include/linux/hugetlb.h

   1 #ifndef _LINUX_HUGETLB_H
   2 #define _LINUX_HUGETLB_H
   3
   4 #include <linux/mm_types.h>
   5 #include <linux/mmdebug.h>
   6 #include <linux/fs.h>
   7 #include <linux/hugetlb_inline.h>
   8 #include <linux/cgroup.h>
   9 #include <linux/list.h>
  10 #include <linux/kref.h>
  11 #include <asm/pgtable.h>
  12
  13 struct ctl_table;
  14 struct user_struct;
  15 struct mmu_gather;
  16
  17 #ifdef CONFIG_HUGETLB_PAGE
  18
  19 #include <linux/mempolicy.h>
  20 #include <linux/shm.h>
  21 #include <asm/tlbflush.h>
  22
  23 struct hugepage_subpool {
  24         spinlock_t lock;
  25         long count;
  26         long max_hpages;        /* Maximum huge pages or -1 if no maximum. */
  27         long used_hpages;       /* Used count against maximum, includes */
  28                                 /* both alloced and reserved pages. */
  29         struct hstate *hstate;
  30         long min_hpages;        /* Minimum huge pages or -1 if no minimum. */
  31         long rsv_hpages;        /* Pages reserved against global pool to */
  32                                 /* sasitfy minimum size. */
  33 };
  34
  35 struct resv_map {
  36         struct kref refs;
  37         spinlock_t lock;
  38         struct list_head regions;
  39         long adds_in_progress;
  40         struct list_head region_cache;
  41         long region_cache_count;
  42 };
  43 extern struct resv_map *resv_map_alloc(void);
  44 void resv_map_release(struct kref *ref);
  45
  46 extern spinlock_t hugetlb_lock;
  47 extern int hugetlb_max_hstate __read_mostly;
  48 #define for_each_hstate(h) \
  49         for ((h) = hstates; (h) < &hstates[hugetlb_max_hstate]; (h)++)
  50
  51 struct hugepage_subpool *hugepage_new_subpool(struct hstate *h, long max_hpages,
  52                                                 long min_hpages);
  53 void hugepage_put_subpool(struct hugepage_subpool *spool);
  54
  55 void reset_vma_resv_huge_pages(struct vm_area_struct *vma);
  56 int hugetlb_sysctl_handler(struct ctl_table *, int, void __user *, size_t *, loff_t *);
  57 int hugetlb_overcommit_handler(struct ctl_table *, int, void __user *, size_t *, loff_t *);
  58 int hugetlb_treat_movable_handler(struct ctl_table *, int, void __user *, size_t *, loff_t *);
  59
  60 #ifdef CONFIG_NUMA
  61 int hugetlb_mempolicy_sysctl_handler(struct ctl_table *, int,
  62                                         void __user *, size_t *, loff_t *);
  63 #endif
  64
  65 int copy_hugetlb_page_range(struct mm_struct *, struct mm_struct *, struct vm_area_struct *);
  66 long follow_hugetlb_page(struct mm_struct *, struct vm_area_struct *,
  67                          struct page **, struct vm_area_struct **,
  68                          unsigned long *, unsigned long *, long, unsigned int);
  69 void unmap_hugepage_range(struct vm_area_struct *,
  70                           unsigned long, unsigned long, struct page *);
  71 void __unmap_hugepage_range_final(struct mmu_gather *tlb,
  72                           struct vm_area_struct *vma,
  73                           unsigned long start, unsigned long end,
  74                           struct page *ref_page);
  75 void __unmap_hugepage_range(struct mmu_gather *tlb, struct vm_area_struct *vma,
  76                                 unsigned long start, unsigned long end,
  77                                 struct page *ref_page);
  78 void hugetlb_report_meminfo(struct seq_file *);
  79 int hugetlb_report_node_meminfo(int, char *);
  80 void hugetlb_show_meminfo(void);
  81 unsigned long hugetlb_total_pages(void);
  82 int hugetlb_fault(struct mm_struct *mm, struct vm_area_struct *vma,
  83                         unsigned long address, unsigned int flags);
  84 int hugetlb_reserve_pages(struct inode *inode, long from, long to,
  85                                                 struct vm_area_struct *vma,
  86                                                 vm_flags_t vm_flags);
  87 long hugetlb_unreserve_pages(struct inode *inode, long start, long end,
  88                                                 long freed);
  89 int dequeue_hwpoisoned_huge_page(struct page *page);
  90 bool isolate_huge_page(struct page *page, struct list_head *list);
  91 void putback_active_hugepage(struct page *page);
  92 void free_huge_page(struct page *page);
  93 void hugetlb_fix_reserve_counts(struct inode *inode, bool restore_reserve);
  94 extern struct mutex *hugetlb_fault_mutex_table;
  95 u32 hugetlb_fault_mutex_hash(struct hstate *h, struct mm_struct *mm,
  96                                 struct vm_area_struct *vma,
  97                                 struct address_space *mapping,
  98                                 pgoff_t idx, unsigned long address);
  99
 100 pte_t *huge_pmd_share(struct mm_struct *mm, unsigned long addr, pud_t *pud);
 101
 102 extern int hugepages_treat_as_movable;
 103 extern int sysctl_hugetlb_shm_group;
 104 extern struct list_head huge_boot_pages;
 105
 106 /* arch callbacks */
 107
 108 pte_t *huge_pte_alloc(struct mm_struct *mm,
 109                         unsigned long addr, unsigned long sz);
 110 pte_t *huge_pte_offset(struct mm_struct *mm, unsigned long addr);
 111 int huge_pmd_unshare(struct mm_struct *mm, unsigned long *addr, pte_t *ptep);
 112 struct page *follow_huge_addr(struct mm_struct *mm, unsigned long address,
 113                               int write);
 114 struct page *follow_huge_pmd(struct mm_struct *mm, unsigned long address,
 115                                 pmd_t *pmd, int flags);
 116 struct page *follow_huge_pud(struct mm_struct *mm, unsigned long address,
 117                                 pud_t *pud, int flags);
 118 int pmd_huge(pmd_t pmd);
 119 int pud_huge(pud_t pmd);
 120 unsigned long hugetlb_change_protection(struct vm_area_struct *vma,
 121                 unsigned long address, unsigned long end, pgprot_t newprot);
 122
 123 #else /* !CONFIG_HUGETLB_PAGE */
 124
 125 static inline void reset_vma_resv_huge_pages(struct vm_area_struct *vma)
 126 {
 127 }
 128
 129 static inline unsigned long hugetlb_total_pages(void)
 130 {
 131         return 0;
 132 }
 133
 134 #define follow_hugetlb_page(m,v,p,vs,a,b,i,w)   ({ BUG(); 0; })
 135 #define follow_huge_addr(mm, addr, write)       ERR_PTR(-EINVAL)
 136 #define copy_hugetlb_page_range(src, dst, vma)  ({ BUG(); 0; })
 137 static inline void hugetlb_report_meminfo(struct seq_file *m)
 138 {
 139 }
 140 #define hugetlb_report_node_meminfo(n, buf)     0
 141 static inline void hugetlb_show_meminfo(void)
 142 {
 143 }
 144 #define follow_huge_pmd(mm, addr, pmd, flags)   NULL
 145 #define follow_huge_pud(mm, addr, pud, flags)   NULL
 146 #define prepare_hugepage_range(file, addr, len) (-EINVAL)
 147 #define pmd_huge(x)     0
 148 #define pud_huge(x)     0
 149 #define is_hugepage_only_range(mm, addr, len)   0
 150 #define hugetlb_free_pgd_range(tlb, addr, end, floor, ceiling) ({BUG(); 0; })
 151 #define hugetlb_fault(mm, vma, addr, flags)     ({ BUG(); 0; })
 152 #define huge_pte_offset(mm, address)    0
 153 static inline int dequeue_hwpoisoned_huge_page(struct page *page)
 154 {
 155         return 0;
 156 }
 157
 158 static inline bool isolate_huge_page(struct page *page, struct list_head *list)
 159 {
 160         return false;
 161 }
 162 #define putback_active_hugepage(p)      do {} while (0)
 163
 164 static inline unsigned long hugetlb_change_protection(struct vm_area_struct *vma,
 165                 unsigned long address, unsigned long end, pgprot_t newprot)
 166 {
 167         return 0;
 168 }
 169
 170 static inline void __unmap_hugepage_range_final(struct mmu_gather *tlb,
 171                         struct vm_area_struct *vma, unsigned long start,
 172                         unsigned long end, struct page *ref_page)
 173 {
 174         BUG();
 175 }
 176
 177 static inline void __unmap_hugepage_range(struct mmu_gather *tlb,
 178                         struct vm_area_struct *vma, unsigned long start,
 179                         unsigned long end, struct page *ref_page)
 180 {
 181         BUG();
 182 }
 183
 184 #endif /* !CONFIG_HUGETLB_PAGE */
 185 /*
 186  * hugepages at page global directory. If arch support
 187  * hugepages at pgd level, they need to define this.
 188  */
 189 #ifndef pgd_huge
 190 #define pgd_huge(x)     0
 191 #endif
 192
 193 #ifndef pgd_write
 194 static inline int pgd_write(pgd_t pgd)
 195 {
 196         BUG();
 197         return 0;
 198 }
 199 #endif
 200
 201 #ifndef pud_write
 202 static inline int pud_write(pud_t pud)
 203 {
 204         BUG();
 205         return 0;
 206 }
 207 #endif
 208
 209 #ifndef is_hugepd
 210 /*
 211  * Some architectures requires a hugepage directory format that is
 212  * required to support multiple hugepage sizes. For example
 213  * a4fe3ce76 "powerpc/mm: Allow more flexible layouts for hugepage pagetables"
 214  * introduced the same on powerpc. This allows for a more flexible hugepage
 215  * pagetable layout.
 216  */
 217 typedef struct { unsigned long pd; } hugepd_t;
 218 #define is_hugepd(hugepd) (0)
 219 #define __hugepd(x) ((hugepd_t) { (x) })
 220 static inline int gup_huge_pd(hugepd_t hugepd, unsigned long addr,
 221                               unsigned pdshift, unsigned long end,
 222                               int write, struct page **pages, int *nr)
 223 {
 224         return 0;
 225 }
 226 #else
 227 extern int gup_huge_pd(hugepd_t hugepd, unsigned long addr,
 228                        unsigned pdshift, unsigned long end,
 229                        int write, struct page **pages, int *nr);
 230 #endif
 231
 232 #define HUGETLB_ANON_FILE "anon_hugepage"
 233
 234 enum {
 235         /*
 236          * The file will be used as an shm file so shmfs accounting rules
 237          * apply
 238          */
 239         HUGETLB_SHMFS_INODE     = 1,
 240         /*
 241          * The file is being created on the internal vfs mount and shmfs
 242          * accounting rules do not apply
 243          */
 244         HUGETLB_ANONHUGE_INODE  = 2,
 245 };
 246
 247 #ifdef CONFIG_HUGETLBFS
 248 struct hugetlbfs_sb_info {
 249         long    max_inodes;   /* inodes allowed */
 250         long    free_inodes;  /* inodes free */
 251         spinlock_t      stat_lock;
 252         struct hstate *hstate;
 253         struct hugepage_subpool *spool;
 254 };
 255
 256 static inline struct hugetlbfs_sb_info *HUGETLBFS_SB(struct super_block *sb)
 257 {
 258         return sb->s_fs_info;
 259 }
 260
 261 extern const struct file_operations hugetlbfs_file_operations;
 262 extern const struct vm_operations_struct hugetlb_vm_ops;
 263 struct file *hugetlb_file_setup(const char *name, size_t size, vm_flags_t acct,
 264                                 struct user_struct **user, int creat_flags,
 265                                 int page_size_log);
 266
 267 static inline bool is_file_hugepages(struct file *file)
 268 {
 269         if (file->f_op == &hugetlbfs_file_operations)
 270                 return true;
 271
 272         return is_file_shm_hugepages(file);
 273 }
 274
 275
 276 #else /* !CONFIG_HUGETLBFS */
 277
 278 #define is_file_hugepages(file)                 false
 279 static inline struct file *
 280 hugetlb_file_setup(const char *name, size_t size, vm_flags_t acctflag,
 281                 struct user_struct **user, int creat_flags,
 282                 int page_size_log)
 283 {
 284         return ERR_PTR(-ENOSYS);
 285 }
 286
 287 #endif /* !CONFIG_HUGETLBFS */
 288
 289 #ifdef HAVE_ARCH_HUGETLB_UNMAPPED_AREA
 290 unsigned long hugetlb_get_unmapped_area(struct file *file, unsigned long addr,
 291                                         unsigned long len, unsigned long pgoff,
 292                                         unsigned long flags);
 293 #endif /* HAVE_ARCH_HUGETLB_UNMAPPED_AREA */
 294
 295 #ifdef CONFIG_HUGETLB_PAGE
 296
 297 #define HSTATE_NAME_LEN 32
 298 /* Defines one hugetlb page size */
 299 struct hstate {
 300         int next_nid_to_alloc;
 301         int next_nid_to_free;
 302         unsigned int order;
 303         unsigned long mask;
 304         unsigned long max_huge_pages;
 305         unsigned long nr_huge_pages;
 306         unsigned long free_huge_pages;
 307         unsigned long resv_huge_pages;
 308         unsigned long surplus_huge_pages;
 309         unsigned long nr_overcommit_huge_pages;
 310         struct list_head hugepage_activelist;
 311         struct list_head hugepage_freelists[MAX_NUMNODES];
 312         unsigned int nr_huge_pages_node[MAX_NUMNODES];
 313         unsigned int free_huge_pages_node[MAX_NUMNODES];
 314         unsigned int surplus_huge_pages_node[MAX_NUMNODES];
 315 #ifdef CONFIG_CGROUP_HUGETLB
 316         /* cgroup control files */
 317         struct cftype cgroup_files[5];
 318 #endif
 319         char name[HSTATE_NAME_LEN];
 320 };
 321
 322 struct huge_bootmem_page {
 323         struct list_head list;
 324         struct hstate *hstate;
 325 #ifdef CONFIG_HIGHMEM
 326         phys_addr_t phys;
 327 #endif
 328 };
 329
 330 struct page *alloc_huge_page(struct vm_area_struct *vma,
 331                                 unsigned long addr, int avoid_reserve);
 332 struct page *alloc_huge_page_node(struct hstate *h, int nid);
 333 struct page *alloc_huge_page_noerr(struct vm_area_struct *vma,
 334                                 unsigned long addr, int avoid_reserve);
 335 int huge_add_to_page_cache(struct page *page, struct address_space *mapping,
 336                         pgoff_t idx);
 337
 338 /* arch callback */
 339 int __init alloc_bootmem_huge_page(struct hstate *h);
 340
 341 void __init hugetlb_bad_size(void);
 342 void __init hugetlb_add_hstate(unsigned order);
 343 struct hstate *size_to_hstate(unsigned long size);
 344
 345 #ifndef HUGE_MAX_HSTATE
 346 #define HUGE_MAX_HSTATE 1
 347 #endif
 348
 349 extern struct hstate hstates[HUGE_MAX_HSTATE];
 350 extern unsigned int default_hstate_idx;
 351
 352 #define default_hstate (hstates[default_hstate_idx])
 353
 354 static inline struct hstate *hstate_inode(struct inode *i)
 355 {
 356         return HUGETLBFS_SB(i->i_sb)->hstate;
 357 }
 358
 359 static inline struct hstate *hstate_file(struct file *f)
 360 {
 361         return hstate_inode(file_inode(f));
 362 }
 363
 364 static inline struct hstate *hstate_sizelog(int page_size_log)
 365 {
 366         if (!page_size_log)
 367                 return &default_hstate;
 368
 369         return size_to_hstate(1UL << page_size_log);
 370 }
 371
 372 static inline struct hstate *hstate_vma(struct vm_area_struct *vma)
 373 {
 374         return hstate_file(vma->vm_file);
 375 }
 376
 377 static inline unsigned long huge_page_size(struct hstate *h)
 378 {
 379         return (unsigned long)PAGE_SIZE << h->order;
 380 }
 381
 382 extern unsigned long vma_kernel_pagesize(struct vm_area_struct *vma);
 383
 384 extern unsigned long vma_mmu_pagesize(struct vm_area_struct *vma);
 385
 386 static inline unsigned long huge_page_mask(struct hstate *h)
 387 {
 388         return h->mask;
 389 }
 390
 391 static inline unsigned int huge_page_order(struct hstate *h)
 392 {
 393         return h->order;
 394 }
 395
 396 static inline unsigned huge_page_shift(struct hstate *h)
 397 {
 398         return h->order + PAGE_SHIFT;
 399 }
 400
 401 static inline bool hstate_is_gigantic(struct hstate *h)
 402 {
 403         return huge_page_order(h) >= MAX_ORDER;
 404 }
 405
 406 static inline unsigned int pages_per_huge_page(struct hstate *h)
 407 {
 408         return 1 << h->order;
 409 }
 410
 411 static inline unsigned int blocks_per_huge_page(struct hstate *h)
 412 {
 413         return huge_page_size(h) / 512;
 414 }
 415
 416 #include <asm/hugetlb.h>
 417
 418 #ifndef arch_make_huge_pte
 419 static inline pte_t arch_make_huge_pte(pte_t entry, struct vm_area_struct *vma,
 420                                        struct page *page, int writable)
 421 {
 422         return entry;
 423 }
 424 #endif
 425
 426 static inline struct hstate *page_hstate(struct page *page)
 427 {
 428         VM_BUG_ON_PAGE(!PageHuge(page), page);
 429         return size_to_hstate(PAGE_SIZE << compound_order(page));
 430 }
 431
 432 static inline unsigned hstate_index_to_shift(unsigned index)
 433 {
 434         return hstates[index].order + PAGE_SHIFT;
 435 }
 436
 437 static inline int hstate_index(struct hstate *h)
 438 {
 439         return h - hstates;
 440 }
 441
 442 pgoff_t __basepage_index(struct page *page);
 443
 444 /* Return page->index in PAGE_SIZE units */
 445 static inline pgoff_t basepage_index(struct page *page)
 446 {
 447         if (!PageCompound(page))
 448                 return page->index;
 449
 450         return __basepage_index(page);
 451 }
 452
 453 extern void dissolve_free_huge_pages(unsigned long start_pfn,
 454                                      unsigned long end_pfn);
 455 static inline bool hugepage_migration_supported(struct hstate *h)
 456 {
 457 #ifdef CONFIG_ARCH_ENABLE_HUGEPAGE_MIGRATION
 458         return huge_page_shift(h) == PMD_SHIFT;
 459 #else
 460         return false;
 461 #endif
 462 }
 463
 464 static inline spinlock_t *huge_pte_lockptr(struct hstate *h,
 465                                            struct mm_struct *mm, pte_t *pte)
 466 {
 467         if (huge_page_size(h) == PMD_SIZE)
 468                 return pmd_lockptr(mm, (pmd_t *) pte);
 469         VM_BUG_ON(huge_page_size(h) == PAGE_SIZE);
 470         return &mm->page_table_lock;
 471 }
 472
 473 #ifndef hugepages_supported
 474 /*
 475  * Some platform decide whether they support huge pages at boot
 476  * time. Some of them, such as powerpc, set HPAGE_SHIFT to 0
 477  * when there is no such support
 478  */
 479 #define hugepages_supported() (HPAGE_SHIFT != 0)
 480 #endif
 481
 482 void hugetlb_report_usage(struct seq_file *m, struct mm_struct *mm);
 483
 484 static inline void hugetlb_count_add(long l, struct mm_struct *mm)
 485 {
 486         atomic_long_add(l, &mm->hugetlb_usage);
 487 }
 488
 489 static inline void hugetlb_count_sub(long l, struct mm_struct *mm)
 490 {
 491         atomic_long_sub(l, &mm->hugetlb_usage);
 492 }
 493 #else   /* CONFIG_HUGETLB_PAGE */
 494 struct hstate {};
 495 #define alloc_huge_page(v, a, r) NULL
 496 #define alloc_huge_page_node(h, nid) NULL
 497 #define alloc_huge_page_noerr(v, a, r) NULL
 498 #define alloc_bootmem_huge_page(h) NULL
 499 #define hstate_file(f) NULL
 500 #define hstate_sizelog(s) NULL
 501 #define hstate_vma(v) NULL
 502 #define hstate_inode(i) NULL
 503 #define page_hstate(page) NULL
 504 #define huge_page_size(h) PAGE_SIZE
 505 #define huge_page_mask(h) PAGE_MASK
 506 #define vma_kernel_pagesize(v) PAGE_SIZE
 507 #define vma_mmu_pagesize(v) PAGE_SIZE
 508 #define huge_page_order(h) 0
 509 #define huge_page_shift(h) PAGE_SHIFT
 510 static inline unsigned int pages_per_huge_page(struct hstate *h)
 511 {
 512         return 1;
 513 }
 514 #define hstate_index_to_shift(index) 0
 515 #define hstate_index(h) 0
 516
 517 static inline pgoff_t basepage_index(struct page *page)
 518 {
 519         return page->index;
 520 }
 521 #define dissolve_free_huge_pages(s, e)  do {} while (0)
 522 #define hugepage_migration_supported(h) false
 523
 524 static inline spinlock_t *huge_pte_lockptr(struct hstate *h,
 525                                            struct mm_struct *mm, pte_t *pte)
 526 {
 527         return &mm->page_table_lock;
 528 }
 529
 530 static inline void hugetlb_report_usage(struct seq_file *f, struct mm_struct *m)
 531 {
 532 }
 533
 534 static inline void hugetlb_count_sub(long l, struct mm_struct *mm)
 535 {
 536 }
 537 #endif  /* CONFIG_HUGETLB_PAGE */
 538
 539 static inline spinlock_t *huge_pte_lock(struct hstate *h,
 540                                         struct mm_struct *mm, pte_t *pte)
 541 {
 542         spinlock_t *ptl;
 543
 544         ptl = huge_pte_lockptr(h, mm, pte);
 545         spin_lock(ptl);
 546         return ptl;
 547 }
 548
 549 #endif /* _LINUX_HUGETLB_H */