include/linux/hugetlb.h

   1 /* SPDX-License-Identifier: GPL-2.0 */
   2 #ifndef _LINUX_HUGETLB_H
   3 #define _LINUX_HUGETLB_H
   4
   5 #include <linux/mm_types.h>
   6 #include <linux/mmdebug.h>
   7 #include <linux/fs.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>
  13
  14 struct ctl_table;
  15 struct user_struct;
  16 struct mmu_gather;
  17
  18 #ifndef is_hugepd
  19 /*
  20  * Some architectures requires a hugepage directory format that is
  21  * required to support multiple hugepage sizes. For example
  22  * a4fe3ce76 "powerpc/mm: Allow more flexible layouts for hugepage pagetables"
  23  * introduced the same on powerpc. This allows for a more flexible hugepage
  24  * pagetable layout.
  25  */
  26 typedef struct { unsigned long pd; } hugepd_t;
  27 #define is_hugepd(hugepd) (0)
  28 #define __hugepd(x) ((hugepd_t) { (x) })
  29 static inline int gup_huge_pd(hugepd_t hugepd, unsigned long addr,
  30                               unsigned pdshift, unsigned long end,
  31                               int write, struct page **pages, int *nr)
  32 {
  33         return 0;
  34 }
  35 #else
  36 extern int gup_huge_pd(hugepd_t hugepd, unsigned long addr,
  37                        unsigned pdshift, unsigned long end,
  38                        int write, struct page **pages, int *nr);
  39 #endif
  40
  41
  42 #ifdef CONFIG_HUGETLB_PAGE
  43
  44 #include <linux/mempolicy.h>
  45 #include <linux/shm.h>
  46 #include <asm/tlbflush.h>
  47
  48 struct hugepage_subpool {
  49         spinlock_t lock;
  50         long count;
  51         long max_hpages;        /* Maximum huge pages or -1 if no maximum. */
  52         long used_hpages;       /* Used count against maximum, includes */
  53                                 /* both alloced and reserved pages. */
  54         struct hstate *hstate;
  55         long min_hpages;        /* Minimum huge pages or -1 if no minimum. */
  56         long rsv_hpages;        /* Pages reserved against global pool to */
  57                                 /* sasitfy minimum size. */
  58 };
  59
  60 struct resv_map {
  61         struct kref refs;
  62         spinlock_t lock;
  63         struct list_head regions;
  64         long adds_in_progress;
  65         struct list_head region_cache;
  66         long region_cache_count;
  67 };
  68 extern struct resv_map *resv_map_alloc(void);
  69 void resv_map_release(struct kref *ref);
  70
  71 extern spinlock_t hugetlb_lock;
  72 extern int hugetlb_max_hstate __read_mostly;
  73 #define for_each_hstate(h) \
  74         for ((h) = hstates; (h) < &hstates[hugetlb_max_hstate]; (h)++)
  75
  76 struct hugepage_subpool *hugepage_new_subpool(struct hstate *h, long max_hpages,
  77                                                 long min_hpages);
  78 void hugepage_put_subpool(struct hugepage_subpool *spool);
  79
  80 void reset_vma_resv_huge_pages(struct vm_area_struct *vma);
  81 int hugetlb_sysctl_handler(struct ctl_table *, int, void __user *, size_t *, loff_t *);
  82 int hugetlb_overcommit_handler(struct ctl_table *, int, void __user *, size_t *, loff_t *);
  83 int hugetlb_treat_movable_handler(struct ctl_table *, int, void __user *, size_t *, loff_t *);
  84
  85 #ifdef CONFIG_NUMA
  86 int hugetlb_mempolicy_sysctl_handler(struct ctl_table *, int,
  87                                         void __user *, size_t *, loff_t *);
  88 #endif
  89
  90 int copy_hugetlb_page_range(struct mm_struct *, struct mm_struct *, struct vm_area_struct *);
  91 long follow_hugetlb_page(struct mm_struct *, struct vm_area_struct *,
  92                          struct page **, struct vm_area_struct **,
  93                          unsigned long *, unsigned long *, long, unsigned int,
  94                          int *);
  95 void unmap_hugepage_range(struct vm_area_struct *,
  96                           unsigned long, unsigned long, struct page *);
  97 void __unmap_hugepage_range_final(struct mmu_gather *tlb,
  98                           struct vm_area_struct *vma,
  99                           unsigned long start, unsigned long end,
 100                           struct page *ref_page);
 101 void __unmap_hugepage_range(struct mmu_gather *tlb, struct vm_area_struct *vma,
 102                                 unsigned long start, unsigned long end,
 103                                 struct page *ref_page);
 104 void hugetlb_report_meminfo(struct seq_file *);
 105 int hugetlb_report_node_meminfo(int, char *);
 106 void hugetlb_show_meminfo(void);
 107 unsigned long hugetlb_total_pages(void);
 108 int hugetlb_fault(struct mm_struct *mm, struct vm_area_struct *vma,
 109                         unsigned long address, unsigned int flags);
 110 int hugetlb_mcopy_atomic_pte(struct mm_struct *dst_mm, pte_t *dst_pte,
 111                                 struct vm_area_struct *dst_vma,
 112                                 unsigned long dst_addr,
 113                                 unsigned long src_addr,
 114                                 struct page **pagep);
 115 int hugetlb_reserve_pages(struct inode *inode, long from, long to,
 116                                                 struct vm_area_struct *vma,
 117                                                 vm_flags_t vm_flags);
 118 long hugetlb_unreserve_pages(struct inode *inode, long start, long end,
 119                                                 long freed);
 120 bool isolate_huge_page(struct page *page, struct list_head *list);
 121 void putback_active_hugepage(struct page *page);
 122 void free_huge_page(struct page *page);
 123 void hugetlb_fix_reserve_counts(struct inode *inode);
 124 extern struct mutex *hugetlb_fault_mutex_table;
 125 u32 hugetlb_fault_mutex_hash(struct hstate *h, struct mm_struct *mm,
 126                                 struct vm_area_struct *vma,
 127                                 struct address_space *mapping,
 128                                 pgoff_t idx, unsigned long address);
 129
 130 pte_t *huge_pmd_share(struct mm_struct *mm, unsigned long addr, pud_t *pud);
 131
 132 extern int hugepages_treat_as_movable;
 133 extern int sysctl_hugetlb_shm_group;
 134 extern struct list_head huge_boot_pages;
 135
 136 /* arch callbacks */
 137
 138 pte_t *huge_pte_alloc(struct mm_struct *mm,
 139                         unsigned long addr, unsigned long sz);
 140 pte_t *huge_pte_offset(struct mm_struct *mm,
 141                        unsigned long addr, unsigned long sz);
 142 int huge_pmd_unshare(struct mm_struct *mm, unsigned long *addr, pte_t *ptep);
 143 struct page *follow_huge_addr(struct mm_struct *mm, unsigned long address,
 144                               int write);
 145 struct page *follow_huge_pd(struct vm_area_struct *vma,
 146                             unsigned long address, hugepd_t hpd,
 147                             int flags, int pdshift);
 148 struct page *follow_huge_pmd(struct mm_struct *mm, unsigned long address,
 149                                 pmd_t *pmd, int flags);
 150 struct page *follow_huge_pud(struct mm_struct *mm, unsigned long address,
 151                                 pud_t *pud, int flags);
 152 struct page *follow_huge_pgd(struct mm_struct *mm, unsigned long address,
 153                              pgd_t *pgd, int flags);
 154
 155 int pmd_huge(pmd_t pmd);
 156 int pud_huge(pud_t pud);
 157 unsigned long hugetlb_change_protection(struct vm_area_struct *vma,
 158                 unsigned long address, unsigned long end, pgprot_t newprot);
 159
 160 bool is_hugetlb_entry_migration(pte_t pte);
 161 #else /* !CONFIG_HUGETLB_PAGE */
 162
 163 static inline void reset_vma_resv_huge_pages(struct vm_area_struct *vma)
 164 {
 165 }
 166
 167 static inline unsigned long hugetlb_total_pages(void)
 168 {
 169         return 0;
 170 }
 171
 172 #define follow_hugetlb_page(m,v,p,vs,a,b,i,w,n) ({ BUG(); 0; })
 173 #define follow_huge_addr(mm, addr, write)       ERR_PTR(-EINVAL)
 174 #define copy_hugetlb_page_range(src, dst, vma)  ({ BUG(); 0; })
 175 static inline void hugetlb_report_meminfo(struct seq_file *m)
 176 {
 177 }
 178 #define hugetlb_report_node_meminfo(n, buf)     0
 179 static inline void hugetlb_show_meminfo(void)
 180 {
 181 }
 182 #define follow_huge_pd(vma, addr, hpd, flags, pdshift) NULL
 183 #define follow_huge_pmd(mm, addr, pmd, flags)   NULL
 184 #define follow_huge_pud(mm, addr, pud, flags)   NULL
 185 #define follow_huge_pgd(mm, addr, pgd, flags)   NULL
 186 #define prepare_hugepage_range(file, addr, len) (-EINVAL)
 187 #define pmd_huge(x)     0
 188 #define pud_huge(x)     0
 189 #define is_hugepage_only_range(mm, addr, len)   0
 190 #define hugetlb_free_pgd_range(tlb, addr, end, floor, ceiling) ({BUG(); 0; })
 191 #define hugetlb_fault(mm, vma, addr, flags)     ({ BUG(); 0; })
 192 #define hugetlb_mcopy_atomic_pte(dst_mm, dst_pte, dst_vma, dst_addr, \
 193                                 src_addr, pagep)        ({ BUG(); 0; })
 194 #define huge_pte_offset(mm, address, sz)        0
 195
 196 static inline bool isolate_huge_page(struct page *page, struct list_head *list)
 197 {
 198         return false;
 199 }
 200 #define putback_active_hugepage(p)      do {} while (0)
 201
 202 static inline unsigned long hugetlb_change_protection(struct vm_area_struct *vma,
 203                 unsigned long address, unsigned long end, pgprot_t newprot)
 204 {
 205         return 0;
 206 }
 207
 208 static inline void __unmap_hugepage_range_final(struct mmu_gather *tlb,
 209                         struct vm_area_struct *vma, unsigned long start,
 210                         unsigned long end, struct page *ref_page)
 211 {
 212         BUG();
 213 }
 214
 215 static inline void __unmap_hugepage_range(struct mmu_gather *tlb,
 216                         struct vm_area_struct *vma, unsigned long start,
 217                         unsigned long end, struct page *ref_page)
 218 {
 219         BUG();
 220 }
 221
 222 #endif /* !CONFIG_HUGETLB_PAGE */
 223 /*
 224  * hugepages at page global directory. If arch support
 225  * hugepages at pgd level, they need to define this.
 226  */
 227 #ifndef pgd_huge
 228 #define pgd_huge(x)     0
 229 #endif
 230 #ifndef p4d_huge
 231 #define p4d_huge(x)     0
 232 #endif
 233
 234 #ifndef pgd_write
 235 static inline int pgd_write(pgd_t pgd)
 236 {
 237         BUG();
 238         return 0;
 239 }
 240 #endif
 241
 242 #define HUGETLB_ANON_FILE "anon_hugepage"
 243
 244 enum {
 245         /*
 246          * The file will be used as an shm file so shmfs accounting rules
 247          * apply
 248          */
 249         HUGETLB_SHMFS_INODE     = 1,
 250         /*
 251          * The file is being created on the internal vfs mount and shmfs
 252          * accounting rules do not apply
 253          */
 254         HUGETLB_ANONHUGE_INODE  = 2,
 255 };
 256
 257 #ifdef CONFIG_HUGETLBFS
 258 struct hugetlbfs_sb_info {
 259         long    max_inodes;   /* inodes allowed */
 260         long    free_inodes;  /* inodes free */
 261         spinlock_t      stat_lock;
 262         struct hstate *hstate;
 263         struct hugepage_subpool *spool;
 264         kuid_t  uid;
 265         kgid_t  gid;
 266         umode_t mode;
 267 };
 268
 269 static inline struct hugetlbfs_sb_info *HUGETLBFS_SB(struct super_block *sb)
 270 {
 271         return sb->s_fs_info;
 272 }
 273
 274 extern const struct file_operations hugetlbfs_file_operations;
 275 extern const struct vm_operations_struct hugetlb_vm_ops;
 276 struct file *hugetlb_file_setup(const char *name, size_t size, vm_flags_t acct,
 277                                 struct user_struct **user, int creat_flags,
 278                                 int page_size_log);
 279
 280 static inline bool is_file_hugepages(struct file *file)
 281 {
 282         if (file->f_op == &hugetlbfs_file_operations)
 283                 return true;
 284
 285         return is_file_shm_hugepages(file);
 286 }
 287
 288
 289 #else /* !CONFIG_HUGETLBFS */
 290
 291 #define is_file_hugepages(file)                 false
 292 static inline struct file *
 293 hugetlb_file_setup(const char *name, size_t size, vm_flags_t acctflag,
 294                 struct user_struct **user, int creat_flags,
 295                 int page_size_log)
 296 {
 297         return ERR_PTR(-ENOSYS);
 298 }
 299
 300 #endif /* !CONFIG_HUGETLBFS */
 301
 302 #ifdef HAVE_ARCH_HUGETLB_UNMAPPED_AREA
 303 unsigned long hugetlb_get_unmapped_area(struct file *file, unsigned long addr,
 304                                         unsigned long len, unsigned long pgoff,
 305                                         unsigned long flags);
 306 #endif /* HAVE_ARCH_HUGETLB_UNMAPPED_AREA */
 307
 308 #ifdef CONFIG_HUGETLB_PAGE
 309
 310 #define HSTATE_NAME_LEN 32
 311 /* Defines one hugetlb page size */
 312 struct hstate {
 313         int next_nid_to_alloc;
 314         int next_nid_to_free;
 315         unsigned int order;
 316         unsigned long mask;
 317         unsigned long max_huge_pages;
 318         unsigned long nr_huge_pages;
 319         unsigned long free_huge_pages;
 320         unsigned long resv_huge_pages;
 321         unsigned long surplus_huge_pages;
 322         unsigned long nr_overcommit_huge_pages;
 323         struct list_head hugepage_activelist;
 324         struct list_head hugepage_freelists[MAX_NUMNODES];
 325         unsigned int nr_huge_pages_node[MAX_NUMNODES];
 326         unsigned int free_huge_pages_node[MAX_NUMNODES];
 327         unsigned int surplus_huge_pages_node[MAX_NUMNODES];
 328 #ifdef CONFIG_CGROUP_HUGETLB
 329         /* cgroup control files */
 330         struct cftype cgroup_files[5];
 331 #endif
 332         char name[HSTATE_NAME_LEN];
 333 };
 334
 335 struct huge_bootmem_page {
 336         struct list_head list;
 337         struct hstate *hstate;
 338 #ifdef CONFIG_HIGHMEM
 339         phys_addr_t phys;
 340 #endif
 341 };
 342
 343 struct page *alloc_huge_page(struct vm_area_struct *vma,
 344                                 unsigned long addr, int avoid_reserve);
 345 struct page *alloc_huge_page_node(struct hstate *h, int nid);
 346 struct page *alloc_huge_page_noerr(struct vm_area_struct *vma,
 347                                 unsigned long addr, int avoid_reserve);
 348 struct page *alloc_huge_page_nodemask(struct hstate *h, int preferred_nid,
 349                                 nodemask_t *nmask);
 350 int huge_add_to_page_cache(struct page *page, struct address_space *mapping,
 351                         pgoff_t idx);
 352
 353 /* arch callback */
 354 int __init __alloc_bootmem_huge_page(struct hstate *h);
 355 int __init alloc_bootmem_huge_page(struct hstate *h);
 356
 357 void __init hugetlb_bad_size(void);
 358 void __init hugetlb_add_hstate(unsigned order);
 359 struct hstate *size_to_hstate(unsigned long size);
 360
 361 #ifndef HUGE_MAX_HSTATE
 362 #define HUGE_MAX_HSTATE 1
 363 #endif
 364
 365 extern struct hstate hstates[HUGE_MAX_HSTATE];
 366 extern unsigned int default_hstate_idx;
 367
 368 #define default_hstate (hstates[default_hstate_idx])
 369
 370 static inline struct hstate *hstate_inode(struct inode *i)
 371 {
 372         return HUGETLBFS_SB(i->i_sb)->hstate;
 373 }
 374
 375 static inline struct hstate *hstate_file(struct file *f)
 376 {
 377         return hstate_inode(file_inode(f));
 378 }
 379
 380 static inline struct hstate *hstate_sizelog(int page_size_log)
 381 {
 382         if (!page_size_log)
 383                 return &default_hstate;
 384
 385         return size_to_hstate(1UL << page_size_log);
 386 }
 387
 388 static inline struct hstate *hstate_vma(struct vm_area_struct *vma)
 389 {
 390         return hstate_file(vma->vm_file);
 391 }
 392
 393 static inline unsigned long huge_page_size(struct hstate *h)
 394 {
 395         return (unsigned long)PAGE_SIZE << h->order;
 396 }
 397
 398 extern unsigned long vma_kernel_pagesize(struct vm_area_struct *vma);
 399
 400 extern unsigned long vma_mmu_pagesize(struct vm_area_struct *vma);
 401
 402 static inline unsigned long huge_page_mask(struct hstate *h)
 403 {
 404         return h->mask;
 405 }
 406
 407 static inline unsigned int huge_page_order(struct hstate *h)
 408 {
 409         return h->order;
 410 }
 411
 412 static inline unsigned huge_page_shift(struct hstate *h)
 413 {
 414         return h->order + PAGE_SHIFT;
 415 }
 416
 417 static inline bool hstate_is_gigantic(struct hstate *h)
 418 {
 419         return huge_page_order(h) >= MAX_ORDER;
 420 }
 421
 422 static inline unsigned int pages_per_huge_page(struct hstate *h)
 423 {
 424         return 1 << h->order;
 425 }
 426
 427 static inline unsigned int blocks_per_huge_page(struct hstate *h)
 428 {
 429         return huge_page_size(h) / 512;
 430 }
 431
 432 #include <asm/hugetlb.h>
 433
 434 #ifndef arch_make_huge_pte
 435 static inline pte_t arch_make_huge_pte(pte_t entry, struct vm_area_struct *vma,
 436                                        struct page *page, int writable)
 437 {
 438         return entry;
 439 }
 440 #endif
 441
 442 static inline struct hstate *page_hstate(struct page *page)
 443 {
 444         VM_BUG_ON_PAGE(!PageHuge(page), page);
 445         return size_to_hstate(PAGE_SIZE << compound_order(page));
 446 }
 447
 448 static inline unsigned hstate_index_to_shift(unsigned index)
 449 {
 450         return hstates[index].order + PAGE_SHIFT;
 451 }
 452
 453 static inline int hstate_index(struct hstate *h)
 454 {
 455         return h - hstates;
 456 }
 457
 458 pgoff_t __basepage_index(struct page *page);
 459
 460 /* Return page->index in PAGE_SIZE units */
 461 static inline pgoff_t basepage_index(struct page *page)
 462 {
 463         if (!PageCompound(page))
 464                 return page->index;
 465
 466         return __basepage_index(page);
 467 }
 468
 469 extern int dissolve_free_huge_page(struct page *page);
 470 extern int dissolve_free_huge_pages(unsigned long start_pfn,
 471                                     unsigned long end_pfn);
 472 static inline bool hugepage_migration_supported(struct hstate *h)
 473 {
 474 #ifdef CONFIG_ARCH_ENABLE_HUGEPAGE_MIGRATION
 475         if ((huge_page_shift(h) == PMD_SHIFT) ||
 476                 (huge_page_shift(h) == PGDIR_SHIFT))
 477                 return true;
 478         else
 479                 return false;
 480 #else
 481         return false;
 482 #endif
 483 }
 484
 485 static inline spinlock_t *huge_pte_lockptr(struct hstate *h,
 486                                            struct mm_struct *mm, pte_t *pte)
 487 {
 488         if (huge_page_size(h) == PMD_SIZE)
 489                 return pmd_lockptr(mm, (pmd_t *) pte);
 490         VM_BUG_ON(huge_page_size(h) == PAGE_SIZE);
 491         return &mm->page_table_lock;
 492 }
 493
 494 #ifndef hugepages_supported
 495 /*
 496  * Some platform decide whether they support huge pages at boot
 497  * time. Some of them, such as powerpc, set HPAGE_SHIFT to 0
 498  * when there is no such support
 499  */
 500 #define hugepages_supported() (HPAGE_SHIFT != 0)
 501 #endif
 502
 503 void hugetlb_report_usage(struct seq_file *m, struct mm_struct *mm);
 504
 505 static inline void hugetlb_count_add(long l, struct mm_struct *mm)
 506 {
 507         atomic_long_add(l, &mm->hugetlb_usage);
 508 }
 509
 510 static inline void hugetlb_count_sub(long l, struct mm_struct *mm)
 511 {
 512         atomic_long_sub(l, &mm->hugetlb_usage);
 513 }
 514
 515 #ifndef set_huge_swap_pte_at
 516 static inline void set_huge_swap_pte_at(struct mm_struct *mm, unsigned long addr,
 517                                         pte_t *ptep, pte_t pte, unsigned long sz)
 518 {
 519         set_huge_pte_at(mm, addr, ptep, pte);
 520 }
 521 #endif
 522 #else   /* CONFIG_HUGETLB_PAGE */
 523 struct hstate {};
 524 #define alloc_huge_page(v, a, r) NULL
 525 #define alloc_huge_page_node(h, nid) NULL
 526 #define alloc_huge_page_nodemask(h, preferred_nid, nmask) NULL
 527 #define alloc_huge_page_noerr(v, a, r) NULL
 528 #define alloc_bootmem_huge_page(h) NULL
 529 #define hstate_file(f) NULL
 530 #define hstate_sizelog(s) NULL
 531 #define hstate_vma(v) NULL
 532 #define hstate_inode(i) NULL
 533 #define page_hstate(page) NULL
 534 #define huge_page_size(h) PAGE_SIZE
 535 #define huge_page_mask(h) PAGE_MASK
 536 #define vma_kernel_pagesize(v) PAGE_SIZE
 537 #define vma_mmu_pagesize(v) PAGE_SIZE
 538 #define huge_page_order(h) 0
 539 #define huge_page_shift(h) PAGE_SHIFT
 540 static inline bool hstate_is_gigantic(struct hstate *h)
 541 {
 542         return false;
 543 }
 544
 545 static inline unsigned int pages_per_huge_page(struct hstate *h)
 546 {
 547         return 1;
 548 }
 549
 550 static inline unsigned hstate_index_to_shift(unsigned index)
 551 {
 552         return 0;
 553 }
 554
 555 static inline int hstate_index(struct hstate *h)
 556 {
 557         return 0;
 558 }
 559
 560 static inline pgoff_t basepage_index(struct page *page)
 561 {
 562         return page->index;
 563 }
 564
 565 static inline int dissolve_free_huge_page(struct page *page)
 566 {
 567         return 0;
 568 }
 569
 570 static inline int dissolve_free_huge_pages(unsigned long start_pfn,
 571                                            unsigned long end_pfn)
 572 {
 573         return 0;
 574 }
 575
 576 static inline bool hugepage_migration_supported(struct hstate *h)
 577 {
 578         return false;
 579 }
 580
 581 static inline spinlock_t *huge_pte_lockptr(struct hstate *h,
 582                                            struct mm_struct *mm, pte_t *pte)
 583 {
 584         return &mm->page_table_lock;
 585 }
 586
 587 static inline void hugetlb_report_usage(struct seq_file *f, struct mm_struct *m)
 588 {
 589 }
 590
 591 static inline void hugetlb_count_sub(long l, struct mm_struct *mm)
 592 {
 593 }
 594
 595 static inline void set_huge_swap_pte_at(struct mm_struct *mm, unsigned long addr,
 596                                         pte_t *ptep, pte_t pte, unsigned long sz)
 597 {
 598 }
 599 #endif  /* CONFIG_HUGETLB_PAGE */
 600
 601 static inline spinlock_t *huge_pte_lock(struct hstate *h,
 602                                         struct mm_struct *mm, pte_t *pte)
 603 {
 604         spinlock_t *ptl;
 605
 606         ptl = huge_pte_lockptr(h, mm, pte);
 607         spin_lock(ptl);
 608         return ptl;
 609 }
 610
 611 #endif /* _LINUX_HUGETLB_H */