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 unsigned int seals;
522};
523
524static inline struct hugetlbfs_inode_info *HUGETLBFS_I(struct inode *inode)
525{
526 return container_of(inode, struct hugetlbfs_inode_info, vfs_inode);
527}
528
529extern const struct vm_operations_struct hugetlb_vm_ops;
530struct file *hugetlb_file_setup(const char *name, size_t size, vma_flags_t acct,
531 int creat_flags, int page_size_log);
532
533static inline bool is_file_hugepages(const struct file *file)
534{
535 return file->f_op->fop_flags & FOP_HUGE_PAGES;
536}
537
538static inline struct hstate *hstate_inode(struct inode *i)
539{
540 return HUGETLBFS_SB(i->i_sb)->hstate;
541}
542#else /* !CONFIG_HUGETLBFS */
543
544#define is_file_hugepages(file) false
545static inline struct file *
546hugetlb_file_setup(const char *name, size_t size, vma_flags_t acctflag,
547 int creat_flags, int page_size_log)
548{
549 return ERR_PTR(-ENOSYS);
550}
551
552static inline struct hstate *hstate_inode(struct inode *i)
553{
554 return NULL;
555}
556#endif /* !CONFIG_HUGETLBFS */
557
558unsigned long
559hugetlb_get_unmapped_area(struct file *file, unsigned long addr,
560 unsigned long len, unsigned long pgoff,
561 unsigned long flags);
562
563/*
564 * huegtlb page specific state flags. These flags are located in page.private
565 * of the hugetlb head page. Functions created via the below macros should be
566 * used to manipulate these flags.
567 *
568 * HPG_restore_reserve - Set when a hugetlb page consumes a reservation at
569 * allocation time. Cleared when page is fully instantiated. Free
570 * routine checks flag to restore a reservation on error paths.
571 * Synchronization: Examined or modified by code that knows it has
572 * the only reference to page. i.e. After allocation but before use
573 * or when the page is being freed.
574 * HPG_migratable - Set after a newly allocated page is added to the page
575 * cache and/or page tables. Indicates the page is a candidate for
576 * migration.
577 * Synchronization: Initially set after new page allocation with no
578 * locking. When examined and modified during migration processing
579 * (isolate, migrate, putback) the hugetlb_lock is held.
580 * HPG_temporary - Set on a page that is temporarily allocated from the buddy
581 * allocator. Typically used for migration target pages when no pages
582 * are available in the pool. The hugetlb free page path will
583 * immediately free pages with this flag set to the buddy allocator.
584 * Synchronization: Can be set after huge page allocation from buddy when
585 * code knows it has only reference. All other examinations and
586 * modifications require hugetlb_lock.
587 * HPG_freed - Set when page is on the free lists.
588 * Synchronization: hugetlb_lock held for examination and modification.
589 * HPG_vmemmap_optimized - Set when the vmemmap pages of the page are freed.
590 * HPG_raw_hwp_unreliable - Set when the hugetlb page has a hwpoison sub-page
591 * that is not tracked by raw_hwp_page list.
592 */
593enum hugetlb_page_flags {
594 HPG_restore_reserve = 0,
595 HPG_migratable,
596 HPG_temporary,
597 HPG_freed,
598 HPG_vmemmap_optimized,
599 HPG_raw_hwp_unreliable,
600 HPG_cma,
601 __NR_HPAGEFLAGS,
602};
603
604/*
605 * Macros to create test, set and clear function definitions for
606 * hugetlb specific page flags.
607 */
608#ifdef CONFIG_HUGETLB_PAGE
609#define TESTHPAGEFLAG(uname, flname) \
610static __always_inline \
611bool folio_test_hugetlb_##flname(struct folio *folio) \
612 { void *private = &folio->private; \
613 return test_bit(HPG_##flname, private); \
614 }
615
616#define SETHPAGEFLAG(uname, flname) \
617static __always_inline \
618void folio_set_hugetlb_##flname(struct folio *folio) \
619 { void *private = &folio->private; \
620 set_bit(HPG_##flname, private); \
621 }
622
623#define CLEARHPAGEFLAG(uname, flname) \
624static __always_inline \
625void folio_clear_hugetlb_##flname(struct folio *folio) \
626 { void *private = &folio->private; \
627 clear_bit(HPG_##flname, private); \
628 }
629#else
630#define TESTHPAGEFLAG(uname, flname) \
631static inline bool \
632folio_test_hugetlb_##flname(struct folio *folio) \
633 { return 0; }
634
635#define SETHPAGEFLAG(uname, flname) \
636static inline void \
637folio_set_hugetlb_##flname(struct folio *folio) \
638 { }
639
640#define CLEARHPAGEFLAG(uname, flname) \
641static inline void \
642folio_clear_hugetlb_##flname(struct folio *folio) \
643 { }
644#endif
645
646#define HPAGEFLAG(uname, flname) \
647 TESTHPAGEFLAG(uname, flname) \
648 SETHPAGEFLAG(uname, flname) \
649 CLEARHPAGEFLAG(uname, flname) \
650
651/*
652 * Create functions associated with hugetlb page flags
653 */
654HPAGEFLAG(RestoreReserve, restore_reserve)
655HPAGEFLAG(Migratable, migratable)
656HPAGEFLAG(Temporary, temporary)
657HPAGEFLAG(Freed, freed)
658HPAGEFLAG(VmemmapOptimized, vmemmap_optimized)
659HPAGEFLAG(RawHwpUnreliable, raw_hwp_unreliable)
660HPAGEFLAG(Cma, cma)
661
662#ifdef CONFIG_HUGETLB_PAGE
663
664#define HSTATE_NAME_LEN 32
665/* Defines one hugetlb page size */
666struct hstate {
667 struct mutex resize_lock;
668 struct lock_class_key resize_key;
669 int next_nid_to_alloc;
670 int next_nid_to_free;
671 unsigned int order;
672 unsigned int demote_order;
673 unsigned long mask;
674 unsigned long max_huge_pages;
675 unsigned long nr_huge_pages;
676 unsigned long free_huge_pages;
677 unsigned long resv_huge_pages;
678 unsigned long surplus_huge_pages;
679 unsigned long nr_overcommit_huge_pages;
680 struct list_head hugepage_activelist;
681 struct list_head hugepage_freelists[MAX_NUMNODES];
682 unsigned int max_huge_pages_node[MAX_NUMNODES];
683 unsigned int nr_huge_pages_node[MAX_NUMNODES];
684 unsigned int free_huge_pages_node[MAX_NUMNODES];
685 unsigned int surplus_huge_pages_node[MAX_NUMNODES];
686 char name[HSTATE_NAME_LEN];
687};
688
689struct cma;
690
691struct huge_bootmem_page {
692 struct list_head list;
693 struct hstate *hstate;
694 unsigned long flags;
695 struct cma *cma;
696};
697
698#define HUGE_BOOTMEM_HVO 0x0001
699#define HUGE_BOOTMEM_ZONES_VALID 0x0002
700#define HUGE_BOOTMEM_CMA 0x0004
701
702bool hugetlb_bootmem_page_zones_valid(int nid, struct huge_bootmem_page *m);
703
704int isolate_or_dissolve_huge_folio(struct folio *folio, struct list_head *list);
705int replace_free_hugepage_folios(unsigned long start_pfn, unsigned long end_pfn);
706void wait_for_freed_hugetlb_folios(void);
707struct folio *alloc_hugetlb_folio(struct vm_area_struct *vma,
708 unsigned long addr, bool cow_from_owner);
709struct folio *alloc_hugetlb_folio_nodemask(struct hstate *h, int preferred_nid,
710 nodemask_t *nmask, gfp_t gfp_mask,
711 bool allow_alloc_fallback);
712struct folio *alloc_hugetlb_folio_reserve(struct hstate *h, int preferred_nid,
713 nodemask_t *nmask, gfp_t gfp_mask);
714
715int hugetlb_add_to_page_cache(struct folio *folio, struct address_space *mapping,
716 pgoff_t idx);
717void restore_reserve_on_error(struct hstate *h, struct vm_area_struct *vma,
718 unsigned long address, struct folio *folio);
719
720/* arch callback */
721int __init __alloc_bootmem_huge_page(struct hstate *h, int nid);
722int __init alloc_bootmem_huge_page(struct hstate *h, int nid);
723bool __init hugetlb_node_alloc_supported(void);
724
725void __init hugetlb_add_hstate(unsigned order);
726bool __init arch_hugetlb_valid_size(unsigned long size);
727struct hstate *size_to_hstate(unsigned long size);
728
729#ifndef HUGE_MAX_HSTATE
730#define HUGE_MAX_HSTATE 1
731#endif
732
733extern struct hstate hstates[HUGE_MAX_HSTATE];
734extern unsigned int default_hstate_idx;
735
736#define default_hstate (hstates[default_hstate_idx])
737
738static inline struct hugepage_subpool *subpool_inode(struct inode *inode)
739{
740 return HUGETLBFS_SB(inode->i_sb)->spool;
741}
742
743static inline struct hugepage_subpool *hugetlb_folio_subpool(struct folio *folio)
744{
745 return folio->_hugetlb_subpool;
746}
747
748static inline void hugetlb_set_folio_subpool(struct folio *folio,
749 struct hugepage_subpool *subpool)
750{
751 folio->_hugetlb_subpool = subpool;
752}
753
754static inline struct hstate *hstate_file(struct file *f)
755{
756 return hstate_inode(file_inode(f));
757}
758
759static inline struct hstate *hstate_sizelog(int page_size_log)
760{
761 if (!page_size_log)
762 return &default_hstate;
763
764 if (page_size_log < BITS_PER_LONG)
765 return size_to_hstate(1UL << page_size_log);
766
767 return NULL;
768}
769
770static inline struct hstate *hstate_vma(struct vm_area_struct *vma)
771{
772 return hstate_file(vma->vm_file);
773}
774
775static inline unsigned long huge_page_size(const struct hstate *h)
776{
777 return (unsigned long)PAGE_SIZE << h->order;
778}
779
780extern unsigned long vma_kernel_pagesize(struct vm_area_struct *vma);
781
782extern unsigned long vma_mmu_pagesize(struct vm_area_struct *vma);
783
784static inline unsigned long huge_page_mask(struct hstate *h)
785{
786 return h->mask;
787}
788
789static inline unsigned int huge_page_order(struct hstate *h)
790{
791 return h->order;
792}
793
794static inline unsigned huge_page_shift(struct hstate *h)
795{
796 return h->order + PAGE_SHIFT;
797}
798
799static inline bool order_is_gigantic(unsigned int order)
800{
801 return order > MAX_PAGE_ORDER;
802}
803
804static inline bool hstate_is_gigantic(struct hstate *h)
805{
806 return order_is_gigantic(huge_page_order(h));
807}
808
809static inline unsigned int pages_per_huge_page(const struct hstate *h)
810{
811 return 1 << h->order;
812}
813
814static inline unsigned int blocks_per_huge_page(struct hstate *h)
815{
816 return huge_page_size(h) / 512;
817}
818
819static inline struct folio *filemap_lock_hugetlb_folio(struct hstate *h,
820 struct address_space *mapping, pgoff_t idx)
821{
822 return filemap_lock_folio(mapping, idx << huge_page_order(h));
823}
824
825#include <asm/hugetlb.h>
826
827#ifndef is_hugepage_only_range
828static inline int is_hugepage_only_range(struct mm_struct *mm,
829 unsigned long addr, unsigned long len)
830{
831 return 0;
832}
833#define is_hugepage_only_range is_hugepage_only_range
834#endif
835
836#ifndef arch_clear_hugetlb_flags
837static inline void arch_clear_hugetlb_flags(struct folio *folio) { }
838#define arch_clear_hugetlb_flags arch_clear_hugetlb_flags
839#endif
840
841#ifndef arch_make_huge_pte
842static inline pte_t arch_make_huge_pte(pte_t entry, unsigned int shift,
843 vm_flags_t flags)
844{
845 return pte_mkhuge(entry);
846}
847#endif
848
849#ifndef arch_has_huge_bootmem_alloc
850/*
851 * Some architectures do their own bootmem allocation, so they can't use
852 * early CMA allocation.
853 */
854static inline bool arch_has_huge_bootmem_alloc(void)
855{
856 return false;
857}
858#endif
859
860static inline struct hstate *folio_hstate(struct folio *folio)
861{
862 VM_BUG_ON_FOLIO(!folio_test_hugetlb(folio), folio);
863 return size_to_hstate(folio_size(folio));
864}
865
866static inline unsigned hstate_index_to_shift(unsigned index)
867{
868 return hstates[index].order + PAGE_SHIFT;
869}
870
871static inline int hstate_index(struct hstate *h)
872{
873 return h - hstates;
874}
875
876int dissolve_free_hugetlb_folio(struct folio *folio);
877int dissolve_free_hugetlb_folios(unsigned long start_pfn,
878 unsigned long end_pfn);
879
880#ifdef CONFIG_MEMORY_FAILURE
881extern void folio_clear_hugetlb_hwpoison(struct folio *folio);
882#else
883static inline void folio_clear_hugetlb_hwpoison(struct folio *folio)
884{
885}
886#endif
887
888#ifdef CONFIG_ARCH_ENABLE_HUGEPAGE_MIGRATION
889#ifndef arch_hugetlb_migration_supported
890static inline bool arch_hugetlb_migration_supported(struct hstate *h)
891{
892 if ((huge_page_shift(h) == PMD_SHIFT) ||
893 (huge_page_shift(h) == PUD_SHIFT) ||
894 (huge_page_shift(h) == PGDIR_SHIFT))
895 return true;
896 else
897 return false;
898}
899#endif
900#else
901static inline bool arch_hugetlb_migration_supported(struct hstate *h)
902{
903 return false;
904}
905#endif
906
907static inline bool hugepage_migration_supported(struct hstate *h)
908{
909 return arch_hugetlb_migration_supported(h);
910}
911
912/*
913 * Movability check is different as compared to migration check.
914 * It determines whether or not a huge page should be placed on
915 * movable zone or not. Movability of any huge page should be
916 * required only if huge page size is supported for migration.
917 * There won't be any reason for the huge page to be movable if
918 * it is not migratable to start with. Also the size of the huge
919 * page should be large enough to be placed under a movable zone
920 * and still feasible enough to be migratable. Just the presence
921 * in movable zone does not make the migration feasible.
922 *
923 * So even though large huge page sizes like the gigantic ones
924 * are migratable they should not be movable because its not
925 * feasible to migrate them from movable zone.
926 */
927static inline bool hugepage_movable_supported(struct hstate *h)
928{
929 if (!hugepage_migration_supported(h))
930 return false;
931
932 if (hstate_is_gigantic(h) && !movable_gigantic_pages)
933 return false;
934 return true;
935}
936
937/* Movability of hugepages depends on migration support. */
938static inline gfp_t htlb_alloc_mask(struct hstate *h)
939{
940 gfp_t gfp = __GFP_COMP | __GFP_NOWARN;
941
942 gfp |= hugepage_movable_supported(h) ? GFP_HIGHUSER_MOVABLE : GFP_HIGHUSER;
943
944 return gfp;
945}
946
947static inline gfp_t htlb_modify_alloc_mask(struct hstate *h, gfp_t gfp_mask)
948{
949 gfp_t modified_mask = htlb_alloc_mask(h);
950
951 /* Some callers might want to enforce node */
952 modified_mask |= (gfp_mask & __GFP_THISNODE);
953
954 modified_mask |= (gfp_mask & __GFP_NOWARN);
955
956 return modified_mask;
957}
958
959static inline bool htlb_allow_alloc_fallback(int reason)
960{
961 bool allowed_fallback = false;
962
963 /*
964 * Note: the memory offline, memory failure and migration syscalls will
965 * be allowed to fallback to other nodes due to lack of a better chioce,
966 * that might break the per-node hugetlb pool. While other cases will
967 * set the __GFP_THISNODE to avoid breaking the per-node hugetlb pool.
968 */
969 switch (reason) {
970 case MR_MEMORY_HOTPLUG:
971 case MR_MEMORY_FAILURE:
972 case MR_SYSCALL:
973 case MR_MEMPOLICY_MBIND:
974 allowed_fallback = true;
975 break;
976 default:
977 break;
978 }
979
980 return allowed_fallback;
981}
982
983static inline spinlock_t *huge_pte_lockptr(struct hstate *h,
984 struct mm_struct *mm, pte_t *pte)
985{
986 const unsigned long size = huge_page_size(h);
987
988 VM_WARN_ON(size == PAGE_SIZE);
989
990 /*
991 * hugetlb must use the exact same PT locks as core-mm page table
992 * walkers would. When modifying a PTE table, hugetlb must take the
993 * PTE PT lock, when modifying a PMD table, hugetlb must take the PMD
994 * PT lock etc.
995 *
996 * The expectation is that any hugetlb folio smaller than a PMD is
997 * always mapped into a single PTE table and that any hugetlb folio
998 * smaller than a PUD (but at least as big as a PMD) is always mapped
999 * into a single PMD table.
1000 *
1001 * If that does not hold for an architecture, then that architecture
1002 * must disable split PT locks such that all *_lockptr() functions
1003 * will give us the same result: the per-MM PT lock.
1004 *
1005 * Note that with e.g., CONFIG_PGTABLE_LEVELS=2 where
1006 * PGDIR_SIZE==P4D_SIZE==PUD_SIZE==PMD_SIZE, we'd use pud_lockptr()
1007 * and core-mm would use pmd_lockptr(). However, in such configurations
1008 * split PMD locks are disabled -- they don't make sense on a single
1009 * PGDIR page table -- and the end result is the same.
1010 */
1011 if (size >= PUD_SIZE)
1012 return pud_lockptr(mm, (pud_t *) pte);
1013 else if (size >= PMD_SIZE || IS_ENABLED(CONFIG_HIGHPTE))
1014 return pmd_lockptr(mm, (pmd_t *) pte);
1015 /* pte_alloc_huge() only applies with !CONFIG_HIGHPTE */
1016 return ptep_lockptr(mm, pte);
1017}
1018
1019#ifndef hugepages_supported
1020/*
1021 * Some platform decide whether they support huge pages at boot
1022 * time. Some of them, such as powerpc, set HPAGE_SHIFT to 0
1023 * when there is no such support
1024 */
1025#define hugepages_supported() (HPAGE_SHIFT != 0)
1026#endif
1027
1028void hugetlb_report_usage(struct seq_file *m, struct mm_struct *mm);
1029
1030static inline void hugetlb_count_init(struct mm_struct *mm)
1031{
1032 atomic_long_set(&mm->hugetlb_usage, 0);
1033}
1034
1035static inline void hugetlb_count_add(long l, struct mm_struct *mm)
1036{
1037 atomic_long_add(l, &mm->hugetlb_usage);
1038}
1039
1040static inline void hugetlb_count_sub(long l, struct mm_struct *mm)
1041{
1042 atomic_long_sub(l, &mm->hugetlb_usage);
1043}
1044
1045#ifndef huge_ptep_modify_prot_start
1046#define huge_ptep_modify_prot_start huge_ptep_modify_prot_start
1047static inline pte_t huge_ptep_modify_prot_start(struct vm_area_struct *vma,
1048 unsigned long addr, pte_t *ptep)
1049{
1050 unsigned long psize = huge_page_size(hstate_vma(vma));
1051
1052 return huge_ptep_get_and_clear(vma->vm_mm, addr, ptep, psize);
1053}
1054#endif
1055
1056#ifndef huge_ptep_modify_prot_commit
1057#define huge_ptep_modify_prot_commit huge_ptep_modify_prot_commit
1058static inline void huge_ptep_modify_prot_commit(struct vm_area_struct *vma,
1059 unsigned long addr, pte_t *ptep,
1060 pte_t old_pte, pte_t pte)
1061{
1062 unsigned long psize = huge_page_size(hstate_vma(vma));
1063
1064 set_huge_pte_at(vma->vm_mm, addr, ptep, pte, psize);
1065}
1066#endif
1067
1068#ifdef CONFIG_NUMA
1069void hugetlb_register_node(struct node *node);
1070void hugetlb_unregister_node(struct node *node);
1071#endif
1072
1073/*
1074 * Check if a given raw @page in a hugepage is HWPOISON.
1075 */
1076bool is_raw_hwpoison_page_in_hugepage(struct page *page);
1077
1078static inline unsigned long huge_page_mask_align(struct file *file)
1079{
1080 return PAGE_MASK & ~huge_page_mask(hstate_file(file));
1081}
1082
1083#else /* CONFIG_HUGETLB_PAGE */
1084struct hstate {};
1085
1086static inline unsigned long huge_page_mask_align(struct file *file)
1087{
1088 return 0;
1089}
1090
1091static inline struct hugepage_subpool *hugetlb_folio_subpool(struct folio *folio)
1092{
1093 return NULL;
1094}
1095
1096static inline struct folio *filemap_lock_hugetlb_folio(struct hstate *h,
1097 struct address_space *mapping, pgoff_t idx)
1098{
1099 return NULL;
1100}
1101
1102static inline int isolate_or_dissolve_huge_folio(struct folio *folio,
1103 struct list_head *list)
1104{
1105 return -ENOMEM;
1106}
1107
1108static inline int replace_free_hugepage_folios(unsigned long start_pfn,
1109 unsigned long end_pfn)
1110{
1111 return 0;
1112}
1113
1114static inline void wait_for_freed_hugetlb_folios(void)
1115{
1116}
1117
1118static inline struct folio *alloc_hugetlb_folio(struct vm_area_struct *vma,
1119 unsigned long addr,
1120 bool cow_from_owner)
1121{
1122 return NULL;
1123}
1124
1125static inline struct folio *
1126alloc_hugetlb_folio_reserve(struct hstate *h, int preferred_nid,
1127 nodemask_t *nmask, gfp_t gfp_mask)
1128{
1129 return NULL;
1130}
1131
1132static inline struct folio *
1133alloc_hugetlb_folio_nodemask(struct hstate *h, int preferred_nid,
1134 nodemask_t *nmask, gfp_t gfp_mask,
1135 bool allow_alloc_fallback)
1136{
1137 return NULL;
1138}
1139
1140static inline int __alloc_bootmem_huge_page(struct hstate *h)
1141{
1142 return 0;
1143}
1144
1145static inline struct hstate *hstate_file(struct file *f)
1146{
1147 return NULL;
1148}
1149
1150static inline struct hstate *hstate_sizelog(int page_size_log)
1151{
1152 return NULL;
1153}
1154
1155static inline struct hstate *hstate_vma(struct vm_area_struct *vma)
1156{
1157 return NULL;
1158}
1159
1160static inline struct hstate *folio_hstate(struct folio *folio)
1161{
1162 return NULL;
1163}
1164
1165static inline struct hstate *size_to_hstate(unsigned long size)
1166{
1167 return NULL;
1168}
1169
1170static inline unsigned long huge_page_size(struct hstate *h)
1171{
1172 return PAGE_SIZE;
1173}
1174
1175static inline unsigned long huge_page_mask(struct hstate *h)
1176{
1177 return PAGE_MASK;
1178}
1179
1180static inline unsigned long vma_kernel_pagesize(struct vm_area_struct *vma)
1181{
1182 return PAGE_SIZE;
1183}
1184
1185static inline unsigned long vma_mmu_pagesize(struct vm_area_struct *vma)
1186{
1187 return PAGE_SIZE;
1188}
1189
1190static inline unsigned int huge_page_order(struct hstate *h)
1191{
1192 return 0;
1193}
1194
1195static inline unsigned int huge_page_shift(struct hstate *h)
1196{
1197 return PAGE_SHIFT;
1198}
1199
1200static inline bool hstate_is_gigantic(struct hstate *h)
1201{
1202 return false;
1203}
1204
1205static inline unsigned int pages_per_huge_page(struct hstate *h)
1206{
1207 return 1;
1208}
1209
1210static inline unsigned hstate_index_to_shift(unsigned index)
1211{
1212 return 0;
1213}
1214
1215static inline int hstate_index(struct hstate *h)
1216{
1217 return 0;
1218}
1219
1220static inline int dissolve_free_hugetlb_folio(struct folio *folio)
1221{
1222 return 0;
1223}
1224
1225static inline int dissolve_free_hugetlb_folios(unsigned long start_pfn,
1226 unsigned long end_pfn)
1227{
1228 return 0;
1229}
1230
1231static inline bool hugepage_migration_supported(struct hstate *h)
1232{
1233 return false;
1234}
1235
1236static inline bool hugepage_movable_supported(struct hstate *h)
1237{
1238 return false;
1239}
1240
1241static inline gfp_t htlb_alloc_mask(struct hstate *h)
1242{
1243 return 0;
1244}
1245
1246static inline gfp_t htlb_modify_alloc_mask(struct hstate *h, gfp_t gfp_mask)
1247{
1248 return 0;
1249}
1250
1251static inline bool htlb_allow_alloc_fallback(int reason)
1252{
1253 return false;
1254}
1255
1256static inline spinlock_t *huge_pte_lockptr(struct hstate *h,
1257 struct mm_struct *mm, pte_t *pte)
1258{
1259 return &mm->page_table_lock;
1260}
1261
1262static inline void hugetlb_count_init(struct mm_struct *mm)
1263{
1264}
1265
1266static inline void hugetlb_report_usage(struct seq_file *f, struct mm_struct *m)
1267{
1268}
1269
1270static inline void hugetlb_count_sub(long l, struct mm_struct *mm)
1271{
1272}
1273
1274static inline pte_t huge_ptep_clear_flush(struct vm_area_struct *vma,
1275 unsigned long addr, pte_t *ptep)
1276{
1277#ifdef CONFIG_MMU
1278 return ptep_get(ptep);
1279#else
1280 return *ptep;
1281#endif
1282}
1283
1284static inline void set_huge_pte_at(struct mm_struct *mm, unsigned long addr,
1285 pte_t *ptep, pte_t pte, unsigned long sz)
1286{
1287}
1288
1289static inline void hugetlb_register_node(struct node *node)
1290{
1291}
1292
1293static inline void hugetlb_unregister_node(struct node *node)
1294{
1295}
1296
1297static inline bool hugetlbfs_pagecache_present(
1298 struct hstate *h, struct vm_area_struct *vma, unsigned long address)
1299{
1300 return false;
1301}
1302
1303static inline void hugetlb_bootmem_alloc(void)
1304{
1305}
1306#endif /* CONFIG_HUGETLB_PAGE */
1307
1308static inline spinlock_t *huge_pte_lock(struct hstate *h,
1309 struct mm_struct *mm, pte_t *pte)
1310{
1311 spinlock_t *ptl;
1312
1313 ptl = huge_pte_lockptr(h, mm, pte);
1314 spin_lock(ptl);
1315 return ptl;
1316}
1317
1318#if defined(CONFIG_HUGETLB_PAGE) && defined(CONFIG_CMA)
1319extern void __init hugetlb_cma_reserve(void);
1320#else
1321static inline __init void hugetlb_cma_reserve(void)
1322{
1323}
1324#endif
1325
1326#ifdef CONFIG_HUGETLB_PMD_PAGE_TABLE_SHARING
1327static inline bool hugetlb_pmd_shared(pte_t *pte)
1328{
1329 return ptdesc_pmd_is_shared(virt_to_ptdesc(pte));
1330}
1331#else
1332static inline bool hugetlb_pmd_shared(pte_t *pte)
1333{
1334 return false;
1335}
1336#endif
1337
1338bool want_pmd_share(struct vm_area_struct *vma, unsigned long addr);
1339
1340#ifndef __HAVE_ARCH_FLUSH_HUGETLB_TLB_RANGE
1341/*
1342 * ARCHes with special requirements for evicting HUGETLB backing TLB entries can
1343 * implement this.
1344 */
1345#define flush_hugetlb_tlb_range(vma, addr, end) flush_tlb_range(vma, addr, end)
1346#endif
1347
1348static inline bool __vma_shareable_lock(struct vm_area_struct *vma)
1349{
1350 return (vma->vm_flags & VM_MAYSHARE) && vma->vm_private_data;
1351}
1352
1353bool __vma_private_lock(struct vm_area_struct *vma);
1354
1355/*
1356 * Safe version of huge_pte_offset() to check the locks. See comments
1357 * above huge_pte_offset().
1358 */
1359static inline pte_t *
1360hugetlb_walk(struct vm_area_struct *vma, unsigned long addr, unsigned long sz)
1361{
1362#if defined(CONFIG_HUGETLB_PMD_PAGE_TABLE_SHARING) && defined(CONFIG_LOCKDEP)
1363 struct hugetlb_vma_lock *vma_lock = vma->vm_private_data;
1364
1365 /*
1366 * If pmd sharing possible, locking needed to safely walk the
1367 * hugetlb pgtables. More information can be found at the comment
1368 * above huge_pte_offset() in the same file.
1369 *
1370 * NOTE: lockdep_is_held() is only defined with CONFIG_LOCKDEP.
1371 */
1372 if (__vma_shareable_lock(vma))
1373 WARN_ON_ONCE(!lockdep_is_held(&vma_lock->rw_sema) &&
1374 !lockdep_is_held(
1375 &vma->vm_file->f_mapping->i_mmap_rwsem));
1376#endif
1377 return huge_pte_offset(vma->vm_mm, addr, sz);
1378}
1379
1380#endif /* _LINUX_HUGETLB_H */