hugetlb.h 15 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570
  1. #ifndef _LINUX_HUGETLB_H
  2. #define _LINUX_HUGETLB_H
  3. #include <linux/mm_types.h>
  4. #include <linux/mmdebug.h>
  5. #include <linux/fs.h>
  6. #include <linux/hugetlb_inline.h>
  7. #include <linux/cgroup.h>
  8. #include <linux/list.h>
  9. #include <linux/kref.h>
  10. #include <asm/pgtable.h>
  11. struct ctl_table;
  12. struct user_struct;
  13. struct mmu_gather;
  14. #ifdef CONFIG_HUGETLB_PAGE
  15. #include <linux/mempolicy.h>
  16. #include <linux/shm.h>
  17. #include <asm/tlbflush.h>
  18. struct hugepage_subpool {
  19. spinlock_t lock;
  20. long count;
  21. long max_hpages; /* Maximum huge pages or -1 if no maximum. */
  22. long used_hpages; /* Used count against maximum, includes */
  23. /* both alloced and reserved pages. */
  24. struct hstate *hstate;
  25. long min_hpages; /* Minimum huge pages or -1 if no minimum. */
  26. long rsv_hpages; /* Pages reserved against global pool to */
  27. /* sasitfy minimum size. */
  28. };
  29. struct resv_map {
  30. struct kref refs;
  31. spinlock_t lock;
  32. struct list_head regions;
  33. long adds_in_progress;
  34. struct list_head region_cache;
  35. long region_cache_count;
  36. };
  37. extern struct resv_map *resv_map_alloc(void);
  38. void resv_map_release(struct kref *ref);
  39. extern spinlock_t hugetlb_lock;
  40. extern int hugetlb_max_hstate __read_mostly;
  41. #define for_each_hstate(h) \
  42. for ((h) = hstates; (h) < &hstates[hugetlb_max_hstate]; (h)++)
  43. struct hugepage_subpool *hugepage_new_subpool(struct hstate *h, long max_hpages,
  44. long min_hpages);
  45. void hugepage_put_subpool(struct hugepage_subpool *spool);
  46. void reset_vma_resv_huge_pages(struct vm_area_struct *vma);
  47. int hugetlb_sysctl_handler(struct ctl_table *, int, void __user *, size_t *, loff_t *);
  48. int hugetlb_overcommit_handler(struct ctl_table *, int, void __user *, size_t *, loff_t *);
  49. int hugetlb_treat_movable_handler(struct ctl_table *, int, void __user *, size_t *, loff_t *);
  50. #ifdef CONFIG_NUMA
  51. int hugetlb_mempolicy_sysctl_handler(struct ctl_table *, int,
  52. void __user *, size_t *, loff_t *);
  53. #endif
  54. int copy_hugetlb_page_range(struct mm_struct *, struct mm_struct *, struct vm_area_struct *);
  55. long follow_hugetlb_page(struct mm_struct *, struct vm_area_struct *,
  56. struct page **, struct vm_area_struct **,
  57. unsigned long *, unsigned long *, long, unsigned int,
  58. int *);
  59. void unmap_hugepage_range(struct vm_area_struct *,
  60. unsigned long, unsigned long, struct page *);
  61. void __unmap_hugepage_range_final(struct mmu_gather *tlb,
  62. struct vm_area_struct *vma,
  63. unsigned long start, unsigned long end,
  64. struct page *ref_page);
  65. void __unmap_hugepage_range(struct mmu_gather *tlb, struct vm_area_struct *vma,
  66. unsigned long start, unsigned long end,
  67. struct page *ref_page);
  68. void hugetlb_report_meminfo(struct seq_file *);
  69. int hugetlb_report_node_meminfo(int, char *);
  70. void hugetlb_show_meminfo(void);
  71. unsigned long hugetlb_total_pages(void);
  72. int hugetlb_fault(struct mm_struct *mm, struct vm_area_struct *vma,
  73. unsigned long address, unsigned int flags);
  74. int hugetlb_mcopy_atomic_pte(struct mm_struct *dst_mm, pte_t *dst_pte,
  75. struct vm_area_struct *dst_vma,
  76. unsigned long dst_addr,
  77. unsigned long src_addr,
  78. struct page **pagep);
  79. int hugetlb_reserve_pages(struct inode *inode, long from, long to,
  80. struct vm_area_struct *vma,
  81. vm_flags_t vm_flags);
  82. long hugetlb_unreserve_pages(struct inode *inode, long start, long end,
  83. long freed);
  84. int dequeue_hwpoisoned_huge_page(struct page *page);
  85. bool isolate_huge_page(struct page *page, struct list_head *list);
  86. void putback_active_hugepage(struct page *page);
  87. void free_huge_page(struct page *page);
  88. void hugetlb_fix_reserve_counts(struct inode *inode);
  89. extern struct mutex *hugetlb_fault_mutex_table;
  90. u32 hugetlb_fault_mutex_hash(struct hstate *h, struct mm_struct *mm,
  91. struct vm_area_struct *vma,
  92. struct address_space *mapping,
  93. pgoff_t idx, unsigned long address);
  94. pte_t *huge_pmd_share(struct mm_struct *mm, unsigned long addr, pud_t *pud);
  95. extern int hugepages_treat_as_movable;
  96. extern int sysctl_hugetlb_shm_group;
  97. extern struct list_head huge_boot_pages;
  98. /* arch callbacks */
  99. pte_t *huge_pte_alloc(struct mm_struct *mm,
  100. unsigned long addr, unsigned long sz);
  101. pte_t *huge_pte_offset(struct mm_struct *mm, unsigned long addr);
  102. int huge_pmd_unshare(struct mm_struct *mm, unsigned long *addr, pte_t *ptep);
  103. struct page *follow_huge_addr(struct mm_struct *mm, unsigned long address,
  104. int write);
  105. struct page *follow_huge_pmd(struct mm_struct *mm, unsigned long address,
  106. pmd_t *pmd, int flags);
  107. struct page *follow_huge_pud(struct mm_struct *mm, unsigned long address,
  108. pud_t *pud, int flags);
  109. int pmd_huge(pmd_t pmd);
  110. int pud_huge(pud_t pud);
  111. unsigned long hugetlb_change_protection(struct vm_area_struct *vma,
  112. unsigned long address, unsigned long end, pgprot_t newprot);
  113. bool is_hugetlb_entry_migration(pte_t pte);
  114. #else /* !CONFIG_HUGETLB_PAGE */
  115. static inline void reset_vma_resv_huge_pages(struct vm_area_struct *vma)
  116. {
  117. }
  118. static inline unsigned long hugetlb_total_pages(void)
  119. {
  120. return 0;
  121. }
  122. #define follow_hugetlb_page(m,v,p,vs,a,b,i,w,n) ({ BUG(); 0; })
  123. #define follow_huge_addr(mm, addr, write) ERR_PTR(-EINVAL)
  124. #define copy_hugetlb_page_range(src, dst, vma) ({ BUG(); 0; })
  125. static inline void hugetlb_report_meminfo(struct seq_file *m)
  126. {
  127. }
  128. #define hugetlb_report_node_meminfo(n, buf) 0
  129. static inline void hugetlb_show_meminfo(void)
  130. {
  131. }
  132. #define follow_huge_pmd(mm, addr, pmd, flags) NULL
  133. #define follow_huge_pud(mm, addr, pud, flags) NULL
  134. #define prepare_hugepage_range(file, addr, len) (-EINVAL)
  135. #define pmd_huge(x) 0
  136. #define pud_huge(x) 0
  137. #define is_hugepage_only_range(mm, addr, len) 0
  138. #define hugetlb_free_pgd_range(tlb, addr, end, floor, ceiling) ({BUG(); 0; })
  139. #define hugetlb_fault(mm, vma, addr, flags) ({ BUG(); 0; })
  140. #define hugetlb_mcopy_atomic_pte(dst_mm, dst_pte, dst_vma, dst_addr, \
  141. src_addr, pagep) ({ BUG(); 0; })
  142. #define huge_pte_offset(mm, address) 0
  143. static inline int dequeue_hwpoisoned_huge_page(struct page *page)
  144. {
  145. return 0;
  146. }
  147. static inline bool isolate_huge_page(struct page *page, struct list_head *list)
  148. {
  149. return false;
  150. }
  151. #define putback_active_hugepage(p) do {} while (0)
  152. static inline unsigned long hugetlb_change_protection(struct vm_area_struct *vma,
  153. unsigned long address, unsigned long end, pgprot_t newprot)
  154. {
  155. return 0;
  156. }
  157. static inline void __unmap_hugepage_range_final(struct mmu_gather *tlb,
  158. struct vm_area_struct *vma, unsigned long start,
  159. unsigned long end, struct page *ref_page)
  160. {
  161. BUG();
  162. }
  163. static inline void __unmap_hugepage_range(struct mmu_gather *tlb,
  164. struct vm_area_struct *vma, unsigned long start,
  165. unsigned long end, struct page *ref_page)
  166. {
  167. BUG();
  168. }
  169. #endif /* !CONFIG_HUGETLB_PAGE */
  170. /*
  171. * hugepages at page global directory. If arch support
  172. * hugepages at pgd level, they need to define this.
  173. */
  174. #ifndef pgd_huge
  175. #define pgd_huge(x) 0
  176. #endif
  177. #ifndef p4d_huge
  178. #define p4d_huge(x) 0
  179. #endif
  180. #ifndef pgd_write
  181. static inline int pgd_write(pgd_t pgd)
  182. {
  183. BUG();
  184. return 0;
  185. }
  186. #endif
  187. #ifndef pud_write
  188. static inline int pud_write(pud_t pud)
  189. {
  190. BUG();
  191. return 0;
  192. }
  193. #endif
  194. #ifndef is_hugepd
  195. /*
  196. * Some architectures requires a hugepage directory format that is
  197. * required to support multiple hugepage sizes. For example
  198. * a4fe3ce76 "powerpc/mm: Allow more flexible layouts for hugepage pagetables"
  199. * introduced the same on powerpc. This allows for a more flexible hugepage
  200. * pagetable layout.
  201. */
  202. typedef struct { unsigned long pd; } hugepd_t;
  203. #define is_hugepd(hugepd) (0)
  204. #define __hugepd(x) ((hugepd_t) { (x) })
  205. static inline int gup_huge_pd(hugepd_t hugepd, unsigned long addr,
  206. unsigned pdshift, unsigned long end,
  207. int write, struct page **pages, int *nr)
  208. {
  209. return 0;
  210. }
  211. #else
  212. extern int gup_huge_pd(hugepd_t hugepd, unsigned long addr,
  213. unsigned pdshift, unsigned long end,
  214. int write, struct page **pages, int *nr);
  215. #endif
  216. #define HUGETLB_ANON_FILE "anon_hugepage"
  217. enum {
  218. /*
  219. * The file will be used as an shm file so shmfs accounting rules
  220. * apply
  221. */
  222. HUGETLB_SHMFS_INODE = 1,
  223. /*
  224. * The file is being created on the internal vfs mount and shmfs
  225. * accounting rules do not apply
  226. */
  227. HUGETLB_ANONHUGE_INODE = 2,
  228. };
  229. #ifdef CONFIG_HUGETLBFS
  230. struct hugetlbfs_sb_info {
  231. long max_inodes; /* inodes allowed */
  232. long free_inodes; /* inodes free */
  233. spinlock_t stat_lock;
  234. struct hstate *hstate;
  235. struct hugepage_subpool *spool;
  236. };
  237. static inline struct hugetlbfs_sb_info *HUGETLBFS_SB(struct super_block *sb)
  238. {
  239. return sb->s_fs_info;
  240. }
  241. extern const struct file_operations hugetlbfs_file_operations;
  242. extern const struct vm_operations_struct hugetlb_vm_ops;
  243. struct file *hugetlb_file_setup(const char *name, size_t size, vm_flags_t acct,
  244. struct user_struct **user, int creat_flags,
  245. int page_size_log);
  246. static inline bool is_file_hugepages(struct file *file)
  247. {
  248. if (file->f_op == &hugetlbfs_file_operations)
  249. return true;
  250. return is_file_shm_hugepages(file);
  251. }
  252. #else /* !CONFIG_HUGETLBFS */
  253. #define is_file_hugepages(file) false
  254. static inline struct file *
  255. hugetlb_file_setup(const char *name, size_t size, vm_flags_t acctflag,
  256. struct user_struct **user, int creat_flags,
  257. int page_size_log)
  258. {
  259. return ERR_PTR(-ENOSYS);
  260. }
  261. #endif /* !CONFIG_HUGETLBFS */
  262. #ifdef HAVE_ARCH_HUGETLB_UNMAPPED_AREA
  263. unsigned long hugetlb_get_unmapped_area(struct file *file, unsigned long addr,
  264. unsigned long len, unsigned long pgoff,
  265. unsigned long flags);
  266. #endif /* HAVE_ARCH_HUGETLB_UNMAPPED_AREA */
  267. #ifdef CONFIG_HUGETLB_PAGE
  268. #define HSTATE_NAME_LEN 32
  269. /* Defines one hugetlb page size */
  270. struct hstate {
  271. int next_nid_to_alloc;
  272. int next_nid_to_free;
  273. unsigned int order;
  274. unsigned long mask;
  275. unsigned long max_huge_pages;
  276. unsigned long nr_huge_pages;
  277. unsigned long free_huge_pages;
  278. unsigned long resv_huge_pages;
  279. unsigned long surplus_huge_pages;
  280. unsigned long nr_overcommit_huge_pages;
  281. struct list_head hugepage_activelist;
  282. struct list_head hugepage_freelists[MAX_NUMNODES];
  283. unsigned int nr_huge_pages_node[MAX_NUMNODES];
  284. unsigned int free_huge_pages_node[MAX_NUMNODES];
  285. unsigned int surplus_huge_pages_node[MAX_NUMNODES];
  286. #ifdef CONFIG_CGROUP_HUGETLB
  287. /* cgroup control files */
  288. struct cftype cgroup_files[5];
  289. #endif
  290. char name[HSTATE_NAME_LEN];
  291. };
  292. struct huge_bootmem_page {
  293. struct list_head list;
  294. struct hstate *hstate;
  295. #ifdef CONFIG_HIGHMEM
  296. phys_addr_t phys;
  297. #endif
  298. };
  299. struct page *alloc_huge_page(struct vm_area_struct *vma,
  300. unsigned long addr, int avoid_reserve);
  301. struct page *alloc_huge_page_node(struct hstate *h, int nid);
  302. struct page *alloc_huge_page_noerr(struct vm_area_struct *vma,
  303. unsigned long addr, int avoid_reserve);
  304. int huge_add_to_page_cache(struct page *page, struct address_space *mapping,
  305. pgoff_t idx);
  306. /* arch callback */
  307. int __init alloc_bootmem_huge_page(struct hstate *h);
  308. void __init hugetlb_bad_size(void);
  309. void __init hugetlb_add_hstate(unsigned order);
  310. struct hstate *size_to_hstate(unsigned long size);
  311. #ifndef HUGE_MAX_HSTATE
  312. #define HUGE_MAX_HSTATE 1
  313. #endif
  314. extern struct hstate hstates[HUGE_MAX_HSTATE];
  315. extern unsigned int default_hstate_idx;
  316. #define default_hstate (hstates[default_hstate_idx])
  317. static inline struct hstate *hstate_inode(struct inode *i)
  318. {
  319. return HUGETLBFS_SB(i->i_sb)->hstate;
  320. }
  321. static inline struct hstate *hstate_file(struct file *f)
  322. {
  323. return hstate_inode(file_inode(f));
  324. }
  325. static inline struct hstate *hstate_sizelog(int page_size_log)
  326. {
  327. if (!page_size_log)
  328. return &default_hstate;
  329. return size_to_hstate(1UL << page_size_log);
  330. }
  331. static inline struct hstate *hstate_vma(struct vm_area_struct *vma)
  332. {
  333. return hstate_file(vma->vm_file);
  334. }
  335. static inline unsigned long huge_page_size(struct hstate *h)
  336. {
  337. return (unsigned long)PAGE_SIZE << h->order;
  338. }
  339. extern unsigned long vma_kernel_pagesize(struct vm_area_struct *vma);
  340. extern unsigned long vma_mmu_pagesize(struct vm_area_struct *vma);
  341. static inline unsigned long huge_page_mask(struct hstate *h)
  342. {
  343. return h->mask;
  344. }
  345. static inline unsigned int huge_page_order(struct hstate *h)
  346. {
  347. return h->order;
  348. }
  349. static inline unsigned huge_page_shift(struct hstate *h)
  350. {
  351. return h->order + PAGE_SHIFT;
  352. }
  353. static inline bool hstate_is_gigantic(struct hstate *h)
  354. {
  355. return huge_page_order(h) >= MAX_ORDER;
  356. }
  357. static inline unsigned int pages_per_huge_page(struct hstate *h)
  358. {
  359. return 1 << h->order;
  360. }
  361. static inline unsigned int blocks_per_huge_page(struct hstate *h)
  362. {
  363. return huge_page_size(h) / 512;
  364. }
  365. #include <asm/hugetlb.h>
  366. #ifndef arch_make_huge_pte
  367. static inline pte_t arch_make_huge_pte(pte_t entry, struct vm_area_struct *vma,
  368. struct page *page, int writable)
  369. {
  370. return entry;
  371. }
  372. #endif
  373. static inline struct hstate *page_hstate(struct page *page)
  374. {
  375. VM_BUG_ON_PAGE(!PageHuge(page), page);
  376. return size_to_hstate(PAGE_SIZE << compound_order(page));
  377. }
  378. static inline unsigned hstate_index_to_shift(unsigned index)
  379. {
  380. return hstates[index].order + PAGE_SHIFT;
  381. }
  382. static inline int hstate_index(struct hstate *h)
  383. {
  384. return h - hstates;
  385. }
  386. pgoff_t __basepage_index(struct page *page);
  387. /* Return page->index in PAGE_SIZE units */
  388. static inline pgoff_t basepage_index(struct page *page)
  389. {
  390. if (!PageCompound(page))
  391. return page->index;
  392. return __basepage_index(page);
  393. }
  394. extern int dissolve_free_huge_pages(unsigned long start_pfn,
  395. unsigned long end_pfn);
  396. static inline bool hugepage_migration_supported(struct hstate *h)
  397. {
  398. #ifdef CONFIG_ARCH_ENABLE_HUGEPAGE_MIGRATION
  399. if ((huge_page_shift(h) == PMD_SHIFT) ||
  400. (huge_page_shift(h) == PGDIR_SHIFT))
  401. return true;
  402. else
  403. return false;
  404. #else
  405. return false;
  406. #endif
  407. }
  408. static inline spinlock_t *huge_pte_lockptr(struct hstate *h,
  409. struct mm_struct *mm, pte_t *pte)
  410. {
  411. if (huge_page_size(h) == PMD_SIZE)
  412. return pmd_lockptr(mm, (pmd_t *) pte);
  413. VM_BUG_ON(huge_page_size(h) == PAGE_SIZE);
  414. return &mm->page_table_lock;
  415. }
  416. #ifndef hugepages_supported
  417. /*
  418. * Some platform decide whether they support huge pages at boot
  419. * time. Some of them, such as powerpc, set HPAGE_SHIFT to 0
  420. * when there is no such support
  421. */
  422. #define hugepages_supported() (HPAGE_SHIFT != 0)
  423. #endif
  424. void hugetlb_report_usage(struct seq_file *m, struct mm_struct *mm);
  425. static inline void hugetlb_count_add(long l, struct mm_struct *mm)
  426. {
  427. atomic_long_add(l, &mm->hugetlb_usage);
  428. }
  429. static inline void hugetlb_count_sub(long l, struct mm_struct *mm)
  430. {
  431. atomic_long_sub(l, &mm->hugetlb_usage);
  432. }
  433. #else /* CONFIG_HUGETLB_PAGE */
  434. struct hstate {};
  435. #define alloc_huge_page(v, a, r) NULL
  436. #define alloc_huge_page_node(h, nid) NULL
  437. #define alloc_huge_page_noerr(v, a, r) NULL
  438. #define alloc_bootmem_huge_page(h) NULL
  439. #define hstate_file(f) NULL
  440. #define hstate_sizelog(s) NULL
  441. #define hstate_vma(v) NULL
  442. #define hstate_inode(i) NULL
  443. #define page_hstate(page) NULL
  444. #define huge_page_size(h) PAGE_SIZE
  445. #define huge_page_mask(h) PAGE_MASK
  446. #define vma_kernel_pagesize(v) PAGE_SIZE
  447. #define vma_mmu_pagesize(v) PAGE_SIZE
  448. #define huge_page_order(h) 0
  449. #define huge_page_shift(h) PAGE_SHIFT
  450. static inline bool hstate_is_gigantic(struct hstate *h)
  451. {
  452. return false;
  453. }
  454. static inline unsigned int pages_per_huge_page(struct hstate *h)
  455. {
  456. return 1;
  457. }
  458. #define hstate_index_to_shift(index) 0
  459. #define hstate_index(h) 0
  460. static inline pgoff_t basepage_index(struct page *page)
  461. {
  462. return page->index;
  463. }
  464. #define dissolve_free_huge_pages(s, e) 0
  465. #define hugepage_migration_supported(h) false
  466. static inline spinlock_t *huge_pte_lockptr(struct hstate *h,
  467. struct mm_struct *mm, pte_t *pte)
  468. {
  469. return &mm->page_table_lock;
  470. }
  471. static inline void hugetlb_report_usage(struct seq_file *f, struct mm_struct *m)
  472. {
  473. }
  474. static inline void hugetlb_count_sub(long l, struct mm_struct *mm)
  475. {
  476. }
  477. #endif /* CONFIG_HUGETLB_PAGE */
  478. static inline spinlock_t *huge_pte_lock(struct hstate *h,
  479. struct mm_struct *mm, pte_t *pte)
  480. {
  481. spinlock_t *ptl;
  482. ptl = huge_pte_lockptr(h, mm, pte);
  483. spin_lock(ptl);
  484. return ptl;
  485. }
  486. #endif /* _LINUX_HUGETLB_H */