xref: /linux-6.15/include/linux/hugetlb.h (revision 564f7dfd)
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 			 int *);
70 void unmap_hugepage_range(struct vm_area_struct *,
71 			  unsigned long, unsigned long, struct page *);
72 void __unmap_hugepage_range_final(struct mmu_gather *tlb,
73 			  struct vm_area_struct *vma,
74 			  unsigned long start, unsigned long end,
75 			  struct page *ref_page);
76 void __unmap_hugepage_range(struct mmu_gather *tlb, struct vm_area_struct *vma,
77 				unsigned long start, unsigned long end,
78 				struct page *ref_page);
79 void hugetlb_report_meminfo(struct seq_file *);
80 int hugetlb_report_node_meminfo(int, char *);
81 void hugetlb_show_meminfo(void);
82 unsigned long hugetlb_total_pages(void);
83 int hugetlb_fault(struct mm_struct *mm, struct vm_area_struct *vma,
84 			unsigned long address, unsigned int flags);
85 int hugetlb_mcopy_atomic_pte(struct mm_struct *dst_mm, pte_t *dst_pte,
86 				struct vm_area_struct *dst_vma,
87 				unsigned long dst_addr,
88 				unsigned long src_addr,
89 				struct page **pagep);
90 int hugetlb_reserve_pages(struct inode *inode, long from, long to,
91 						struct vm_area_struct *vma,
92 						vm_flags_t vm_flags);
93 long hugetlb_unreserve_pages(struct inode *inode, long start, long end,
94 						long freed);
95 int dequeue_hwpoisoned_huge_page(struct page *page);
96 bool isolate_huge_page(struct page *page, struct list_head *list);
97 void putback_active_hugepage(struct page *page);
98 void free_huge_page(struct page *page);
99 void hugetlb_fix_reserve_counts(struct inode *inode);
100 extern struct mutex *hugetlb_fault_mutex_table;
101 u32 hugetlb_fault_mutex_hash(struct hstate *h, struct mm_struct *mm,
102 				struct vm_area_struct *vma,
103 				struct address_space *mapping,
104 				pgoff_t idx, unsigned long address);
105 
106 pte_t *huge_pmd_share(struct mm_struct *mm, unsigned long addr, pud_t *pud);
107 
108 extern int hugepages_treat_as_movable;
109 extern int sysctl_hugetlb_shm_group;
110 extern struct list_head huge_boot_pages;
111 
112 /* arch callbacks */
113 
114 pte_t *huge_pte_alloc(struct mm_struct *mm,
115 			unsigned long addr, unsigned long sz);
116 pte_t *huge_pte_offset(struct mm_struct *mm, unsigned long addr);
117 int huge_pmd_unshare(struct mm_struct *mm, unsigned long *addr, pte_t *ptep);
118 struct page *follow_huge_addr(struct mm_struct *mm, unsigned long address,
119 			      int write);
120 struct page *follow_huge_pmd(struct mm_struct *mm, unsigned long address,
121 				pmd_t *pmd, int flags);
122 struct page *follow_huge_pud(struct mm_struct *mm, unsigned long address,
123 				pud_t *pud, int flags);
124 int pmd_huge(pmd_t pmd);
125 int pud_huge(pud_t pud);
126 unsigned long hugetlb_change_protection(struct vm_area_struct *vma,
127 		unsigned long address, unsigned long end, pgprot_t newprot);
128 
129 #else /* !CONFIG_HUGETLB_PAGE */
130 
131 static inline void reset_vma_resv_huge_pages(struct vm_area_struct *vma)
132 {
133 }
134 
135 static inline unsigned long hugetlb_total_pages(void)
136 {
137 	return 0;
138 }
139 
140 #define follow_hugetlb_page(m,v,p,vs,a,b,i,w,n)	({ BUG(); 0; })
141 #define follow_huge_addr(mm, addr, write)	ERR_PTR(-EINVAL)
142 #define copy_hugetlb_page_range(src, dst, vma)	({ BUG(); 0; })
143 static inline void hugetlb_report_meminfo(struct seq_file *m)
144 {
145 }
146 #define hugetlb_report_node_meminfo(n, buf)	0
147 static inline void hugetlb_show_meminfo(void)
148 {
149 }
150 #define follow_huge_pmd(mm, addr, pmd, flags)	NULL
151 #define follow_huge_pud(mm, addr, pud, flags)	NULL
152 #define prepare_hugepage_range(file, addr, len)	(-EINVAL)
153 #define pmd_huge(x)	0
154 #define pud_huge(x)	0
155 #define is_hugepage_only_range(mm, addr, len)	0
156 #define hugetlb_free_pgd_range(tlb, addr, end, floor, ceiling) ({BUG(); 0; })
157 #define hugetlb_fault(mm, vma, addr, flags)	({ BUG(); 0; })
158 #define hugetlb_mcopy_atomic_pte(dst_mm, dst_pte, dst_vma, dst_addr, \
159 				src_addr, pagep)	({ BUG(); 0; })
160 #define huge_pte_offset(mm, address)	0
161 static inline int dequeue_hwpoisoned_huge_page(struct page *page)
162 {
163 	return 0;
164 }
165 
166 static inline bool isolate_huge_page(struct page *page, struct list_head *list)
167 {
168 	return false;
169 }
170 #define putback_active_hugepage(p)	do {} while (0)
171 
172 static inline unsigned long hugetlb_change_protection(struct vm_area_struct *vma,
173 		unsigned long address, unsigned long end, pgprot_t newprot)
174 {
175 	return 0;
176 }
177 
178 static inline void __unmap_hugepage_range_final(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 static inline void __unmap_hugepage_range(struct mmu_gather *tlb,
186 			struct vm_area_struct *vma, unsigned long start,
187 			unsigned long end, struct page *ref_page)
188 {
189 	BUG();
190 }
191 
192 #endif /* !CONFIG_HUGETLB_PAGE */
193 /*
194  * hugepages at page global directory. If arch support
195  * hugepages at pgd level, they need to define this.
196  */
197 #ifndef pgd_huge
198 #define pgd_huge(x)	0
199 #endif
200 #ifndef p4d_huge
201 #define p4d_huge(x)	0
202 #endif
203 
204 #ifndef pgd_write
205 static inline int pgd_write(pgd_t pgd)
206 {
207 	BUG();
208 	return 0;
209 }
210 #endif
211 
212 #ifndef pud_write
213 static inline int pud_write(pud_t pud)
214 {
215 	BUG();
216 	return 0;
217 }
218 #endif
219 
220 #ifndef is_hugepd
221 /*
222  * Some architectures requires a hugepage directory format that is
223  * required to support multiple hugepage sizes. For example
224  * a4fe3ce76 "powerpc/mm: Allow more flexible layouts for hugepage pagetables"
225  * introduced the same on powerpc. This allows for a more flexible hugepage
226  * pagetable layout.
227  */
228 typedef struct { unsigned long pd; } hugepd_t;
229 #define is_hugepd(hugepd) (0)
230 #define __hugepd(x) ((hugepd_t) { (x) })
231 static inline int gup_huge_pd(hugepd_t hugepd, unsigned long addr,
232 			      unsigned pdshift, unsigned long end,
233 			      int write, struct page **pages, int *nr)
234 {
235 	return 0;
236 }
237 #else
238 extern int gup_huge_pd(hugepd_t hugepd, unsigned long addr,
239 		       unsigned pdshift, unsigned long end,
240 		       int write, struct page **pages, int *nr);
241 #endif
242 
243 #define HUGETLB_ANON_FILE "anon_hugepage"
244 
245 enum {
246 	/*
247 	 * The file will be used as an shm file so shmfs accounting rules
248 	 * apply
249 	 */
250 	HUGETLB_SHMFS_INODE     = 1,
251 	/*
252 	 * The file is being created on the internal vfs mount and shmfs
253 	 * accounting rules do not apply
254 	 */
255 	HUGETLB_ANONHUGE_INODE  = 2,
256 };
257 
258 #ifdef CONFIG_HUGETLBFS
259 struct hugetlbfs_sb_info {
260 	long	max_inodes;   /* inodes allowed */
261 	long	free_inodes;  /* inodes free */
262 	spinlock_t	stat_lock;
263 	struct hstate *hstate;
264 	struct hugepage_subpool *spool;
265 };
266 
267 static inline struct hugetlbfs_sb_info *HUGETLBFS_SB(struct super_block *sb)
268 {
269 	return sb->s_fs_info;
270 }
271 
272 extern const struct file_operations hugetlbfs_file_operations;
273 extern const struct vm_operations_struct hugetlb_vm_ops;
274 struct file *hugetlb_file_setup(const char *name, size_t size, vm_flags_t acct,
275 				struct user_struct **user, int creat_flags,
276 				int page_size_log);
277 
278 static inline bool is_file_hugepages(struct file *file)
279 {
280 	if (file->f_op == &hugetlbfs_file_operations)
281 		return true;
282 
283 	return is_file_shm_hugepages(file);
284 }
285 
286 
287 #else /* !CONFIG_HUGETLBFS */
288 
289 #define is_file_hugepages(file)			false
290 static inline struct file *
291 hugetlb_file_setup(const char *name, size_t size, vm_flags_t acctflag,
292 		struct user_struct **user, int creat_flags,
293 		int page_size_log)
294 {
295 	return ERR_PTR(-ENOSYS);
296 }
297 
298 #endif /* !CONFIG_HUGETLBFS */
299 
300 #ifdef HAVE_ARCH_HUGETLB_UNMAPPED_AREA
301 unsigned long hugetlb_get_unmapped_area(struct file *file, unsigned long addr,
302 					unsigned long len, unsigned long pgoff,
303 					unsigned long flags);
304 #endif /* HAVE_ARCH_HUGETLB_UNMAPPED_AREA */
305 
306 #ifdef CONFIG_HUGETLB_PAGE
307 
308 #define HSTATE_NAME_LEN 32
309 /* Defines one hugetlb page size */
310 struct hstate {
311 	int next_nid_to_alloc;
312 	int next_nid_to_free;
313 	unsigned int order;
314 	unsigned long mask;
315 	unsigned long max_huge_pages;
316 	unsigned long nr_huge_pages;
317 	unsigned long free_huge_pages;
318 	unsigned long resv_huge_pages;
319 	unsigned long surplus_huge_pages;
320 	unsigned long nr_overcommit_huge_pages;
321 	struct list_head hugepage_activelist;
322 	struct list_head hugepage_freelists[MAX_NUMNODES];
323 	unsigned int nr_huge_pages_node[MAX_NUMNODES];
324 	unsigned int free_huge_pages_node[MAX_NUMNODES];
325 	unsigned int surplus_huge_pages_node[MAX_NUMNODES];
326 #ifdef CONFIG_CGROUP_HUGETLB
327 	/* cgroup control files */
328 	struct cftype cgroup_files[5];
329 #endif
330 	char name[HSTATE_NAME_LEN];
331 };
332 
333 struct huge_bootmem_page {
334 	struct list_head list;
335 	struct hstate *hstate;
336 #ifdef CONFIG_HIGHMEM
337 	phys_addr_t phys;
338 #endif
339 };
340 
341 struct page *alloc_huge_page(struct vm_area_struct *vma,
342 				unsigned long addr, int avoid_reserve);
343 struct page *alloc_huge_page_node(struct hstate *h, int nid);
344 struct page *alloc_huge_page_noerr(struct vm_area_struct *vma,
345 				unsigned long addr, int avoid_reserve);
346 int huge_add_to_page_cache(struct page *page, struct address_space *mapping,
347 			pgoff_t idx);
348 
349 /* arch callback */
350 int __init alloc_bootmem_huge_page(struct hstate *h);
351 
352 void __init hugetlb_bad_size(void);
353 void __init hugetlb_add_hstate(unsigned order);
354 struct hstate *size_to_hstate(unsigned long size);
355 
356 #ifndef HUGE_MAX_HSTATE
357 #define HUGE_MAX_HSTATE 1
358 #endif
359 
360 extern struct hstate hstates[HUGE_MAX_HSTATE];
361 extern unsigned int default_hstate_idx;
362 
363 #define default_hstate (hstates[default_hstate_idx])
364 
365 static inline struct hstate *hstate_inode(struct inode *i)
366 {
367 	return HUGETLBFS_SB(i->i_sb)->hstate;
368 }
369 
370 static inline struct hstate *hstate_file(struct file *f)
371 {
372 	return hstate_inode(file_inode(f));
373 }
374 
375 static inline struct hstate *hstate_sizelog(int page_size_log)
376 {
377 	if (!page_size_log)
378 		return &default_hstate;
379 
380 	return size_to_hstate(1UL << page_size_log);
381 }
382 
383 static inline struct hstate *hstate_vma(struct vm_area_struct *vma)
384 {
385 	return hstate_file(vma->vm_file);
386 }
387 
388 static inline unsigned long huge_page_size(struct hstate *h)
389 {
390 	return (unsigned long)PAGE_SIZE << h->order;
391 }
392 
393 extern unsigned long vma_kernel_pagesize(struct vm_area_struct *vma);
394 
395 extern unsigned long vma_mmu_pagesize(struct vm_area_struct *vma);
396 
397 static inline unsigned long huge_page_mask(struct hstate *h)
398 {
399 	return h->mask;
400 }
401 
402 static inline unsigned int huge_page_order(struct hstate *h)
403 {
404 	return h->order;
405 }
406 
407 static inline unsigned huge_page_shift(struct hstate *h)
408 {
409 	return h->order + PAGE_SHIFT;
410 }
411 
412 static inline bool hstate_is_gigantic(struct hstate *h)
413 {
414 	return huge_page_order(h) >= MAX_ORDER;
415 }
416 
417 static inline unsigned int pages_per_huge_page(struct hstate *h)
418 {
419 	return 1 << h->order;
420 }
421 
422 static inline unsigned int blocks_per_huge_page(struct hstate *h)
423 {
424 	return huge_page_size(h) / 512;
425 }
426 
427 #include <asm/hugetlb.h>
428 
429 #ifndef arch_make_huge_pte
430 static inline pte_t arch_make_huge_pte(pte_t entry, struct vm_area_struct *vma,
431 				       struct page *page, int writable)
432 {
433 	return entry;
434 }
435 #endif
436 
437 static inline struct hstate *page_hstate(struct page *page)
438 {
439 	VM_BUG_ON_PAGE(!PageHuge(page), page);
440 	return size_to_hstate(PAGE_SIZE << compound_order(page));
441 }
442 
443 static inline unsigned hstate_index_to_shift(unsigned index)
444 {
445 	return hstates[index].order + PAGE_SHIFT;
446 }
447 
448 static inline int hstate_index(struct hstate *h)
449 {
450 	return h - hstates;
451 }
452 
453 pgoff_t __basepage_index(struct page *page);
454 
455 /* Return page->index in PAGE_SIZE units */
456 static inline pgoff_t basepage_index(struct page *page)
457 {
458 	if (!PageCompound(page))
459 		return page->index;
460 
461 	return __basepage_index(page);
462 }
463 
464 extern int dissolve_free_huge_pages(unsigned long start_pfn,
465 				    unsigned long end_pfn);
466 static inline bool hugepage_migration_supported(struct hstate *h)
467 {
468 #ifdef CONFIG_ARCH_ENABLE_HUGEPAGE_MIGRATION
469 	return huge_page_shift(h) == PMD_SHIFT;
470 #else
471 	return false;
472 #endif
473 }
474 
475 static inline spinlock_t *huge_pte_lockptr(struct hstate *h,
476 					   struct mm_struct *mm, pte_t *pte)
477 {
478 	if (huge_page_size(h) == PMD_SIZE)
479 		return pmd_lockptr(mm, (pmd_t *) pte);
480 	VM_BUG_ON(huge_page_size(h) == PAGE_SIZE);
481 	return &mm->page_table_lock;
482 }
483 
484 #ifndef hugepages_supported
485 /*
486  * Some platform decide whether they support huge pages at boot
487  * time. Some of them, such as powerpc, set HPAGE_SHIFT to 0
488  * when there is no such support
489  */
490 #define hugepages_supported() (HPAGE_SHIFT != 0)
491 #endif
492 
493 void hugetlb_report_usage(struct seq_file *m, struct mm_struct *mm);
494 
495 static inline void hugetlb_count_add(long l, struct mm_struct *mm)
496 {
497 	atomic_long_add(l, &mm->hugetlb_usage);
498 }
499 
500 static inline void hugetlb_count_sub(long l, struct mm_struct *mm)
501 {
502 	atomic_long_sub(l, &mm->hugetlb_usage);
503 }
504 #else	/* CONFIG_HUGETLB_PAGE */
505 struct hstate {};
506 #define alloc_huge_page(v, a, r) NULL
507 #define alloc_huge_page_node(h, nid) NULL
508 #define alloc_huge_page_noerr(v, a, r) NULL
509 #define alloc_bootmem_huge_page(h) NULL
510 #define hstate_file(f) NULL
511 #define hstate_sizelog(s) NULL
512 #define hstate_vma(v) NULL
513 #define hstate_inode(i) NULL
514 #define page_hstate(page) NULL
515 #define huge_page_size(h) PAGE_SIZE
516 #define huge_page_mask(h) PAGE_MASK
517 #define vma_kernel_pagesize(v) PAGE_SIZE
518 #define vma_mmu_pagesize(v) PAGE_SIZE
519 #define huge_page_order(h) 0
520 #define huge_page_shift(h) PAGE_SHIFT
521 static inline unsigned int pages_per_huge_page(struct hstate *h)
522 {
523 	return 1;
524 }
525 #define hstate_index_to_shift(index) 0
526 #define hstate_index(h) 0
527 
528 static inline pgoff_t basepage_index(struct page *page)
529 {
530 	return page->index;
531 }
532 #define dissolve_free_huge_pages(s, e)	0
533 #define hugepage_migration_supported(h)	false
534 
535 static inline spinlock_t *huge_pte_lockptr(struct hstate *h,
536 					   struct mm_struct *mm, pte_t *pte)
537 {
538 	return &mm->page_table_lock;
539 }
540 
541 static inline void hugetlb_report_usage(struct seq_file *f, struct mm_struct *m)
542 {
543 }
544 
545 static inline void hugetlb_count_sub(long l, struct mm_struct *mm)
546 {
547 }
548 #endif	/* CONFIG_HUGETLB_PAGE */
549 
550 static inline spinlock_t *huge_pte_lock(struct hstate *h,
551 					struct mm_struct *mm, pte_t *pte)
552 {
553 	spinlock_t *ptl;
554 
555 	ptl = huge_pte_lockptr(h, mm, pte);
556 	spin_lock(ptl);
557 	return ptl;
558 }
559 
560 #endif /* _LINUX_HUGETLB_H */
561