb page flags
654 */
655HPAGEFLAG(RestoreReserve, restore_reserve)
656HPAGEFLAG(Migratable, migratable)
657HPAGEFLAG(Temporary, temporary)
658HPAGEFLAG(Freed, freed)
659HPAGEFLAG(VmemmapOptimized, vmemmap_optimized)
660HPAGEFLAG(RawHwpUnreliable, raw_hwp_unreliable)
661HPAGEFLAG(Cma, cma)
662
663#ifdef CONFIG_HUGETLB_PAGE
664
665#define HSTATE_NAME_LEN 32
666/* Defines one hugetlb page size */
667struct hstate {
668 struct mutex resize_lock;
669 struct lock_class_key resize_key;
670 int next_nid_to_alloc;
671 int next_nid_to_free;
672 unsigned int order;
673 unsigned int demote_order;
674 unsigned long mask;
675 unsigned long max_huge_pages;
676 unsigned long nr_huge_pages;
677 unsigned long free_huge_pages;
678 unsigned long resv_huge_pages;
679 unsigned long surplus_huge_pages;
680 unsigned long nr_overcommit_huge_pages;
681 struct list_head hugepage_activelist;
682 struct list_head hugepage_freelists[MAX_NUMNODES];
683 unsigned int max_huge_pages_node[MAX_NUMNODES];
684 unsigned int nr_huge_pages_node[MAX_NUMNODES];
685 unsigned int free_huge_pages_node[MAX_NUMNODES];
686 unsigned int surplus_huge_pages_node[MAX_NUMNODES];
687 char name[HSTATE_NAME_LEN];
688};
689
690struct cma;
691
692struct huge_bootmem_page {
693 struct list_head list;
694 struct hstate *hstate;
695 unsigned long flags;
696 struct cma *cma;
697};
698
699#define HUGE_BOOTMEM_HVO 0x0001
700#define HUGE_BOOTMEM_ZONES_VALID 0x0002
701#define HUGE_BOOTMEM_CMA 0x0004
702
703bool hugetlb_bootmem_page_zones_valid(int nid, struct huge_bootmem_page *m);
704
705int isolate_or_dissolve_huge_folio(struct folio *folio, struct list_head *list);
706int replace_free_hugepage_folios(unsigned long start_pfn, unsigned long end_pfn);
707void wait_for_freed_hugetlb_folios(void);
708struct folio *alloc_hugetlb_folio(struct vm_area_struct *vma,
709 unsigned long addr, bool cow_from_owner);
710struct folio *alloc_hugetlb_folio_nodemask(struct hstate *h, int preferred_nid,
711 nodemask_t *nmask, gfp_t gfp_mask,
712 bool allow_alloc_fallback);
713struct folio *alloc_hugetlb_folio_reserve(struct hstate *h, int preferred_nid,
714 nodemask_t *nmask, gfp_t gfp_mask);
715
716int hugetlb_add_to_page_cache(struct folio *folio, struct address_space *mapping,
717 pgoff_t idx);
718void restore_reserve_on_error(struct hstate *h, struct vm_area_struct *vma,
719 unsigned long address, struct folio *folio);
720
721/* arch callback */
722int __init __alloc_bootmem_huge_page(struct hstate *h, int nid);
723int __init alloc_bootmem_huge_page(struct hstate *h, int nid);
724bool __init hugetlb_node_alloc_supported(void);
725
726void __init hugetlb_add_hstate(unsigned order);
727bool __init arch_hugetlb_valid_size(unsigned long size);
728struct hstate *size_to_hstate(unsigned long size);
729
730#ifndef HUGE_MAX_HSTATE
731#define HUGE_MAX_HSTATE 1
732#endif
733
734extern struct hstate hstates[HUGE_MAX_HSTATE];
735extern unsigned int default_hstate_idx;
736
737#define default_hstate (hstates[default_hstate_idx])
738
739static inline struct hugepage_subpool *subpool_inode(struct inode *inode)
740{
741 return HUGETLBFS_SB(inode->i_sb)->spool;
742}
743
744static inline struct hugepage_subpool *hugetlb_folio_subpool(struct folio *folio)
745{
746 return folio->_hugetlb_subpool;
747}
748
749static inline void hugetlb_set_folio_subpool(struct folio *folio,
750 struct hugepage_subpool *subpool)
751{
752 folio->_hugetlb_subpool = subpool;
753}
754
755static inline struct hstate *hstate_file(struct file *f)
756{
757 return hstate_inode(file_inode(f));
758}
759
760static inline struct hstate *hstate_sizelog(int page_size_log)
761{
762 if (!page_size_log)
763 return &default_hstate;
764
765 if (page_size_log < BITS_PER_LONG)
766 return size_to_hstate(1UL << page_size_log);
767
768 return NULL;
769}
770
771static inline struct hstate *hstate_vma(struct vm_area_struct *vma)
772{
773 return hstate_file(vma->vm_file);
774}
775
776static inline unsigned long huge_page_size(const struct hstate *h)
777{
778 return (unsigned long)PAGE_SIZE << h->order;
779}
780
781static inline unsigned long huge_page_mask(struct hstate *h)
782{
783 return h->mask;
784}
785
786static inline unsigned int huge_page_order(struct hstate *h)
787{
788 return h->order;
789}
790
791static inline unsigned huge_page_shift(struct hstate *h)
792{
793 return h->order + PAGE_SHIFT;
794}
795
796/**
797 * hugetlb_linear_page_index() - linear_page_index() but in hugetlb
798 * page size granularity.
799 * @vma: the hugetlb VMA
800 * @address: the virtual address within the VMA
801 *
802 * Return: the page offset within the mapping in huge page units.
803 */
804static inline pgoff_t hugetlb_linear_page_index(struct vm_area_struct *vma,
805 unsigned long address)
806{
807 struct hstate *h = hstate_vma(vma);
808
809 return ((address - vma->vm_start) >> huge_page_shift(h)) +
810 (vma->vm_pgoff >> huge_page_order(h));
811}
812
813static inline bool order_is_gigantic(unsigned int order)
814{
815 return order > MAX_PAGE_ORDER;
816}
817
818static inline bool hstate_is_gigantic(struct hstate *h)
819{
820 return order_is_gigantic(huge_page_order(h));
821}
822
823static inline unsigned int pages_per_huge_page(const struct hstate *h)
824{
825 return 1 << h->order;
826}
827
828static inline unsigned int blocks_per_huge_page(struct hstate *h)
829{
830 return huge_page_size(h) / 512;
831}
832
833static inline struct folio *filemap_lock_hugetlb_folio(struct hstate *h,
834 struct address_space *mapping, pgoff_t idx)
835{
836 return filemap_lock_folio(mapping, idx << huge_page_order(h));
837}
838
839#include <asm/hugetlb.h>
840
841#ifndef is_hugepage_only_range
842static inline int is_hugepage_only_range(struct mm_struct *mm,
843 unsigned long addr, unsigned long len)
844{
845 return 0;
846}
847#define is_hugepage_only_range is_hugepage_only_range
848#endif
849
850#ifndef arch_clear_hugetlb_flags
851static inline void arch_clear_hugetlb_flags(struct folio *folio) { }
852#define arch_clear_hugetlb_flags arch_clear_hugetlb_flags
853#endif
854
855#ifndef arch_make_huge_pte
856static inline pte_t arch_make_huge_pte(pte_t entry, unsigned int shift,
857 vm_flags_t flags)
858{
859 return pte_mkhuge(entry);
860}
861#endif
862
863#ifndef arch_has_huge_bootmem_alloc
864/*
865 * Some architectures do their own bootmem allocation, so they can't use
866 * early CMA allocation.
867 */
868static inline bool arch_has_huge_bootmem_alloc(void)
869{
870 return false;
871}
872#endif
873
874static inline struct hstate *folio_hstate(struct folio *folio)
875{
876 VM_BUG_ON_FOLIO(!folio_test_hugetlb(folio), folio);
877 return size_to_hstate(folio_size(folio));
878}
879
880static inline unsigned hstate_index_to_shift(unsigned index)
881{
882 return hstates[index].order + PAGE_SHIFT;
883}
884
885static inline int hstate_index(struct hstate *h)
886{
887 return h - hstates;
888}
889
890int dissolve_free_hugetlb_folio(struct folio *folio);
891int dissolve_free_hugetlb_folios(unsigned long start_pfn,
892 unsigned long end_pfn);
893
894#ifdef CONFIG_MEMORY_FAILURE
895extern void folio_clear_hugetlb_hwpoison(struct folio *folio);
896#else
897static inline void folio_clear_hugetlb_hwpoison(struct folio *folio)
898{
899}
900#endif
901
902#ifdef CONFIG_ARCH_ENABLE_HUGEPAGE_MIGRATION
903#ifndef arch_hugetlb_migration_supported
904static inline bool arch_hugetlb_migration_supported(struct hstate *h)
905{
906 if ((huge_page_shift(h) == PMD_SHIFT) ||
907 (huge_page_shift(h) == PUD_SHIFT) ||
908 (huge_page_shift(h) == PGDIR_SHIFT))
909 return true;
910 else
911 return false;
912}
913#endif
914#else
915static inline bool arch_hugetlb_migration_supported(struct hstate *h)
916{
917 return false;
918}
919#endif
920
921static inline bool hugepage_migration_supported(struct hstate *h)
922{
923 return arch_hugetlb_migration_supported(h);
924}
925
926/*
927 * Movability check is different as compared to migration check.
928 * It determines whether or not a huge page should be placed on
929 * movable zone or not. Movability of any huge page should be
930 * required only if huge page size is supported for migration.
931 * There won't be any reason for the huge page to be movable if
932 * it is not migratable to start with. Also the size of the huge
933 * page should be large enough to be placed under a movable zone
934 * and still feasible enough to be migratable. Just the presence
935 * in movable zone does not make the migration feasible.
936 *
937 * So even though large huge page sizes like the gigantic ones
938 * are migratable they should not be movable because its not
939 * feasible to migrate them from movable zone.
940 */
941static inline bool hugepage_movable_supported(struct hstate *h)
942{
943 if (!hugepage_migration_supported(h))
944 return false;
945
946 if (hstate_is_gigantic(h) && !movable_gigantic_pages)
947 return false;
948 return true;
949}
950
951/* Movability of hugepages depends on migration support. */
952static inline gfp_t htlb_alloc_mask(struct hstate *h)
953{
954 gfp_t gfp = __GFP_COMP | __GFP_NOWARN;
955
956 gfp |= hugepage_movable_supported(h) ? GFP_HIGHUSER_MOVABLE : GFP_HIGHUSER;
957
958 return gfp;
959}
960
961static inline gfp_t htlb_modify_alloc_mask(struct hstate *h, gfp_t gfp_mask)
962{
963 gfp_t modified_mask = htlb_alloc_mask(h);
964
965 /* Some callers might want to enforce node */
966 modified_mask |= (gfp_mask & __GFP_THISNODE);
967
968 modified_mask |= (gfp_mask & __GFP_NOWARN);
969
970 return modified_mask;
971}
972
973static inline bool htlb_allow_alloc_fallback(int reason)
974{
975 bool allowed_fallback = false;
976
977 /*
978 * Note: the memory offline, memory failure and migration syscalls will
979 * be allowed to fallback to other nodes due to lack of a better chioce,
980 * that might break the per-node hugetlb pool. While other cases will
981 * set the __GFP_THISNODE to avoid breaking the per-node hugetlb pool.
982 */
983 switch (reason) {
984 case MR_MEMORY_HOTPLUG:
985 case MR_MEMORY_FAILURE:
986 case MR_SYSCALL:
987 case MR_MEMPOLICY_MBIND:
988 allowed_fallback = true;
989 break;
990 default:
991 break;
992 }
993
994 return allowed_fallback;
995}
996
997static inline spinlock_t *huge_pte_lockptr(struct hstate *h,
998 struct mm_struct *mm, pte_t *pte)
999{
1000 const unsigned long size = huge_page_size(h);
1001
1002 VM_WARN_ON(size == PAGE_SIZE);
1003
1004 /*
1005 * hugetlb must use the exact same PT locks as core-mm page table
1006 * walkers would. When modifying a PTE table, hugetlb must take the
1007 * PTE PT lock, when modifying a PMD table, hugetlb must take the PMD
1008 * PT lock etc.
1009 *
1010 * The expectation is that any hugetlb folio smaller than a PMD is
1011 * always mapped into a single PTE table and that any hugetlb folio
1012 * smaller than a PUD (but at least as big as a PMD) is always mapped
1013 * into a single PMD table.
1014 *
1015 * If that does not hold for an architecture, then that architecture
1016 * must disable split PT locks such that all *_lockptr() functions
1017 * will give us the same result: the per-MM PT lock.
1018 *
1019 * Note that with e.g., CONFIG_PGTABLE_LEVELS=2 where
1020 * PGDIR_SIZE==P4D_SIZE==PUD_SIZE==PMD_SIZE, we'd use pud_lockptr()
1021 * and core-mm would use pmd_lockptr(). However, in such configurations
1022 * split PMD locks are disabled -- they don't make sense on a single
1023 * PGDIR page table -- and the end result is the same.
1024 */
1025 if (size >= PUD_SIZE)
1026 return pud_lockptr(mm, (pud_t *) pte);
1027 else if (size >= PMD_SIZE || IS_ENABLED(CONFIG_HIGHPTE))
1028 return pmd_lockptr(mm, (pmd_t *) pte);
1029 /* pte_alloc_huge() only applies with !CONFIG_HIGHPTE */
1030 return ptep_lockptr(mm, pte);
1031}
1032
1033#ifndef hugepages_supported
1034/*
1035 * Some platform decide whether they support huge pages at boot
1036 * time. Some of them, such as powerpc, set HPAGE_SHIFT to 0
1037 * when there is no such support
1038 */
1039#define hugepages_supported() (HPAGE_SHIFT != 0)
1040#endif
1041
1042void hugetlb_report_usage(struct seq_file *m, struct mm_struct *mm);
1043
1044static inline void hugetlb_count_init(struct mm_struct *mm)
1045{
1046 atomic_long_set(&mm->hugetlb_usage, 0);
1047}
1048
1049static inline void hugetlb_count_add(long l, struct mm_struct *mm)
1050{
1051 atomic_long_add(l, &mm->hugetlb_usage);
1052}
1053
1054static inline void hugetlb_count_sub(long l, struct mm_struct *mm)
1055{
1056 atomic_long_sub(l, &mm->hugetlb_usage);
1057}
1058
1059#ifndef huge_ptep_modify_prot_start
1060#define huge_ptep_modify_prot_start huge_ptep_modify_prot_start
1061static inline pte_t huge_ptep_modify_prot_start(struct vm_area_struct *vma,
1062 unsigned long addr, pte_t *ptep)
1063{
1064 unsigned long psize = huge_page_size(hstate_vma(vma));
1065
1066 return huge_ptep_get_and_clear(vma->vm_mm, addr, ptep, psize);
1067}
1068#endif
1069
1070#ifndef huge_ptep_modify_prot_commit
1071#define huge_ptep_modify_prot_commit huge_ptep_modify_prot_commit
1072static inline void huge_ptep_modify_prot_commit(struct vm_area_struct *vma,
1073 unsigned long addr, pte_t *ptep,
1074 pte_t old_pte, pte_t pte)
1075{
1076 unsigned long psize = huge_page_size(hstate_vma(vma));
1077
1078 set_huge_pte_at(vma->vm_mm, addr, ptep, pte, psize);
1079}
1080#endif
1081
1082#ifdef CONFIG_NUMA
1083void hugetlb_register_node(struct node *node);
1084void hugetlb_unregister_node(struct node *node);
1085#endif
1086
1087/*
1088 * Check if a given raw @page in a hugepage is HWPOISON.
1089 */
1090bool is_raw_hwpoison_page_in_hugepage(struct page *page);
1091
1092static inline unsigned long huge_page_mask_align(struct file *file)
1093{
1094 return PAGE_MASK & ~huge_page_mask(hstate_file(file));
1095}
1096
1097#else /* CONFIG_HUGETLB_PAGE */
1098struct hstate {};
1099
1100static inline unsigned long huge_page_mask_align(struct file *file)
1101{
1102 return 0;
1103}
1104
1105static inline struct hugepage_subpool *hugetlb_folio_subpool(struct folio *folio)
1106{
1107 return NULL;
1108}
1109
1110static inline struct folio *filemap_lock_hugetlb_folio(struct hstate *h,
1111 struct address_space *mapping, pgoff_t idx)
1112{
1113 return NULL;
1114}
1115
1116static inline int isolate_or_dissolve_huge_folio(struct folio *folio,
1117 struct list_head *list)
1118{
1119 return -ENOMEM;
1120}
1121
1122static inline int replace_free_hugepage_folios(unsigned long start_pfn,
1123 unsigned long end_pfn)
1124{
1125 return 0;
1126}
1127
1128static inline void wait_for_freed_hugetlb_folios(void)
1129{
1130}
1131
1132static inline struct folio *alloc_hugetlb_folio(struct vm_area_struct *vma,
1133 unsigned long addr,
1134 bool cow_from_owner)
1135{
1136 return NULL;
1137}
1138
1139static inline struct folio *
1140alloc_hugetlb_folio_reserve(struct hstate *h, int preferred_nid,
1141 nodemask_t *nmask, gfp_t gfp_mask)
1142{
1143 return NULL;
1144}
1145
1146static inline struct folio *
1147alloc_hugetlb_folio_nodemask(struct hstate *h, int preferred_nid,
1148 nodemask_t *nmask, gfp_t gfp_mask,
1149 bool allow_alloc_fallback)
1150{
1151 return NULL;
1152}
1153
1154static inline int __alloc_bootmem_huge_page(struct hstate *h)
1155{
1156 return 0;
1157}
1158
1159static inline struct hstate *hstate_file(struct file *f)
1160{
1161 return NULL;
1162}
1163
1164static inline struct hstate *hstate_sizelog(int page_size_log)
1165{
1166 return NULL;
1167}
1168
1169static inline struct hstate *hstate_vma(struct vm_area_struct *vma)
1170{
1171 return NULL;
1172}
1173
1174static inline struct hstate *folio_hstate(struct folio *folio)
1175{
1176 return NULL;
1177}
1178
1179static inline struct hstate *size_to_hstate(unsigned long size)
1180{
1181 return NULL;
1182}
1183
1184static inline unsigned long huge_page_size(struct hstate *h)
1185{
1186 return PAGE_SIZE;
1187}
1188
1189static inline unsigned long huge_page_mask(struct hstate *h)
1190{
1191 return PAGE_MASK;
1192}
1193
1194static inline unsigned int huge_page_order(struct hstate *h)
1195{
1196 return 0;
1197}
1198
1199static inline unsigned int huge_page_shift(struct hstate *h)
1200{
1201 return PAGE_SHIFT;
1202}
1203
1204static inline bool hstate_is_gigantic(struct hstate *h)
1205{
1206 return false;
1207}
1208
1209static inline unsigned int pages_per_huge_page(struct hstate *h)
1210{
1211 return 1;
1212}
1213
1214static inline unsigned hstate_index_to_shift(unsigned index)
1215{
1216 return 0;
1217}
1218
1219static inline int hstate_index(struct hstate *h)
1220{
1221 return 0;
1222}
1223
1224static inline int dissolve_free_hugetlb_folio(struct folio *folio)
1225{
1226 return 0;
1227}
1228
1229static inline int dissolve_free_hugetlb_folios(unsigned long start_pfn,
1230 unsigned long end_pfn)
1231{
1232 return 0;
1233}
1234
1235static inline bool hugepage_migration_supported(struct hstate *h)
1236{
1237 return false;
1238}
1239
1240static inline bool hugepage_movable_supported(struct hstate *h)
1241{
1242 return false;
1243}
1244
1245static inline gfp_t htlb_alloc_mask(struct hstate *h)
1246{
1247 return 0;
1248}
1249
1250static inline gfp_t htlb_modify_alloc_mask(struct hstate *h, gfp_t gfp_mask)
1251{
1252 return 0;
1253}
1254
1255static inline bool htlb_allow_alloc_fallback(int reason)
1256{
1257 return false;
1258}
1259
1260static inline spinlock_t *huge_pte_lockptr(struct hstate *h,
1261 struct mm_struct *mm, pte_t *pte)
1262{
1263 return &mm->page_table_lock;
1264}
1265
1266static inline void hugetlb_count_init(struct mm_struct *mm)
1267{
1268}
1269
1270static inline void hugetlb_report_usage(struct seq_file *f, struct mm_struct *m)
1271{
1272}
1273
1274static inline void hugetlb_count_sub(long l, struct mm_struct *mm)
1275{
1276}
1277
1278static inline pte_t huge_ptep_clear_flush(struct vm_area_struct *vma,
1279 unsigned long addr, pte_t *ptep)
1280{
1281#ifdef CONFIG_MMU
1282 return ptep_get(ptep);
1283#else
1284 return *ptep;
1285#endif
1286}
1287
1288static inline void set_huge_pte_at(struct mm_struct *mm, unsigned long addr,
1289 pte_t *ptep, pte_t pte, unsigned long sz)
1290{
1291}
1292
1293static inline void hugetlb_register_node(struct node *node)
1294{
1295}
1296
1297static inline void hugetlb_unregister_node(struct node *node)
1298{
1299}
1300
1301static inline bool hugetlbfs_pagecache_present(
1302 struct hstate *h, struct vm_area_struct *vma, unsigned long address)
1303{
1304 return false;
1305}
1306
1307static inline void hugetlb_bootmem_alloc(void)
1308{
1309}
1310#endif /* CONFIG_HUGETLB_PAGE */
1311
1312static inline spinlock_t *huge_pte_lock(struct hstate *h,
1313 struct mm_struct *mm, pte_t *pte)
1314{
1315 spinlock_t *ptl;
1316
1317 ptl = huge_pte_lockptr(h, mm, pte);
1318 spin_lock(ptl);
1319 return ptl;
1320}
1321
1322#if defined(CONFIG_HUGETLB_PAGE) && defined(CONFIG_CMA)
1323extern void __init hugetlb_cma_reserve(void);
1324#else
1325static inline __init void hugetlb_cma_reserve(void)
1326{
1327}
1328#endif
1329
1330#ifdef CONFIG_HUGETLB_PMD_PAGE_TABLE_SHARING
1331static inline bool hugetlb_pmd_shared(pte_t *pte)
1332{
1333 return ptdesc_pmd_is_shared(virt_to_ptdesc(pte));
1334}
1335#else
1336static inline bool hugetlb_pmd_shared(pte_t *pte)
1337{
1338 return false;
1339}
1340#endif
1341
1342bool want_pmd_share(struct vm_area_struct *vma, unsigned long addr);
1343
1344#ifndef __HAVE_ARCH_FLUSH_HUGETLB_TLB_RANGE
1345/*
1346 * ARCHes with special requirements for evicting HUGETLB backing TLB entries can
1347 * implement this.
1348 */
1349#define flush_hugetlb_tlb_range(vma, addr, end) flush_tlb_range(vma, addr, end)
1350#endif
1351
1352static inline bool __vma_shareable_lock(struct vm_area_struct *vma)
1353{
1354 return (vma->vm_flags & VM_MAYSHARE) && vma->vm_private_data;
1355}
1356
1357bool __vma_private_lock(struct vm_area_struct *vma);
1358
1359/*
1360 * Safe version of huge_pte_offset() to check the locks. See comments
1361 * above huge_pte_offset().
1362 */
1363static inline pte_t *
1364hugetlb_walk(struct vm_area_struct *vma, unsigned long addr, unsigned long sz)
1365{
1366#if defined(CONFIG_HUGETLB_PMD_PAGE_TABLE_SHARING) && defined(CONFIG_LOCKDEP)
1367 struct hugetlb_vma_lock *vma_lock = vma->vm_private_data;
1368
1369 /*
1370 * If pmd sharing possible, locking needed to safely walk the
1371 * hugetlb pgtables. More information can be found at the comment
1372 * above huge_pte_offset() in the same file.
1373 *
1374 * NOTE: lockdep_is_held() is only defined with CONFIG_LOCKDEP.
1375 */
1376 if (__vma_shareable_lock(vma))
1377 WARN_ON_ONCE(!lockdep_is_held(&vma_lock->rw_sema) &&
1378 !lockdep_is_held(
1379 &vma->vm_file->f_mapping->i_mmap_rwsem));
1380#endif
1381 return huge_pte_offset(vma->vm_mm, addr, sz);
1382}
1383
1384#endif /* _LINUX_HUGETLB_H */