xref: /linux-6.15/include/linux/hugetlb.h (revision bb3fc5dd)
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 
12 struct ctl_table;
13 struct user_struct;
14 struct mmu_gather;
15 
16 #ifdef CONFIG_HUGETLB_PAGE
17 
18 #include <linux/mempolicy.h>
19 #include <linux/shm.h>
20 #include <asm/tlbflush.h>
21 
22 struct hugepage_subpool {
23 	spinlock_t lock;
24 	long count;
25 	long max_hpages;	/* Maximum huge pages or -1 if no maximum. */
26 	long used_hpages;	/* Used count against maximum, includes */
27 				/* both alloced and reserved pages. */
28 	struct hstate *hstate;
29 	long min_hpages;	/* Minimum huge pages or -1 if no minimum. */
30 	long rsv_hpages;	/* Pages reserved against global pool to */
31 				/* sasitfy minimum size. */
32 };
33 
34 struct resv_map {
35 	struct kref refs;
36 	spinlock_t lock;
37 	struct list_head regions;
38 	long adds_in_progress;
39 	struct list_head region_cache;
40 	long region_cache_count;
41 };
42 extern struct resv_map *resv_map_alloc(void);
43 void resv_map_release(struct kref *ref);
44 
45 extern spinlock_t hugetlb_lock;
46 extern int hugetlb_max_hstate __read_mostly;
47 #define for_each_hstate(h) \
48 	for ((h) = hstates; (h) < &hstates[hugetlb_max_hstate]; (h)++)
49 
50 struct hugepage_subpool *hugepage_new_subpool(struct hstate *h, long max_hpages,
51 						long min_hpages);
52 void hugepage_put_subpool(struct hugepage_subpool *spool);
53 
54 void reset_vma_resv_huge_pages(struct vm_area_struct *vma);
55 int hugetlb_sysctl_handler(struct ctl_table *, int, void __user *, size_t *, loff_t *);
56 int hugetlb_overcommit_handler(struct ctl_table *, int, void __user *, size_t *, loff_t *);
57 int hugetlb_treat_movable_handler(struct ctl_table *, int, void __user *, size_t *, loff_t *);
58 
59 #ifdef CONFIG_NUMA
60 int hugetlb_mempolicy_sysctl_handler(struct ctl_table *, int,
61 					void __user *, size_t *, loff_t *);
62 #endif
63 
64 int copy_hugetlb_page_range(struct mm_struct *, struct mm_struct *, struct vm_area_struct *);
65 long follow_hugetlb_page(struct mm_struct *, struct vm_area_struct *,
66 			 struct page **, struct vm_area_struct **,
67 			 unsigned long *, unsigned long *, long, unsigned int);
68 void unmap_hugepage_range(struct vm_area_struct *,
69 			  unsigned long, unsigned long, struct page *);
70 void __unmap_hugepage_range_final(struct mmu_gather *tlb,
71 			  struct vm_area_struct *vma,
72 			  unsigned long start, unsigned long end,
73 			  struct page *ref_page);
74 void __unmap_hugepage_range(struct mmu_gather *tlb, struct vm_area_struct *vma,
75 				unsigned long start, unsigned long end,
76 				struct page *ref_page);
77 void hugetlb_report_meminfo(struct seq_file *);
78 int hugetlb_report_node_meminfo(int, char *);
79 void hugetlb_show_meminfo(void);
80 unsigned long hugetlb_total_pages(void);
81 int hugetlb_fault(struct mm_struct *mm, struct vm_area_struct *vma,
82 			unsigned long address, unsigned int flags);
83 int hugetlb_reserve_pages(struct inode *inode, long from, long to,
84 						struct vm_area_struct *vma,
85 						vm_flags_t vm_flags);
86 long hugetlb_unreserve_pages(struct inode *inode, long start, long end,
87 						long freed);
88 int dequeue_hwpoisoned_huge_page(struct page *page);
89 bool isolate_huge_page(struct page *page, struct list_head *list);
90 void putback_active_hugepage(struct page *page);
91 void free_huge_page(struct page *page);
92 void hugetlb_fix_reserve_counts(struct inode *inode, bool restore_reserve);
93 extern struct mutex *hugetlb_fault_mutex_table;
94 u32 hugetlb_fault_mutex_hash(struct hstate *h, struct mm_struct *mm,
95 				struct vm_area_struct *vma,
96 				struct address_space *mapping,
97 				pgoff_t idx, unsigned long address);
98 
99 #ifdef CONFIG_ARCH_WANT_HUGE_PMD_SHARE
100 pte_t *huge_pmd_share(struct mm_struct *mm, unsigned long addr, pud_t *pud);
101 #endif
102 
103 extern int hugepages_treat_as_movable;
104 extern int sysctl_hugetlb_shm_group;
105 extern struct list_head huge_boot_pages;
106 
107 /* arch callbacks */
108 
109 pte_t *huge_pte_alloc(struct mm_struct *mm,
110 			unsigned long addr, unsigned long sz);
111 pte_t *huge_pte_offset(struct mm_struct *mm, unsigned long addr);
112 int huge_pmd_unshare(struct mm_struct *mm, unsigned long *addr, pte_t *ptep);
113 struct page *follow_huge_addr(struct mm_struct *mm, unsigned long address,
114 			      int write);
115 struct page *follow_huge_pmd(struct mm_struct *mm, unsigned long address,
116 				pmd_t *pmd, int flags);
117 struct page *follow_huge_pud(struct mm_struct *mm, unsigned long address,
118 				pud_t *pud, int flags);
119 int pmd_huge(pmd_t pmd);
120 int pud_huge(pud_t pmd);
121 unsigned long hugetlb_change_protection(struct vm_area_struct *vma,
122 		unsigned long address, unsigned long end, pgprot_t newprot);
123 
124 #else /* !CONFIG_HUGETLB_PAGE */
125 
126 static inline void reset_vma_resv_huge_pages(struct vm_area_struct *vma)
127 {
128 }
129 
130 static inline unsigned long hugetlb_total_pages(void)
131 {
132 	return 0;
133 }
134 
135 #define follow_hugetlb_page(m,v,p,vs,a,b,i,w)	({ BUG(); 0; })
136 #define follow_huge_addr(mm, addr, write)	ERR_PTR(-EINVAL)
137 #define copy_hugetlb_page_range(src, dst, vma)	({ BUG(); 0; })
138 static inline void hugetlb_report_meminfo(struct seq_file *m)
139 {
140 }
141 #define hugetlb_report_node_meminfo(n, buf)	0
142 static inline void hugetlb_show_meminfo(void)
143 {
144 }
145 #define follow_huge_pmd(mm, addr, pmd, flags)	NULL
146 #define follow_huge_pud(mm, addr, pud, flags)	NULL
147 #define prepare_hugepage_range(file, addr, len)	(-EINVAL)
148 #define pmd_huge(x)	0
149 #define pud_huge(x)	0
150 #define is_hugepage_only_range(mm, addr, len)	0
151 #define hugetlb_free_pgd_range(tlb, addr, end, floor, ceiling) ({BUG(); 0; })
152 #define hugetlb_fault(mm, vma, addr, flags)	({ BUG(); 0; })
153 #define huge_pte_offset(mm, address)	0
154 static inline int dequeue_hwpoisoned_huge_page(struct page *page)
155 {
156 	return 0;
157 }
158 
159 static inline bool isolate_huge_page(struct page *page, struct list_head *list)
160 {
161 	return false;
162 }
163 #define putback_active_hugepage(p)	do {} while (0)
164 
165 static inline unsigned long hugetlb_change_protection(struct vm_area_struct *vma,
166 		unsigned long address, unsigned long end, pgprot_t newprot)
167 {
168 	return 0;
169 }
170 
171 static inline void __unmap_hugepage_range_final(struct mmu_gather *tlb,
172 			struct vm_area_struct *vma, unsigned long start,
173 			unsigned long end, struct page *ref_page)
174 {
175 	BUG();
176 }
177 
178 static inline void __unmap_hugepage_range(struct mmu_gather *tlb,
179 			struct vm_area_struct *vma, unsigned long start,
180 			unsigned long end, struct page *ref_page)
181 {
182 	BUG();
183 }
184 
185 #endif /* !CONFIG_HUGETLB_PAGE */
186 /*
187  * hugepages at page global directory. If arch support
188  * hugepages at pgd level, they need to define this.
189  */
190 #ifndef pgd_huge
191 #define pgd_huge(x)	0
192 #endif
193 
194 #ifndef pgd_write
195 static inline int pgd_write(pgd_t pgd)
196 {
197 	BUG();
198 	return 0;
199 }
200 #endif
201 
202 #ifndef pud_write
203 static inline int pud_write(pud_t pud)
204 {
205 	BUG();
206 	return 0;
207 }
208 #endif
209 
210 #ifndef is_hugepd
211 /*
212  * Some architectures requires a hugepage directory format that is
213  * required to support multiple hugepage sizes. For example
214  * a4fe3ce76 "powerpc/mm: Allow more flexible layouts for hugepage pagetables"
215  * introduced the same on powerpc. This allows for a more flexible hugepage
216  * pagetable layout.
217  */
218 typedef struct { unsigned long pd; } hugepd_t;
219 #define is_hugepd(hugepd) (0)
220 #define __hugepd(x) ((hugepd_t) { (x) })
221 static inline int gup_huge_pd(hugepd_t hugepd, unsigned long addr,
222 			      unsigned pdshift, unsigned long end,
223 			      int write, struct page **pages, int *nr)
224 {
225 	return 0;
226 }
227 #else
228 extern int gup_huge_pd(hugepd_t hugepd, unsigned long addr,
229 		       unsigned pdshift, unsigned long end,
230 		       int write, struct page **pages, int *nr);
231 #endif
232 
233 #define HUGETLB_ANON_FILE "anon_hugepage"
234 
235 enum {
236 	/*
237 	 * The file will be used as an shm file so shmfs accounting rules
238 	 * apply
239 	 */
240 	HUGETLB_SHMFS_INODE     = 1,
241 	/*
242 	 * The file is being created on the internal vfs mount and shmfs
243 	 * accounting rules do not apply
244 	 */
245 	HUGETLB_ANONHUGE_INODE  = 2,
246 };
247 
248 #ifdef CONFIG_HUGETLBFS
249 struct hugetlbfs_sb_info {
250 	long	max_inodes;   /* inodes allowed */
251 	long	free_inodes;  /* inodes free */
252 	spinlock_t	stat_lock;
253 	struct hstate *hstate;
254 	struct hugepage_subpool *spool;
255 };
256 
257 static inline struct hugetlbfs_sb_info *HUGETLBFS_SB(struct super_block *sb)
258 {
259 	return sb->s_fs_info;
260 }
261 
262 extern const struct file_operations hugetlbfs_file_operations;
263 extern const struct vm_operations_struct hugetlb_vm_ops;
264 struct file *hugetlb_file_setup(const char *name, size_t size, vm_flags_t acct,
265 				struct user_struct **user, int creat_flags,
266 				int page_size_log);
267 
268 static inline int is_file_hugepages(struct file *file)
269 {
270 	if (file->f_op == &hugetlbfs_file_operations)
271 		return 1;
272 	if (is_file_shm_hugepages(file))
273 		return 1;
274 
275 	return 0;
276 }
277 
278 
279 #else /* !CONFIG_HUGETLBFS */
280 
281 #define is_file_hugepages(file)			0
282 static inline struct file *
283 hugetlb_file_setup(const char *name, size_t size, vm_flags_t acctflag,
284 		struct user_struct **user, int creat_flags,
285 		int page_size_log)
286 {
287 	return ERR_PTR(-ENOSYS);
288 }
289 
290 #endif /* !CONFIG_HUGETLBFS */
291 
292 #ifdef HAVE_ARCH_HUGETLB_UNMAPPED_AREA
293 unsigned long hugetlb_get_unmapped_area(struct file *file, unsigned long addr,
294 					unsigned long len, unsigned long pgoff,
295 					unsigned long flags);
296 #endif /* HAVE_ARCH_HUGETLB_UNMAPPED_AREA */
297 
298 #ifdef CONFIG_HUGETLB_PAGE
299 
300 #define HSTATE_NAME_LEN 32
301 /* Defines one hugetlb page size */
302 struct hstate {
303 	int next_nid_to_alloc;
304 	int next_nid_to_free;
305 	unsigned int order;
306 	unsigned long mask;
307 	unsigned long max_huge_pages;
308 	unsigned long nr_huge_pages;
309 	unsigned long free_huge_pages;
310 	unsigned long resv_huge_pages;
311 	unsigned long surplus_huge_pages;
312 	unsigned long nr_overcommit_huge_pages;
313 	struct list_head hugepage_activelist;
314 	struct list_head hugepage_freelists[MAX_NUMNODES];
315 	unsigned int nr_huge_pages_node[MAX_NUMNODES];
316 	unsigned int free_huge_pages_node[MAX_NUMNODES];
317 	unsigned int surplus_huge_pages_node[MAX_NUMNODES];
318 #ifdef CONFIG_CGROUP_HUGETLB
319 	/* cgroup control files */
320 	struct cftype cgroup_files[5];
321 #endif
322 	char name[HSTATE_NAME_LEN];
323 };
324 
325 struct huge_bootmem_page {
326 	struct list_head list;
327 	struct hstate *hstate;
328 #ifdef CONFIG_HIGHMEM
329 	phys_addr_t phys;
330 #endif
331 };
332 
333 struct page *alloc_huge_page(struct vm_area_struct *vma,
334 				unsigned long addr, int avoid_reserve);
335 struct page *alloc_huge_page_node(struct hstate *h, int nid);
336 struct page *alloc_huge_page_noerr(struct vm_area_struct *vma,
337 				unsigned long addr, int avoid_reserve);
338 int huge_add_to_page_cache(struct page *page, struct address_space *mapping,
339 			pgoff_t idx);
340 
341 /* arch callback */
342 int __init alloc_bootmem_huge_page(struct hstate *h);
343 
344 void __init hugetlb_add_hstate(unsigned order);
345 struct hstate *size_to_hstate(unsigned long size);
346 
347 #ifndef HUGE_MAX_HSTATE
348 #define HUGE_MAX_HSTATE 1
349 #endif
350 
351 extern struct hstate hstates[HUGE_MAX_HSTATE];
352 extern unsigned int default_hstate_idx;
353 
354 #define default_hstate (hstates[default_hstate_idx])
355 
356 static inline struct hstate *hstate_inode(struct inode *i)
357 {
358 	struct hugetlbfs_sb_info *hsb;
359 	hsb = HUGETLBFS_SB(i->i_sb);
360 	return hsb->hstate;
361 }
362 
363 static inline struct hstate *hstate_file(struct file *f)
364 {
365 	return hstate_inode(file_inode(f));
366 }
367 
368 static inline struct hstate *hstate_sizelog(int page_size_log)
369 {
370 	if (!page_size_log)
371 		return &default_hstate;
372 
373 	return size_to_hstate(1UL << page_size_log);
374 }
375 
376 static inline struct hstate *hstate_vma(struct vm_area_struct *vma)
377 {
378 	return hstate_file(vma->vm_file);
379 }
380 
381 static inline unsigned long huge_page_size(struct hstate *h)
382 {
383 	return (unsigned long)PAGE_SIZE << h->order;
384 }
385 
386 extern unsigned long vma_kernel_pagesize(struct vm_area_struct *vma);
387 
388 extern unsigned long vma_mmu_pagesize(struct vm_area_struct *vma);
389 
390 static inline unsigned long huge_page_mask(struct hstate *h)
391 {
392 	return h->mask;
393 }
394 
395 static inline unsigned int huge_page_order(struct hstate *h)
396 {
397 	return h->order;
398 }
399 
400 static inline unsigned huge_page_shift(struct hstate *h)
401 {
402 	return h->order + PAGE_SHIFT;
403 }
404 
405 static inline bool hstate_is_gigantic(struct hstate *h)
406 {
407 	return huge_page_order(h) >= MAX_ORDER;
408 }
409 
410 static inline unsigned int pages_per_huge_page(struct hstate *h)
411 {
412 	return 1 << h->order;
413 }
414 
415 static inline unsigned int blocks_per_huge_page(struct hstate *h)
416 {
417 	return huge_page_size(h) / 512;
418 }
419 
420 #include <asm/hugetlb.h>
421 
422 #ifndef arch_make_huge_pte
423 static inline pte_t arch_make_huge_pte(pte_t entry, struct vm_area_struct *vma,
424 				       struct page *page, int writable)
425 {
426 	return entry;
427 }
428 #endif
429 
430 static inline struct hstate *page_hstate(struct page *page)
431 {
432 	VM_BUG_ON_PAGE(!PageHuge(page), page);
433 	return size_to_hstate(PAGE_SIZE << compound_order(page));
434 }
435 
436 static inline unsigned hstate_index_to_shift(unsigned index)
437 {
438 	return hstates[index].order + PAGE_SHIFT;
439 }
440 
441 static inline int hstate_index(struct hstate *h)
442 {
443 	return h - hstates;
444 }
445 
446 pgoff_t __basepage_index(struct page *page);
447 
448 /* Return page->index in PAGE_SIZE units */
449 static inline pgoff_t basepage_index(struct page *page)
450 {
451 	if (!PageCompound(page))
452 		return page->index;
453 
454 	return __basepage_index(page);
455 }
456 
457 extern void dissolve_free_huge_pages(unsigned long start_pfn,
458 				     unsigned long end_pfn);
459 static inline int hugepage_migration_supported(struct hstate *h)
460 {
461 #ifdef CONFIG_ARCH_ENABLE_HUGEPAGE_MIGRATION
462 	return huge_page_shift(h) == PMD_SHIFT;
463 #else
464 	return 0;
465 #endif
466 }
467 
468 static inline spinlock_t *huge_pte_lockptr(struct hstate *h,
469 					   struct mm_struct *mm, pte_t *pte)
470 {
471 	if (huge_page_size(h) == PMD_SIZE)
472 		return pmd_lockptr(mm, (pmd_t *) pte);
473 	VM_BUG_ON(huge_page_size(h) == PAGE_SIZE);
474 	return &mm->page_table_lock;
475 }
476 
477 #ifndef hugepages_supported
478 /*
479  * Some platform decide whether they support huge pages at boot
480  * time. Some of them, such as powerpc, set HPAGE_SHIFT to 0
481  * when there is no such support
482  */
483 #define hugepages_supported() (HPAGE_SHIFT != 0)
484 #endif
485 
486 #else	/* CONFIG_HUGETLB_PAGE */
487 struct hstate {};
488 #define alloc_huge_page(v, a, r) NULL
489 #define alloc_huge_page_node(h, nid) NULL
490 #define alloc_huge_page_noerr(v, a, r) NULL
491 #define alloc_bootmem_huge_page(h) NULL
492 #define hstate_file(f) NULL
493 #define hstate_sizelog(s) NULL
494 #define hstate_vma(v) NULL
495 #define hstate_inode(i) NULL
496 #define page_hstate(page) NULL
497 #define huge_page_size(h) PAGE_SIZE
498 #define huge_page_mask(h) PAGE_MASK
499 #define vma_kernel_pagesize(v) PAGE_SIZE
500 #define vma_mmu_pagesize(v) PAGE_SIZE
501 #define huge_page_order(h) 0
502 #define huge_page_shift(h) PAGE_SHIFT
503 static inline unsigned int pages_per_huge_page(struct hstate *h)
504 {
505 	return 1;
506 }
507 #define hstate_index_to_shift(index) 0
508 #define hstate_index(h) 0
509 
510 static inline pgoff_t basepage_index(struct page *page)
511 {
512 	return page->index;
513 }
514 #define dissolve_free_huge_pages(s, e)	do {} while (0)
515 #define hugepage_migration_supported(h)	0
516 
517 static inline spinlock_t *huge_pte_lockptr(struct hstate *h,
518 					   struct mm_struct *mm, pte_t *pte)
519 {
520 	return &mm->page_table_lock;
521 }
522 #endif	/* CONFIG_HUGETLB_PAGE */
523 
524 static inline spinlock_t *huge_pte_lock(struct hstate *h,
525 					struct mm_struct *mm, pte_t *pte)
526 {
527 	spinlock_t *ptl;
528 
529 	ptl = huge_pte_lockptr(h, mm, pte);
530 	spin_lock(ptl);
531 	return ptl;
532 }
533 
534 #endif /* _LINUX_HUGETLB_H */
535