1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef _LINUX_HUGETLB_H
3 #define _LINUX_HUGETLB_H
5 #include <linux/mm_types.h>
6 #include <linux/mmdebug.h>
8 #include <linux/hugetlb_inline.h>
9 #include <linux/cgroup.h>
10 #include <linux/list.h>
11 #include <linux/kref.h>
12 #include <asm/pgtable.h>
19 typedef struct { unsigned long pd; } hugepd_t;
20 #define is_hugepd(hugepd) (0)
21 #define __hugepd(x) ((hugepd_t) { (x) })
24 #ifdef CONFIG_HUGETLB_PAGE
26 #include <linux/mempolicy.h>
27 #include <linux/shm.h>
28 #include <asm/tlbflush.h>
30 struct hugepage_subpool {
33 long max_hpages; /* Maximum huge pages or -1 if no maximum. */
34 long used_hpages; /* Used count against maximum, includes */
35 /* both alloced and reserved pages. */
36 struct hstate *hstate;
37 long min_hpages; /* Minimum huge pages or -1 if no minimum. */
38 long rsv_hpages; /* Pages reserved against global pool to */
39 /* sasitfy minimum size. */
45 struct list_head regions;
46 long adds_in_progress;
47 struct list_head region_cache;
48 long region_cache_count;
49 #ifdef CONFIG_CGROUP_HUGETLB
51 * On private mappings, the counter to uncharge reservations is stored
52 * here. If these fields are 0, then either the mapping is shared, or
53 * cgroup accounting is disabled for this resv_map.
55 struct page_counter *reservation_counter;
56 unsigned long pages_per_hpage;
57 struct cgroup_subsys_state *css;
62 * Region tracking -- allows tracking of reservations and instantiated pages
63 * across the pages in a mapping.
65 * The region data structures are embedded into a resv_map and protected
66 * by a resv_map's lock. The set of regions within the resv_map represent
67 * reservations for huge pages, or huge pages that have already been
68 * instantiated within the map. The from and to elements are huge page
69 * indicies into the associated mapping. from indicates the starting index
70 * of the region. to represents the first index past the end of the region.
72 * For example, a file region structure with from == 0 and to == 4 represents
73 * four huge pages in a mapping. It is important to note that the to element
74 * represents the first element past the end of the region. This is used in
75 * arithmetic as 4(to) - 0(from) = 4 huge pages in the region.
77 * Interval notation of the form [from, to) will be used to indicate that
78 * the endpoint from is inclusive and to is exclusive.
81 struct list_head link;
84 #ifdef CONFIG_CGROUP_HUGETLB
86 * On shared mappings, each reserved region appears as a struct
87 * file_region in resv_map. These fields hold the info needed to
88 * uncharge each reservation.
90 struct page_counter *reservation_counter;
91 struct cgroup_subsys_state *css;
95 extern struct resv_map *resv_map_alloc(void);
96 void resv_map_release(struct kref *ref);
98 extern spinlock_t hugetlb_lock;
99 extern int hugetlb_max_hstate __read_mostly;
100 #define for_each_hstate(h) \
101 for ((h) = hstates; (h) < &hstates[hugetlb_max_hstate]; (h)++)
103 struct hugepage_subpool *hugepage_new_subpool(struct hstate *h, long max_hpages,
105 void hugepage_put_subpool(struct hugepage_subpool *spool);
107 void reset_vma_resv_huge_pages(struct vm_area_struct *vma);
108 int hugetlb_sysctl_handler(struct ctl_table *, int, void __user *, size_t *, loff_t *);
109 int hugetlb_overcommit_handler(struct ctl_table *, int, void __user *, size_t *, loff_t *);
110 int hugetlb_treat_movable_handler(struct ctl_table *, int, void __user *, size_t *, loff_t *);
113 int hugetlb_mempolicy_sysctl_handler(struct ctl_table *, int,
114 void __user *, size_t *, loff_t *);
117 int copy_hugetlb_page_range(struct mm_struct *, struct mm_struct *, struct vm_area_struct *);
118 long follow_hugetlb_page(struct mm_struct *, struct vm_area_struct *,
119 struct page **, struct vm_area_struct **,
120 unsigned long *, unsigned long *, long, unsigned int,
122 void unmap_hugepage_range(struct vm_area_struct *,
123 unsigned long, unsigned long, struct page *);
124 void __unmap_hugepage_range_final(struct mmu_gather *tlb,
125 struct vm_area_struct *vma,
126 unsigned long start, unsigned long end,
127 struct page *ref_page);
128 void __unmap_hugepage_range(struct mmu_gather *tlb, struct vm_area_struct *vma,
129 unsigned long start, unsigned long end,
130 struct page *ref_page);
131 void hugetlb_report_meminfo(struct seq_file *);
132 int hugetlb_report_node_meminfo(int, char *);
133 void hugetlb_show_meminfo(void);
134 unsigned long hugetlb_total_pages(void);
135 vm_fault_t hugetlb_fault(struct mm_struct *mm, struct vm_area_struct *vma,
136 unsigned long address, unsigned int flags);
137 int hugetlb_mcopy_atomic_pte(struct mm_struct *dst_mm, pte_t *dst_pte,
138 struct vm_area_struct *dst_vma,
139 unsigned long dst_addr,
140 unsigned long src_addr,
141 struct page **pagep);
142 int hugetlb_reserve_pages(struct inode *inode, long from, long to,
143 struct vm_area_struct *vma,
144 vm_flags_t vm_flags);
145 long hugetlb_unreserve_pages(struct inode *inode, long start, long end,
147 bool isolate_huge_page(struct page *page, struct list_head *list);
148 void putback_active_hugepage(struct page *page);
149 void move_hugetlb_state(struct page *oldpage, struct page *newpage, int reason);
150 void free_huge_page(struct page *page);
151 void hugetlb_fix_reserve_counts(struct inode *inode);
152 extern struct mutex *hugetlb_fault_mutex_table;
153 u32 hugetlb_fault_mutex_hash(struct address_space *mapping, pgoff_t idx);
155 pte_t *huge_pmd_share(struct mm_struct *mm, unsigned long addr, pud_t *pud);
157 struct address_space *hugetlb_page_mapping_lock_write(struct page *hpage);
159 extern int sysctl_hugetlb_shm_group;
160 extern struct list_head huge_boot_pages;
164 pte_t *huge_pte_alloc(struct mm_struct *mm,
165 unsigned long addr, unsigned long sz);
166 pte_t *huge_pte_offset(struct mm_struct *mm,
167 unsigned long addr, unsigned long sz);
168 int huge_pmd_unshare(struct mm_struct *mm, unsigned long *addr, pte_t *ptep);
169 void adjust_range_if_pmd_sharing_possible(struct vm_area_struct *vma,
170 unsigned long *start, unsigned long *end);
171 struct page *follow_huge_addr(struct mm_struct *mm, unsigned long address,
173 struct page *follow_huge_pd(struct vm_area_struct *vma,
174 unsigned long address, hugepd_t hpd,
175 int flags, int pdshift);
176 struct page *follow_huge_pmd(struct mm_struct *mm, unsigned long address,
177 pmd_t *pmd, int flags);
178 struct page *follow_huge_pud(struct mm_struct *mm, unsigned long address,
179 pud_t *pud, int flags);
180 struct page *follow_huge_pgd(struct mm_struct *mm, unsigned long address,
181 pgd_t *pgd, int flags);
183 int pmd_huge(pmd_t pmd);
184 int pud_huge(pud_t pud);
185 unsigned long hugetlb_change_protection(struct vm_area_struct *vma,
186 unsigned long address, unsigned long end, pgprot_t newprot);
188 bool is_hugetlb_entry_migration(pte_t pte);
190 #else /* !CONFIG_HUGETLB_PAGE */
192 static inline void reset_vma_resv_huge_pages(struct vm_area_struct *vma)
196 static inline unsigned long hugetlb_total_pages(void)
201 static inline struct address_space *hugetlb_page_mapping_lock_write(
207 static inline int huge_pmd_unshare(struct mm_struct *mm, unsigned long *addr,
213 static inline void adjust_range_if_pmd_sharing_possible(
214 struct vm_area_struct *vma,
215 unsigned long *start, unsigned long *end)
219 static inline long follow_hugetlb_page(struct mm_struct *mm,
220 struct vm_area_struct *vma, struct page **pages,
221 struct vm_area_struct **vmas, unsigned long *position,
222 unsigned long *nr_pages, long i, unsigned int flags,
229 static inline struct page *follow_huge_addr(struct mm_struct *mm,
230 unsigned long address, int write)
232 return ERR_PTR(-EINVAL);
235 static inline int copy_hugetlb_page_range(struct mm_struct *dst,
236 struct mm_struct *src, struct vm_area_struct *vma)
242 static inline void hugetlb_report_meminfo(struct seq_file *m)
246 static inline int hugetlb_report_node_meminfo(int nid, char *buf)
251 static inline void hugetlb_show_meminfo(void)
255 static inline struct page *follow_huge_pd(struct vm_area_struct *vma,
256 unsigned long address, hugepd_t hpd, int flags,
262 static inline struct page *follow_huge_pmd(struct mm_struct *mm,
263 unsigned long address, pmd_t *pmd, int flags)
268 static inline struct page *follow_huge_pud(struct mm_struct *mm,
269 unsigned long address, pud_t *pud, int flags)
274 static inline struct page *follow_huge_pgd(struct mm_struct *mm,
275 unsigned long address, pgd_t *pgd, int flags)
280 static inline int prepare_hugepage_range(struct file *file,
281 unsigned long addr, unsigned long len)
286 static inline int pmd_huge(pmd_t pmd)
291 static inline int pud_huge(pud_t pud)
296 static inline int is_hugepage_only_range(struct mm_struct *mm,
297 unsigned long addr, unsigned long len)
302 static inline void hugetlb_free_pgd_range(struct mmu_gather *tlb,
303 unsigned long addr, unsigned long end,
304 unsigned long floor, unsigned long ceiling)
309 static inline int hugetlb_mcopy_atomic_pte(struct mm_struct *dst_mm,
311 struct vm_area_struct *dst_vma,
312 unsigned long dst_addr,
313 unsigned long src_addr,
320 static inline pte_t *huge_pte_offset(struct mm_struct *mm, unsigned long addr,
326 static inline bool isolate_huge_page(struct page *page, struct list_head *list)
331 static inline void putback_active_hugepage(struct page *page)
335 static inline void move_hugetlb_state(struct page *oldpage,
336 struct page *newpage, int reason)
340 static inline unsigned long hugetlb_change_protection(
341 struct vm_area_struct *vma, unsigned long address,
342 unsigned long end, pgprot_t newprot)
347 static inline void __unmap_hugepage_range_final(struct mmu_gather *tlb,
348 struct vm_area_struct *vma, unsigned long start,
349 unsigned long end, struct page *ref_page)
354 static inline void __unmap_hugepage_range(struct mmu_gather *tlb,
355 struct vm_area_struct *vma, unsigned long start,
356 unsigned long end, struct page *ref_page)
361 static inline vm_fault_t hugetlb_fault(struct mm_struct *mm,
362 struct vm_area_struct *vma, unsigned long address,
369 #endif /* !CONFIG_HUGETLB_PAGE */
371 * hugepages at page global directory. If arch support
372 * hugepages at pgd level, they need to define this.
375 #define pgd_huge(x) 0
378 #define p4d_huge(x) 0
382 static inline int pgd_write(pgd_t pgd)
389 #define HUGETLB_ANON_FILE "anon_hugepage"
393 * The file will be used as an shm file so shmfs accounting rules
396 HUGETLB_SHMFS_INODE = 1,
398 * The file is being created on the internal vfs mount and shmfs
399 * accounting rules do not apply
401 HUGETLB_ANONHUGE_INODE = 2,
404 #ifdef CONFIG_HUGETLBFS
405 struct hugetlbfs_sb_info {
406 long max_inodes; /* inodes allowed */
407 long free_inodes; /* inodes free */
408 spinlock_t stat_lock;
409 struct hstate *hstate;
410 struct hugepage_subpool *spool;
416 static inline struct hugetlbfs_sb_info *HUGETLBFS_SB(struct super_block *sb)
418 return sb->s_fs_info;
421 struct hugetlbfs_inode_info {
422 struct shared_policy policy;
423 struct inode vfs_inode;
427 static inline struct hugetlbfs_inode_info *HUGETLBFS_I(struct inode *inode)
429 return container_of(inode, struct hugetlbfs_inode_info, vfs_inode);
432 extern const struct file_operations hugetlbfs_file_operations;
433 extern const struct vm_operations_struct hugetlb_vm_ops;
434 struct file *hugetlb_file_setup(const char *name, size_t size, vm_flags_t acct,
435 struct user_struct **user, int creat_flags,
438 static inline bool is_file_hugepages(struct file *file)
440 if (file->f_op == &hugetlbfs_file_operations)
443 return is_file_shm_hugepages(file);
446 static inline struct hstate *hstate_inode(struct inode *i)
448 return HUGETLBFS_SB(i->i_sb)->hstate;
450 #else /* !CONFIG_HUGETLBFS */
452 #define is_file_hugepages(file) false
453 static inline struct file *
454 hugetlb_file_setup(const char *name, size_t size, vm_flags_t acctflag,
455 struct user_struct **user, int creat_flags,
458 return ERR_PTR(-ENOSYS);
461 static inline struct hstate *hstate_inode(struct inode *i)
465 #endif /* !CONFIG_HUGETLBFS */
467 #ifdef HAVE_ARCH_HUGETLB_UNMAPPED_AREA
468 unsigned long hugetlb_get_unmapped_area(struct file *file, unsigned long addr,
469 unsigned long len, unsigned long pgoff,
470 unsigned long flags);
471 #endif /* HAVE_ARCH_HUGETLB_UNMAPPED_AREA */
473 #ifdef CONFIG_HUGETLB_PAGE
475 #define HSTATE_NAME_LEN 32
476 /* Defines one hugetlb page size */
478 int next_nid_to_alloc;
479 int next_nid_to_free;
482 unsigned long max_huge_pages;
483 unsigned long nr_huge_pages;
484 unsigned long free_huge_pages;
485 unsigned long resv_huge_pages;
486 unsigned long surplus_huge_pages;
487 unsigned long nr_overcommit_huge_pages;
488 struct list_head hugepage_activelist;
489 struct list_head hugepage_freelists[MAX_NUMNODES];
490 unsigned int nr_huge_pages_node[MAX_NUMNODES];
491 unsigned int free_huge_pages_node[MAX_NUMNODES];
492 unsigned int surplus_huge_pages_node[MAX_NUMNODES];
493 #ifdef CONFIG_CGROUP_HUGETLB
494 /* cgroup control files */
495 struct cftype cgroup_files_dfl[7];
496 struct cftype cgroup_files_legacy[9];
498 char name[HSTATE_NAME_LEN];
501 struct huge_bootmem_page {
502 struct list_head list;
503 struct hstate *hstate;
506 struct page *alloc_huge_page(struct vm_area_struct *vma,
507 unsigned long addr, int avoid_reserve);
508 struct page *alloc_huge_page_node(struct hstate *h, int nid);
509 struct page *alloc_huge_page_nodemask(struct hstate *h, int preferred_nid,
511 struct page *alloc_huge_page_vma(struct hstate *h, struct vm_area_struct *vma,
512 unsigned long address);
513 struct page *alloc_migrate_huge_page(struct hstate *h, gfp_t gfp_mask,
514 int nid, nodemask_t *nmask);
515 int huge_add_to_page_cache(struct page *page, struct address_space *mapping,
519 int __init __alloc_bootmem_huge_page(struct hstate *h);
520 int __init alloc_bootmem_huge_page(struct hstate *h);
522 void __init hugetlb_bad_size(void);
523 void __init hugetlb_add_hstate(unsigned order);
524 struct hstate *size_to_hstate(unsigned long size);
526 #ifndef HUGE_MAX_HSTATE
527 #define HUGE_MAX_HSTATE 1
530 extern struct hstate hstates[HUGE_MAX_HSTATE];
531 extern unsigned int default_hstate_idx;
533 #define default_hstate (hstates[default_hstate_idx])
535 static inline struct hstate *hstate_file(struct file *f)
537 return hstate_inode(file_inode(f));
540 static inline struct hstate *hstate_sizelog(int page_size_log)
543 return &default_hstate;
545 return size_to_hstate(1UL << page_size_log);
548 static inline struct hstate *hstate_vma(struct vm_area_struct *vma)
550 return hstate_file(vma->vm_file);
553 static inline unsigned long huge_page_size(struct hstate *h)
555 return (unsigned long)PAGE_SIZE << h->order;
558 extern unsigned long vma_kernel_pagesize(struct vm_area_struct *vma);
560 extern unsigned long vma_mmu_pagesize(struct vm_area_struct *vma);
562 static inline unsigned long huge_page_mask(struct hstate *h)
567 static inline unsigned int huge_page_order(struct hstate *h)
572 static inline unsigned huge_page_shift(struct hstate *h)
574 return h->order + PAGE_SHIFT;
577 static inline bool hstate_is_gigantic(struct hstate *h)
579 return huge_page_order(h) >= MAX_ORDER;
582 static inline unsigned int pages_per_huge_page(struct hstate *h)
584 return 1 << h->order;
587 static inline unsigned int blocks_per_huge_page(struct hstate *h)
589 return huge_page_size(h) / 512;
592 #include <asm/hugetlb.h>
594 #ifndef arch_make_huge_pte
595 static inline pte_t arch_make_huge_pte(pte_t entry, struct vm_area_struct *vma,
596 struct page *page, int writable)
602 static inline struct hstate *page_hstate(struct page *page)
604 VM_BUG_ON_PAGE(!PageHuge(page), page);
605 return size_to_hstate(page_size(page));
608 static inline unsigned hstate_index_to_shift(unsigned index)
610 return hstates[index].order + PAGE_SHIFT;
613 static inline int hstate_index(struct hstate *h)
618 pgoff_t __basepage_index(struct page *page);
620 /* Return page->index in PAGE_SIZE units */
621 static inline pgoff_t basepage_index(struct page *page)
623 if (!PageCompound(page))
626 return __basepage_index(page);
629 extern int dissolve_free_huge_page(struct page *page);
630 extern int dissolve_free_huge_pages(unsigned long start_pfn,
631 unsigned long end_pfn);
633 #ifdef CONFIG_ARCH_ENABLE_HUGEPAGE_MIGRATION
634 #ifndef arch_hugetlb_migration_supported
635 static inline bool arch_hugetlb_migration_supported(struct hstate *h)
637 if ((huge_page_shift(h) == PMD_SHIFT) ||
638 (huge_page_shift(h) == PUD_SHIFT) ||
639 (huge_page_shift(h) == PGDIR_SHIFT))
646 static inline bool arch_hugetlb_migration_supported(struct hstate *h)
652 static inline bool hugepage_migration_supported(struct hstate *h)
654 return arch_hugetlb_migration_supported(h);
658 * Movability check is different as compared to migration check.
659 * It determines whether or not a huge page should be placed on
660 * movable zone or not. Movability of any huge page should be
661 * required only if huge page size is supported for migration.
662 * There wont be any reason for the huge page to be movable if
663 * it is not migratable to start with. Also the size of the huge
664 * page should be large enough to be placed under a movable zone
665 * and still feasible enough to be migratable. Just the presence
666 * in movable zone does not make the migration feasible.
668 * So even though large huge page sizes like the gigantic ones
669 * are migratable they should not be movable because its not
670 * feasible to migrate them from movable zone.
672 static inline bool hugepage_movable_supported(struct hstate *h)
674 if (!hugepage_migration_supported(h))
677 if (hstate_is_gigantic(h))
682 static inline spinlock_t *huge_pte_lockptr(struct hstate *h,
683 struct mm_struct *mm, pte_t *pte)
685 if (huge_page_size(h) == PMD_SIZE)
686 return pmd_lockptr(mm, (pmd_t *) pte);
687 VM_BUG_ON(huge_page_size(h) == PAGE_SIZE);
688 return &mm->page_table_lock;
691 #ifndef hugepages_supported
693 * Some platform decide whether they support huge pages at boot
694 * time. Some of them, such as powerpc, set HPAGE_SHIFT to 0
695 * when there is no such support
697 #define hugepages_supported() (HPAGE_SHIFT != 0)
700 void hugetlb_report_usage(struct seq_file *m, struct mm_struct *mm);
702 static inline void hugetlb_count_add(long l, struct mm_struct *mm)
704 atomic_long_add(l, &mm->hugetlb_usage);
707 static inline void hugetlb_count_sub(long l, struct mm_struct *mm)
709 atomic_long_sub(l, &mm->hugetlb_usage);
712 #ifndef set_huge_swap_pte_at
713 static inline void set_huge_swap_pte_at(struct mm_struct *mm, unsigned long addr,
714 pte_t *ptep, pte_t pte, unsigned long sz)
716 set_huge_pte_at(mm, addr, ptep, pte);
720 #ifndef huge_ptep_modify_prot_start
721 #define huge_ptep_modify_prot_start huge_ptep_modify_prot_start
722 static inline pte_t huge_ptep_modify_prot_start(struct vm_area_struct *vma,
723 unsigned long addr, pte_t *ptep)
725 return huge_ptep_get_and_clear(vma->vm_mm, addr, ptep);
729 #ifndef huge_ptep_modify_prot_commit
730 #define huge_ptep_modify_prot_commit huge_ptep_modify_prot_commit
731 static inline void huge_ptep_modify_prot_commit(struct vm_area_struct *vma,
732 unsigned long addr, pte_t *ptep,
733 pte_t old_pte, pte_t pte)
735 set_huge_pte_at(vma->vm_mm, addr, ptep, pte);
739 #else /* CONFIG_HUGETLB_PAGE */
742 static inline struct page *alloc_huge_page(struct vm_area_struct *vma,
749 static inline struct page *alloc_huge_page_node(struct hstate *h, int nid)
754 static inline struct page *
755 alloc_huge_page_nodemask(struct hstate *h, int preferred_nid, nodemask_t *nmask)
760 static inline struct page *alloc_huge_page_vma(struct hstate *h,
761 struct vm_area_struct *vma,
762 unsigned long address)
767 static inline int __alloc_bootmem_huge_page(struct hstate *h)
772 static inline struct hstate *hstate_file(struct file *f)
777 static inline struct hstate *hstate_sizelog(int page_size_log)
782 static inline struct hstate *hstate_vma(struct vm_area_struct *vma)
787 static inline struct hstate *page_hstate(struct page *page)
792 static inline unsigned long huge_page_size(struct hstate *h)
797 static inline unsigned long huge_page_mask(struct hstate *h)
802 static inline unsigned long vma_kernel_pagesize(struct vm_area_struct *vma)
807 static inline unsigned long vma_mmu_pagesize(struct vm_area_struct *vma)
812 static inline unsigned int huge_page_order(struct hstate *h)
817 static inline unsigned int huge_page_shift(struct hstate *h)
822 static inline bool hstate_is_gigantic(struct hstate *h)
827 static inline unsigned int pages_per_huge_page(struct hstate *h)
832 static inline unsigned hstate_index_to_shift(unsigned index)
837 static inline int hstate_index(struct hstate *h)
842 static inline pgoff_t basepage_index(struct page *page)
847 static inline int dissolve_free_huge_page(struct page *page)
852 static inline int dissolve_free_huge_pages(unsigned long start_pfn,
853 unsigned long end_pfn)
858 static inline bool hugepage_migration_supported(struct hstate *h)
863 static inline bool hugepage_movable_supported(struct hstate *h)
868 static inline spinlock_t *huge_pte_lockptr(struct hstate *h,
869 struct mm_struct *mm, pte_t *pte)
871 return &mm->page_table_lock;
874 static inline void hugetlb_report_usage(struct seq_file *f, struct mm_struct *m)
878 static inline void hugetlb_count_sub(long l, struct mm_struct *mm)
882 static inline void set_huge_swap_pte_at(struct mm_struct *mm, unsigned long addr,
883 pte_t *ptep, pte_t pte, unsigned long sz)
886 #endif /* CONFIG_HUGETLB_PAGE */
888 static inline spinlock_t *huge_pte_lock(struct hstate *h,
889 struct mm_struct *mm, pte_t *pte)
893 ptl = huge_pte_lockptr(h, mm, pte);
898 #if defined(CONFIG_HUGETLB_PAGE) && defined(CONFIG_CMA)
899 extern void __init hugetlb_cma_reserve(int order);
900 extern void __init hugetlb_cma_check(void);
902 static inline __init void hugetlb_cma_reserve(int order)
905 static inline __init void hugetlb_cma_check(void)
910 #endif /* _LINUX_HUGETLB_H */