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