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