Loading...
1#ifndef _LINUX_MM_H
2#define _LINUX_MM_H
3
4#include <linux/errno.h>
5
6#ifdef __KERNEL__
7
8#include <linux/mmdebug.h>
9#include <linux/gfp.h>
10#include <linux/bug.h>
11#include <linux/list.h>
12#include <linux/mmzone.h>
13#include <linux/rbtree.h>
14#include <linux/atomic.h>
15#include <linux/debug_locks.h>
16#include <linux/mm_types.h>
17#include <linux/range.h>
18#include <linux/pfn.h>
19#include <linux/percpu-refcount.h>
20#include <linux/bit_spinlock.h>
21#include <linux/shrinker.h>
22#include <linux/resource.h>
23#include <linux/page_ext.h>
24#include <linux/err.h>
25#include <linux/page_ref.h>
26
27struct mempolicy;
28struct anon_vma;
29struct anon_vma_chain;
30struct file_ra_state;
31struct user_struct;
32struct writeback_control;
33struct bdi_writeback;
34
35#ifndef CONFIG_NEED_MULTIPLE_NODES /* Don't use mapnrs, do it properly */
36extern unsigned long max_mapnr;
37
38static inline void set_max_mapnr(unsigned long limit)
39{
40 max_mapnr = limit;
41}
42#else
43static inline void set_max_mapnr(unsigned long limit) { }
44#endif
45
46extern unsigned long totalram_pages;
47extern void * high_memory;
48extern int page_cluster;
49
50#ifdef CONFIG_SYSCTL
51extern int sysctl_legacy_va_layout;
52#else
53#define sysctl_legacy_va_layout 0
54#endif
55
56#ifdef CONFIG_HAVE_ARCH_MMAP_RND_BITS
57extern const int mmap_rnd_bits_min;
58extern const int mmap_rnd_bits_max;
59extern int mmap_rnd_bits __read_mostly;
60#endif
61#ifdef CONFIG_HAVE_ARCH_MMAP_RND_COMPAT_BITS
62extern const int mmap_rnd_compat_bits_min;
63extern const int mmap_rnd_compat_bits_max;
64extern int mmap_rnd_compat_bits __read_mostly;
65#endif
66
67#include <asm/page.h>
68#include <asm/pgtable.h>
69#include <asm/processor.h>
70
71#ifndef __pa_symbol
72#define __pa_symbol(x) __pa(RELOC_HIDE((unsigned long)(x), 0))
73#endif
74
75#ifndef page_to_virt
76#define page_to_virt(x) __va(PFN_PHYS(page_to_pfn(x)))
77#endif
78
79/*
80 * To prevent common memory management code establishing
81 * a zero page mapping on a read fault.
82 * This macro should be defined within <asm/pgtable.h>.
83 * s390 does this to prevent multiplexing of hardware bits
84 * related to the physical page in case of virtualization.
85 */
86#ifndef mm_forbids_zeropage
87#define mm_forbids_zeropage(X) (0)
88#endif
89
90/*
91 * Default maximum number of active map areas, this limits the number of vmas
92 * per mm struct. Users can overwrite this number by sysctl but there is a
93 * problem.
94 *
95 * When a program's coredump is generated as ELF format, a section is created
96 * per a vma. In ELF, the number of sections is represented in unsigned short.
97 * This means the number of sections should be smaller than 65535 at coredump.
98 * Because the kernel adds some informative sections to a image of program at
99 * generating coredump, we need some margin. The number of extra sections is
100 * 1-3 now and depends on arch. We use "5" as safe margin, here.
101 *
102 * ELF extended numbering allows more than 65535 sections, so 16-bit bound is
103 * not a hard limit any more. Although some userspace tools can be surprised by
104 * that.
105 */
106#define MAPCOUNT_ELF_CORE_MARGIN (5)
107#define DEFAULT_MAX_MAP_COUNT (USHRT_MAX - MAPCOUNT_ELF_CORE_MARGIN)
108
109extern int sysctl_max_map_count;
110
111extern unsigned long sysctl_user_reserve_kbytes;
112extern unsigned long sysctl_admin_reserve_kbytes;
113
114extern int sysctl_overcommit_memory;
115extern int sysctl_overcommit_ratio;
116extern unsigned long sysctl_overcommit_kbytes;
117
118extern int overcommit_ratio_handler(struct ctl_table *, int, void __user *,
119 size_t *, loff_t *);
120extern int overcommit_kbytes_handler(struct ctl_table *, int, void __user *,
121 size_t *, loff_t *);
122
123#define nth_page(page,n) pfn_to_page(page_to_pfn((page)) + (n))
124
125/* to align the pointer to the (next) page boundary */
126#define PAGE_ALIGN(addr) ALIGN(addr, PAGE_SIZE)
127
128/* test whether an address (unsigned long or pointer) is aligned to PAGE_SIZE */
129#define PAGE_ALIGNED(addr) IS_ALIGNED((unsigned long)(addr), PAGE_SIZE)
130
131/*
132 * Linux kernel virtual memory manager primitives.
133 * The idea being to have a "virtual" mm in the same way
134 * we have a virtual fs - giving a cleaner interface to the
135 * mm details, and allowing different kinds of memory mappings
136 * (from shared memory to executable loading to arbitrary
137 * mmap() functions).
138 */
139
140extern struct kmem_cache *vm_area_cachep;
141
142#ifndef CONFIG_MMU
143extern struct rb_root nommu_region_tree;
144extern struct rw_semaphore nommu_region_sem;
145
146extern unsigned int kobjsize(const void *objp);
147#endif
148
149/*
150 * vm_flags in vm_area_struct, see mm_types.h.
151 * When changing, update also include/trace/events/mmflags.h
152 */
153#define VM_NONE 0x00000000
154
155#define VM_READ 0x00000001 /* currently active flags */
156#define VM_WRITE 0x00000002
157#define VM_EXEC 0x00000004
158#define VM_SHARED 0x00000008
159
160/* mprotect() hardcodes VM_MAYREAD >> 4 == VM_READ, and so for r/w/x bits. */
161#define VM_MAYREAD 0x00000010 /* limits for mprotect() etc */
162#define VM_MAYWRITE 0x00000020
163#define VM_MAYEXEC 0x00000040
164#define VM_MAYSHARE 0x00000080
165
166#define VM_GROWSDOWN 0x00000100 /* general info on the segment */
167#define VM_UFFD_MISSING 0x00000200 /* missing pages tracking */
168#define VM_PFNMAP 0x00000400 /* Page-ranges managed without "struct page", just pure PFN */
169#define VM_DENYWRITE 0x00000800 /* ETXTBSY on write attempts.. */
170#define VM_UFFD_WP 0x00001000 /* wrprotect pages tracking */
171
172#define VM_LOCKED 0x00002000
173#define VM_IO 0x00004000 /* Memory mapped I/O or similar */
174
175 /* Used by sys_madvise() */
176#define VM_SEQ_READ 0x00008000 /* App will access data sequentially */
177#define VM_RAND_READ 0x00010000 /* App will not benefit from clustered reads */
178
179#define VM_DONTCOPY 0x00020000 /* Do not copy this vma on fork */
180#define VM_DONTEXPAND 0x00040000 /* Cannot expand with mremap() */
181#define VM_LOCKONFAULT 0x00080000 /* Lock the pages covered when they are faulted in */
182#define VM_ACCOUNT 0x00100000 /* Is a VM accounted object */
183#define VM_NORESERVE 0x00200000 /* should the VM suppress accounting */
184#define VM_HUGETLB 0x00400000 /* Huge TLB Page VM */
185#define VM_ARCH_1 0x01000000 /* Architecture-specific flag */
186#define VM_ARCH_2 0x02000000
187#define VM_DONTDUMP 0x04000000 /* Do not include in the core dump */
188
189#ifdef CONFIG_MEM_SOFT_DIRTY
190# define VM_SOFTDIRTY 0x08000000 /* Not soft dirty clean area */
191#else
192# define VM_SOFTDIRTY 0
193#endif
194
195#define VM_MIXEDMAP 0x10000000 /* Can contain "struct page" and pure PFN pages */
196#define VM_HUGEPAGE 0x20000000 /* MADV_HUGEPAGE marked this vma */
197#define VM_NOHUGEPAGE 0x40000000 /* MADV_NOHUGEPAGE marked this vma */
198#define VM_MERGEABLE 0x80000000 /* KSM may merge identical pages */
199
200#ifdef CONFIG_ARCH_USES_HIGH_VMA_FLAGS
201#define VM_HIGH_ARCH_BIT_0 32 /* bit only usable on 64-bit architectures */
202#define VM_HIGH_ARCH_BIT_1 33 /* bit only usable on 64-bit architectures */
203#define VM_HIGH_ARCH_BIT_2 34 /* bit only usable on 64-bit architectures */
204#define VM_HIGH_ARCH_BIT_3 35 /* bit only usable on 64-bit architectures */
205#define VM_HIGH_ARCH_0 BIT(VM_HIGH_ARCH_BIT_0)
206#define VM_HIGH_ARCH_1 BIT(VM_HIGH_ARCH_BIT_1)
207#define VM_HIGH_ARCH_2 BIT(VM_HIGH_ARCH_BIT_2)
208#define VM_HIGH_ARCH_3 BIT(VM_HIGH_ARCH_BIT_3)
209#endif /* CONFIG_ARCH_USES_HIGH_VMA_FLAGS */
210
211#if defined(CONFIG_X86)
212# define VM_PAT VM_ARCH_1 /* PAT reserves whole VMA at once (x86) */
213#if defined (CONFIG_X86_INTEL_MEMORY_PROTECTION_KEYS)
214# define VM_PKEY_SHIFT VM_HIGH_ARCH_BIT_0
215# define VM_PKEY_BIT0 VM_HIGH_ARCH_0 /* A protection key is a 4-bit value */
216# define VM_PKEY_BIT1 VM_HIGH_ARCH_1
217# define VM_PKEY_BIT2 VM_HIGH_ARCH_2
218# define VM_PKEY_BIT3 VM_HIGH_ARCH_3
219#endif
220#elif defined(CONFIG_PPC)
221# define VM_SAO VM_ARCH_1 /* Strong Access Ordering (powerpc) */
222#elif defined(CONFIG_PARISC)
223# define VM_GROWSUP VM_ARCH_1
224#elif defined(CONFIG_METAG)
225# define VM_GROWSUP VM_ARCH_1
226#elif defined(CONFIG_IA64)
227# define VM_GROWSUP VM_ARCH_1
228#elif !defined(CONFIG_MMU)
229# define VM_MAPPED_COPY VM_ARCH_1 /* T if mapped copy of data (nommu mmap) */
230#endif
231
232#if defined(CONFIG_X86)
233/* MPX specific bounds table or bounds directory */
234# define VM_MPX VM_ARCH_2
235#endif
236
237#ifndef VM_GROWSUP
238# define VM_GROWSUP VM_NONE
239#endif
240
241/* Bits set in the VMA until the stack is in its final location */
242#define VM_STACK_INCOMPLETE_SETUP (VM_RAND_READ | VM_SEQ_READ)
243
244#ifndef VM_STACK_DEFAULT_FLAGS /* arch can override this */
245#define VM_STACK_DEFAULT_FLAGS VM_DATA_DEFAULT_FLAGS
246#endif
247
248#ifdef CONFIG_STACK_GROWSUP
249#define VM_STACK VM_GROWSUP
250#else
251#define VM_STACK VM_GROWSDOWN
252#endif
253
254#define VM_STACK_FLAGS (VM_STACK | VM_STACK_DEFAULT_FLAGS | VM_ACCOUNT)
255
256/*
257 * Special vmas that are non-mergable, non-mlock()able.
258 * Note: mm/huge_memory.c VM_NO_THP depends on this definition.
259 */
260#define VM_SPECIAL (VM_IO | VM_DONTEXPAND | VM_PFNMAP | VM_MIXEDMAP)
261
262/* This mask defines which mm->def_flags a process can inherit its parent */
263#define VM_INIT_DEF_MASK VM_NOHUGEPAGE
264
265/* This mask is used to clear all the VMA flags used by mlock */
266#define VM_LOCKED_CLEAR_MASK (~(VM_LOCKED | VM_LOCKONFAULT))
267
268/*
269 * mapping from the currently active vm_flags protection bits (the
270 * low four bits) to a page protection mask..
271 */
272extern pgprot_t protection_map[16];
273
274#define FAULT_FLAG_WRITE 0x01 /* Fault was a write access */
275#define FAULT_FLAG_MKWRITE 0x02 /* Fault was mkwrite of existing pte */
276#define FAULT_FLAG_ALLOW_RETRY 0x04 /* Retry fault if blocking */
277#define FAULT_FLAG_RETRY_NOWAIT 0x08 /* Don't drop mmap_sem and wait when retrying */
278#define FAULT_FLAG_KILLABLE 0x10 /* The fault task is in SIGKILL killable region */
279#define FAULT_FLAG_TRIED 0x20 /* Second try */
280#define FAULT_FLAG_USER 0x40 /* The fault originated in userspace */
281#define FAULT_FLAG_REMOTE 0x80 /* faulting for non current tsk/mm */
282#define FAULT_FLAG_INSTRUCTION 0x100 /* The fault was during an instruction fetch */
283
284/*
285 * vm_fault is filled by the the pagefault handler and passed to the vma's
286 * ->fault function. The vma's ->fault is responsible for returning a bitmask
287 * of VM_FAULT_xxx flags that give details about how the fault was handled.
288 *
289 * MM layer fills up gfp_mask for page allocations but fault handler might
290 * alter it if its implementation requires a different allocation context.
291 *
292 * pgoff should be used in favour of virtual_address, if possible.
293 */
294struct vm_fault {
295 struct vm_area_struct *vma; /* Target VMA */
296 unsigned int flags; /* FAULT_FLAG_xxx flags */
297 gfp_t gfp_mask; /* gfp mask to be used for allocations */
298 pgoff_t pgoff; /* Logical page offset based on vma */
299 unsigned long address; /* Faulting virtual address */
300 pmd_t *pmd; /* Pointer to pmd entry matching
301 * the 'address' */
302 pte_t orig_pte; /* Value of PTE at the time of fault */
303
304 struct page *cow_page; /* Page handler may use for COW fault */
305 struct mem_cgroup *memcg; /* Cgroup cow_page belongs to */
306 struct page *page; /* ->fault handlers should return a
307 * page here, unless VM_FAULT_NOPAGE
308 * is set (which is also implied by
309 * VM_FAULT_ERROR).
310 */
311 /* These three entries are valid only while holding ptl lock */
312 pte_t *pte; /* Pointer to pte entry matching
313 * the 'address'. NULL if the page
314 * table hasn't been allocated.
315 */
316 spinlock_t *ptl; /* Page table lock.
317 * Protects pte page table if 'pte'
318 * is not NULL, otherwise pmd.
319 */
320 pgtable_t prealloc_pte; /* Pre-allocated pte page table.
321 * vm_ops->map_pages() calls
322 * alloc_set_pte() from atomic context.
323 * do_fault_around() pre-allocates
324 * page table to avoid allocation from
325 * atomic context.
326 */
327};
328
329/*
330 * These are the virtual MM functions - opening of an area, closing and
331 * unmapping it (needed to keep files on disk up-to-date etc), pointer
332 * to the functions called when a no-page or a wp-page exception occurs.
333 */
334struct vm_operations_struct {
335 void (*open)(struct vm_area_struct * area);
336 void (*close)(struct vm_area_struct * area);
337 int (*mremap)(struct vm_area_struct * area);
338 int (*fault)(struct vm_area_struct *vma, struct vm_fault *vmf);
339 int (*pmd_fault)(struct vm_area_struct *, unsigned long address,
340 pmd_t *, unsigned int flags);
341 void (*map_pages)(struct vm_fault *vmf,
342 pgoff_t start_pgoff, pgoff_t end_pgoff);
343
344 /* notification that a previously read-only page is about to become
345 * writable, if an error is returned it will cause a SIGBUS */
346 int (*page_mkwrite)(struct vm_area_struct *vma, struct vm_fault *vmf);
347
348 /* same as page_mkwrite when using VM_PFNMAP|VM_MIXEDMAP */
349 int (*pfn_mkwrite)(struct vm_area_struct *vma, struct vm_fault *vmf);
350
351 /* called by access_process_vm when get_user_pages() fails, typically
352 * for use by special VMAs that can switch between memory and hardware
353 */
354 int (*access)(struct vm_area_struct *vma, unsigned long addr,
355 void *buf, int len, int write);
356
357 /* Called by the /proc/PID/maps code to ask the vma whether it
358 * has a special name. Returning non-NULL will also cause this
359 * vma to be dumped unconditionally. */
360 const char *(*name)(struct vm_area_struct *vma);
361
362#ifdef CONFIG_NUMA
363 /*
364 * set_policy() op must add a reference to any non-NULL @new mempolicy
365 * to hold the policy upon return. Caller should pass NULL @new to
366 * remove a policy and fall back to surrounding context--i.e. do not
367 * install a MPOL_DEFAULT policy, nor the task or system default
368 * mempolicy.
369 */
370 int (*set_policy)(struct vm_area_struct *vma, struct mempolicy *new);
371
372 /*
373 * get_policy() op must add reference [mpol_get()] to any policy at
374 * (vma,addr) marked as MPOL_SHARED. The shared policy infrastructure
375 * in mm/mempolicy.c will do this automatically.
376 * get_policy() must NOT add a ref if the policy at (vma,addr) is not
377 * marked as MPOL_SHARED. vma policies are protected by the mmap_sem.
378 * If no [shared/vma] mempolicy exists at the addr, get_policy() op
379 * must return NULL--i.e., do not "fallback" to task or system default
380 * policy.
381 */
382 struct mempolicy *(*get_policy)(struct vm_area_struct *vma,
383 unsigned long addr);
384#endif
385 /*
386 * Called by vm_normal_page() for special PTEs to find the
387 * page for @addr. This is useful if the default behavior
388 * (using pte_page()) would not find the correct page.
389 */
390 struct page *(*find_special_page)(struct vm_area_struct *vma,
391 unsigned long addr);
392};
393
394struct mmu_gather;
395struct inode;
396
397#define page_private(page) ((page)->private)
398#define set_page_private(page, v) ((page)->private = (v))
399
400#if !defined(__HAVE_ARCH_PTE_DEVMAP) || !defined(CONFIG_TRANSPARENT_HUGEPAGE)
401static inline int pmd_devmap(pmd_t pmd)
402{
403 return 0;
404}
405#endif
406
407/*
408 * FIXME: take this include out, include page-flags.h in
409 * files which need it (119 of them)
410 */
411#include <linux/page-flags.h>
412#include <linux/huge_mm.h>
413
414/*
415 * Methods to modify the page usage count.
416 *
417 * What counts for a page usage:
418 * - cache mapping (page->mapping)
419 * - private data (page->private)
420 * - page mapped in a task's page tables, each mapping
421 * is counted separately
422 *
423 * Also, many kernel routines increase the page count before a critical
424 * routine so they can be sure the page doesn't go away from under them.
425 */
426
427/*
428 * Drop a ref, return true if the refcount fell to zero (the page has no users)
429 */
430static inline int put_page_testzero(struct page *page)
431{
432 VM_BUG_ON_PAGE(page_ref_count(page) == 0, page);
433 return page_ref_dec_and_test(page);
434}
435
436/*
437 * Try to grab a ref unless the page has a refcount of zero, return false if
438 * that is the case.
439 * This can be called when MMU is off so it must not access
440 * any of the virtual mappings.
441 */
442static inline int get_page_unless_zero(struct page *page)
443{
444 return page_ref_add_unless(page, 1, 0);
445}
446
447extern int page_is_ram(unsigned long pfn);
448
449enum {
450 REGION_INTERSECTS,
451 REGION_DISJOINT,
452 REGION_MIXED,
453};
454
455int region_intersects(resource_size_t offset, size_t size, unsigned long flags,
456 unsigned long desc);
457
458/* Support for virtually mapped pages */
459struct page *vmalloc_to_page(const void *addr);
460unsigned long vmalloc_to_pfn(const void *addr);
461
462/*
463 * Determine if an address is within the vmalloc range
464 *
465 * On nommu, vmalloc/vfree wrap through kmalloc/kfree directly, so there
466 * is no special casing required.
467 */
468static inline bool is_vmalloc_addr(const void *x)
469{
470#ifdef CONFIG_MMU
471 unsigned long addr = (unsigned long)x;
472
473 return addr >= VMALLOC_START && addr < VMALLOC_END;
474#else
475 return false;
476#endif
477}
478#ifdef CONFIG_MMU
479extern int is_vmalloc_or_module_addr(const void *x);
480#else
481static inline int is_vmalloc_or_module_addr(const void *x)
482{
483 return 0;
484}
485#endif
486
487extern void kvfree(const void *addr);
488
489static inline atomic_t *compound_mapcount_ptr(struct page *page)
490{
491 return &page[1].compound_mapcount;
492}
493
494static inline int compound_mapcount(struct page *page)
495{
496 VM_BUG_ON_PAGE(!PageCompound(page), page);
497 page = compound_head(page);
498 return atomic_read(compound_mapcount_ptr(page)) + 1;
499}
500
501/*
502 * The atomic page->_mapcount, starts from -1: so that transitions
503 * both from it and to it can be tracked, using atomic_inc_and_test
504 * and atomic_add_negative(-1).
505 */
506static inline void page_mapcount_reset(struct page *page)
507{
508 atomic_set(&(page)->_mapcount, -1);
509}
510
511int __page_mapcount(struct page *page);
512
513static inline int page_mapcount(struct page *page)
514{
515 VM_BUG_ON_PAGE(PageSlab(page), page);
516
517 if (unlikely(PageCompound(page)))
518 return __page_mapcount(page);
519 return atomic_read(&page->_mapcount) + 1;
520}
521
522#ifdef CONFIG_TRANSPARENT_HUGEPAGE
523int total_mapcount(struct page *page);
524int page_trans_huge_mapcount(struct page *page, int *total_mapcount);
525#else
526static inline int total_mapcount(struct page *page)
527{
528 return page_mapcount(page);
529}
530static inline int page_trans_huge_mapcount(struct page *page,
531 int *total_mapcount)
532{
533 int mapcount = page_mapcount(page);
534 if (total_mapcount)
535 *total_mapcount = mapcount;
536 return mapcount;
537}
538#endif
539
540static inline struct page *virt_to_head_page(const void *x)
541{
542 struct page *page = virt_to_page(x);
543
544 return compound_head(page);
545}
546
547void __put_page(struct page *page);
548
549void put_pages_list(struct list_head *pages);
550
551void split_page(struct page *page, unsigned int order);
552
553/*
554 * Compound pages have a destructor function. Provide a
555 * prototype for that function and accessor functions.
556 * These are _only_ valid on the head of a compound page.
557 */
558typedef void compound_page_dtor(struct page *);
559
560/* Keep the enum in sync with compound_page_dtors array in mm/page_alloc.c */
561enum compound_dtor_id {
562 NULL_COMPOUND_DTOR,
563 COMPOUND_PAGE_DTOR,
564#ifdef CONFIG_HUGETLB_PAGE
565 HUGETLB_PAGE_DTOR,
566#endif
567#ifdef CONFIG_TRANSPARENT_HUGEPAGE
568 TRANSHUGE_PAGE_DTOR,
569#endif
570 NR_COMPOUND_DTORS,
571};
572extern compound_page_dtor * const compound_page_dtors[];
573
574static inline void set_compound_page_dtor(struct page *page,
575 enum compound_dtor_id compound_dtor)
576{
577 VM_BUG_ON_PAGE(compound_dtor >= NR_COMPOUND_DTORS, page);
578 page[1].compound_dtor = compound_dtor;
579}
580
581static inline compound_page_dtor *get_compound_page_dtor(struct page *page)
582{
583 VM_BUG_ON_PAGE(page[1].compound_dtor >= NR_COMPOUND_DTORS, page);
584 return compound_page_dtors[page[1].compound_dtor];
585}
586
587static inline unsigned int compound_order(struct page *page)
588{
589 if (!PageHead(page))
590 return 0;
591 return page[1].compound_order;
592}
593
594static inline void set_compound_order(struct page *page, unsigned int order)
595{
596 page[1].compound_order = order;
597}
598
599void free_compound_page(struct page *page);
600
601#ifdef CONFIG_MMU
602/*
603 * Do pte_mkwrite, but only if the vma says VM_WRITE. We do this when
604 * servicing faults for write access. In the normal case, do always want
605 * pte_mkwrite. But get_user_pages can cause write faults for mappings
606 * that do not have writing enabled, when used by access_process_vm.
607 */
608static inline pte_t maybe_mkwrite(pte_t pte, struct vm_area_struct *vma)
609{
610 if (likely(vma->vm_flags & VM_WRITE))
611 pte = pte_mkwrite(pte);
612 return pte;
613}
614
615int alloc_set_pte(struct vm_fault *vmf, struct mem_cgroup *memcg,
616 struct page *page);
617int finish_fault(struct vm_fault *vmf);
618int finish_mkwrite_fault(struct vm_fault *vmf);
619#endif
620
621/*
622 * Multiple processes may "see" the same page. E.g. for untouched
623 * mappings of /dev/null, all processes see the same page full of
624 * zeroes, and text pages of executables and shared libraries have
625 * only one copy in memory, at most, normally.
626 *
627 * For the non-reserved pages, page_count(page) denotes a reference count.
628 * page_count() == 0 means the page is free. page->lru is then used for
629 * freelist management in the buddy allocator.
630 * page_count() > 0 means the page has been allocated.
631 *
632 * Pages are allocated by the slab allocator in order to provide memory
633 * to kmalloc and kmem_cache_alloc. In this case, the management of the
634 * page, and the fields in 'struct page' are the responsibility of mm/slab.c
635 * unless a particular usage is carefully commented. (the responsibility of
636 * freeing the kmalloc memory is the caller's, of course).
637 *
638 * A page may be used by anyone else who does a __get_free_page().
639 * In this case, page_count still tracks the references, and should only
640 * be used through the normal accessor functions. The top bits of page->flags
641 * and page->virtual store page management information, but all other fields
642 * are unused and could be used privately, carefully. The management of this
643 * page is the responsibility of the one who allocated it, and those who have
644 * subsequently been given references to it.
645 *
646 * The other pages (we may call them "pagecache pages") are completely
647 * managed by the Linux memory manager: I/O, buffers, swapping etc.
648 * The following discussion applies only to them.
649 *
650 * A pagecache page contains an opaque `private' member, which belongs to the
651 * page's address_space. Usually, this is the address of a circular list of
652 * the page's disk buffers. PG_private must be set to tell the VM to call
653 * into the filesystem to release these pages.
654 *
655 * A page may belong to an inode's memory mapping. In this case, page->mapping
656 * is the pointer to the inode, and page->index is the file offset of the page,
657 * in units of PAGE_SIZE.
658 *
659 * If pagecache pages are not associated with an inode, they are said to be
660 * anonymous pages. These may become associated with the swapcache, and in that
661 * case PG_swapcache is set, and page->private is an offset into the swapcache.
662 *
663 * In either case (swapcache or inode backed), the pagecache itself holds one
664 * reference to the page. Setting PG_private should also increment the
665 * refcount. The each user mapping also has a reference to the page.
666 *
667 * The pagecache pages are stored in a per-mapping radix tree, which is
668 * rooted at mapping->page_tree, and indexed by offset.
669 * Where 2.4 and early 2.6 kernels kept dirty/clean pages in per-address_space
670 * lists, we instead now tag pages as dirty/writeback in the radix tree.
671 *
672 * All pagecache pages may be subject to I/O:
673 * - inode pages may need to be read from disk,
674 * - inode pages which have been modified and are MAP_SHARED may need
675 * to be written back to the inode on disk,
676 * - anonymous pages (including MAP_PRIVATE file mappings) which have been
677 * modified may need to be swapped out to swap space and (later) to be read
678 * back into memory.
679 */
680
681/*
682 * The zone field is never updated after free_area_init_core()
683 * sets it, so none of the operations on it need to be atomic.
684 */
685
686/* Page flags: | [SECTION] | [NODE] | ZONE | [LAST_CPUPID] | ... | FLAGS | */
687#define SECTIONS_PGOFF ((sizeof(unsigned long)*8) - SECTIONS_WIDTH)
688#define NODES_PGOFF (SECTIONS_PGOFF - NODES_WIDTH)
689#define ZONES_PGOFF (NODES_PGOFF - ZONES_WIDTH)
690#define LAST_CPUPID_PGOFF (ZONES_PGOFF - LAST_CPUPID_WIDTH)
691
692/*
693 * Define the bit shifts to access each section. For non-existent
694 * sections we define the shift as 0; that plus a 0 mask ensures
695 * the compiler will optimise away reference to them.
696 */
697#define SECTIONS_PGSHIFT (SECTIONS_PGOFF * (SECTIONS_WIDTH != 0))
698#define NODES_PGSHIFT (NODES_PGOFF * (NODES_WIDTH != 0))
699#define ZONES_PGSHIFT (ZONES_PGOFF * (ZONES_WIDTH != 0))
700#define LAST_CPUPID_PGSHIFT (LAST_CPUPID_PGOFF * (LAST_CPUPID_WIDTH != 0))
701
702/* NODE:ZONE or SECTION:ZONE is used to ID a zone for the buddy allocator */
703#ifdef NODE_NOT_IN_PAGE_FLAGS
704#define ZONEID_SHIFT (SECTIONS_SHIFT + ZONES_SHIFT)
705#define ZONEID_PGOFF ((SECTIONS_PGOFF < ZONES_PGOFF)? \
706 SECTIONS_PGOFF : ZONES_PGOFF)
707#else
708#define ZONEID_SHIFT (NODES_SHIFT + ZONES_SHIFT)
709#define ZONEID_PGOFF ((NODES_PGOFF < ZONES_PGOFF)? \
710 NODES_PGOFF : ZONES_PGOFF)
711#endif
712
713#define ZONEID_PGSHIFT (ZONEID_PGOFF * (ZONEID_SHIFT != 0))
714
715#if SECTIONS_WIDTH+NODES_WIDTH+ZONES_WIDTH > BITS_PER_LONG - NR_PAGEFLAGS
716#error SECTIONS_WIDTH+NODES_WIDTH+ZONES_WIDTH > BITS_PER_LONG - NR_PAGEFLAGS
717#endif
718
719#define ZONES_MASK ((1UL << ZONES_WIDTH) - 1)
720#define NODES_MASK ((1UL << NODES_WIDTH) - 1)
721#define SECTIONS_MASK ((1UL << SECTIONS_WIDTH) - 1)
722#define LAST_CPUPID_MASK ((1UL << LAST_CPUPID_SHIFT) - 1)
723#define ZONEID_MASK ((1UL << ZONEID_SHIFT) - 1)
724
725static inline enum zone_type page_zonenum(const struct page *page)
726{
727 return (page->flags >> ZONES_PGSHIFT) & ZONES_MASK;
728}
729
730#ifdef CONFIG_ZONE_DEVICE
731void get_zone_device_page(struct page *page);
732void put_zone_device_page(struct page *page);
733static inline bool is_zone_device_page(const struct page *page)
734{
735 return page_zonenum(page) == ZONE_DEVICE;
736}
737#else
738static inline void get_zone_device_page(struct page *page)
739{
740}
741static inline void put_zone_device_page(struct page *page)
742{
743}
744static inline bool is_zone_device_page(const struct page *page)
745{
746 return false;
747}
748#endif
749
750static inline void get_page(struct page *page)
751{
752 page = compound_head(page);
753 /*
754 * Getting a normal page or the head of a compound page
755 * requires to already have an elevated page->_refcount.
756 */
757 VM_BUG_ON_PAGE(page_ref_count(page) <= 0, page);
758 page_ref_inc(page);
759
760 if (unlikely(is_zone_device_page(page)))
761 get_zone_device_page(page);
762}
763
764static inline void put_page(struct page *page)
765{
766 page = compound_head(page);
767
768 if (put_page_testzero(page))
769 __put_page(page);
770
771 if (unlikely(is_zone_device_page(page)))
772 put_zone_device_page(page);
773}
774
775#if defined(CONFIG_SPARSEMEM) && !defined(CONFIG_SPARSEMEM_VMEMMAP)
776#define SECTION_IN_PAGE_FLAGS
777#endif
778
779/*
780 * The identification function is mainly used by the buddy allocator for
781 * determining if two pages could be buddies. We are not really identifying
782 * the zone since we could be using the section number id if we do not have
783 * node id available in page flags.
784 * We only guarantee that it will return the same value for two combinable
785 * pages in a zone.
786 */
787static inline int page_zone_id(struct page *page)
788{
789 return (page->flags >> ZONEID_PGSHIFT) & ZONEID_MASK;
790}
791
792static inline int zone_to_nid(struct zone *zone)
793{
794#ifdef CONFIG_NUMA
795 return zone->node;
796#else
797 return 0;
798#endif
799}
800
801#ifdef NODE_NOT_IN_PAGE_FLAGS
802extern int page_to_nid(const struct page *page);
803#else
804static inline int page_to_nid(const struct page *page)
805{
806 return (page->flags >> NODES_PGSHIFT) & NODES_MASK;
807}
808#endif
809
810#ifdef CONFIG_NUMA_BALANCING
811static inline int cpu_pid_to_cpupid(int cpu, int pid)
812{
813 return ((cpu & LAST__CPU_MASK) << LAST__PID_SHIFT) | (pid & LAST__PID_MASK);
814}
815
816static inline int cpupid_to_pid(int cpupid)
817{
818 return cpupid & LAST__PID_MASK;
819}
820
821static inline int cpupid_to_cpu(int cpupid)
822{
823 return (cpupid >> LAST__PID_SHIFT) & LAST__CPU_MASK;
824}
825
826static inline int cpupid_to_nid(int cpupid)
827{
828 return cpu_to_node(cpupid_to_cpu(cpupid));
829}
830
831static inline bool cpupid_pid_unset(int cpupid)
832{
833 return cpupid_to_pid(cpupid) == (-1 & LAST__PID_MASK);
834}
835
836static inline bool cpupid_cpu_unset(int cpupid)
837{
838 return cpupid_to_cpu(cpupid) == (-1 & LAST__CPU_MASK);
839}
840
841static inline bool __cpupid_match_pid(pid_t task_pid, int cpupid)
842{
843 return (task_pid & LAST__PID_MASK) == cpupid_to_pid(cpupid);
844}
845
846#define cpupid_match_pid(task, cpupid) __cpupid_match_pid(task->pid, cpupid)
847#ifdef LAST_CPUPID_NOT_IN_PAGE_FLAGS
848static inline int page_cpupid_xchg_last(struct page *page, int cpupid)
849{
850 return xchg(&page->_last_cpupid, cpupid & LAST_CPUPID_MASK);
851}
852
853static inline int page_cpupid_last(struct page *page)
854{
855 return page->_last_cpupid;
856}
857static inline void page_cpupid_reset_last(struct page *page)
858{
859 page->_last_cpupid = -1 & LAST_CPUPID_MASK;
860}
861#else
862static inline int page_cpupid_last(struct page *page)
863{
864 return (page->flags >> LAST_CPUPID_PGSHIFT) & LAST_CPUPID_MASK;
865}
866
867extern int page_cpupid_xchg_last(struct page *page, int cpupid);
868
869static inline void page_cpupid_reset_last(struct page *page)
870{
871 page->flags |= LAST_CPUPID_MASK << LAST_CPUPID_PGSHIFT;
872}
873#endif /* LAST_CPUPID_NOT_IN_PAGE_FLAGS */
874#else /* !CONFIG_NUMA_BALANCING */
875static inline int page_cpupid_xchg_last(struct page *page, int cpupid)
876{
877 return page_to_nid(page); /* XXX */
878}
879
880static inline int page_cpupid_last(struct page *page)
881{
882 return page_to_nid(page); /* XXX */
883}
884
885static inline int cpupid_to_nid(int cpupid)
886{
887 return -1;
888}
889
890static inline int cpupid_to_pid(int cpupid)
891{
892 return -1;
893}
894
895static inline int cpupid_to_cpu(int cpupid)
896{
897 return -1;
898}
899
900static inline int cpu_pid_to_cpupid(int nid, int pid)
901{
902 return -1;
903}
904
905static inline bool cpupid_pid_unset(int cpupid)
906{
907 return 1;
908}
909
910static inline void page_cpupid_reset_last(struct page *page)
911{
912}
913
914static inline bool cpupid_match_pid(struct task_struct *task, int cpupid)
915{
916 return false;
917}
918#endif /* CONFIG_NUMA_BALANCING */
919
920static inline struct zone *page_zone(const struct page *page)
921{
922 return &NODE_DATA(page_to_nid(page))->node_zones[page_zonenum(page)];
923}
924
925static inline pg_data_t *page_pgdat(const struct page *page)
926{
927 return NODE_DATA(page_to_nid(page));
928}
929
930#ifdef SECTION_IN_PAGE_FLAGS
931static inline void set_page_section(struct page *page, unsigned long section)
932{
933 page->flags &= ~(SECTIONS_MASK << SECTIONS_PGSHIFT);
934 page->flags |= (section & SECTIONS_MASK) << SECTIONS_PGSHIFT;
935}
936
937static inline unsigned long page_to_section(const struct page *page)
938{
939 return (page->flags >> SECTIONS_PGSHIFT) & SECTIONS_MASK;
940}
941#endif
942
943static inline void set_page_zone(struct page *page, enum zone_type zone)
944{
945 page->flags &= ~(ZONES_MASK << ZONES_PGSHIFT);
946 page->flags |= (zone & ZONES_MASK) << ZONES_PGSHIFT;
947}
948
949static inline void set_page_node(struct page *page, unsigned long node)
950{
951 page->flags &= ~(NODES_MASK << NODES_PGSHIFT);
952 page->flags |= (node & NODES_MASK) << NODES_PGSHIFT;
953}
954
955static inline void set_page_links(struct page *page, enum zone_type zone,
956 unsigned long node, unsigned long pfn)
957{
958 set_page_zone(page, zone);
959 set_page_node(page, node);
960#ifdef SECTION_IN_PAGE_FLAGS
961 set_page_section(page, pfn_to_section_nr(pfn));
962#endif
963}
964
965#ifdef CONFIG_MEMCG
966static inline struct mem_cgroup *page_memcg(struct page *page)
967{
968 return page->mem_cgroup;
969}
970static inline struct mem_cgroup *page_memcg_rcu(struct page *page)
971{
972 WARN_ON_ONCE(!rcu_read_lock_held());
973 return READ_ONCE(page->mem_cgroup);
974}
975#else
976static inline struct mem_cgroup *page_memcg(struct page *page)
977{
978 return NULL;
979}
980static inline struct mem_cgroup *page_memcg_rcu(struct page *page)
981{
982 WARN_ON_ONCE(!rcu_read_lock_held());
983 return NULL;
984}
985#endif
986
987/*
988 * Some inline functions in vmstat.h depend on page_zone()
989 */
990#include <linux/vmstat.h>
991
992static __always_inline void *lowmem_page_address(const struct page *page)
993{
994 return page_to_virt(page);
995}
996
997#if defined(CONFIG_HIGHMEM) && !defined(WANT_PAGE_VIRTUAL)
998#define HASHED_PAGE_VIRTUAL
999#endif
1000
1001#if defined(WANT_PAGE_VIRTUAL)
1002static inline void *page_address(const struct page *page)
1003{
1004 return page->virtual;
1005}
1006static inline void set_page_address(struct page *page, void *address)
1007{
1008 page->virtual = address;
1009}
1010#define page_address_init() do { } while(0)
1011#endif
1012
1013#if defined(HASHED_PAGE_VIRTUAL)
1014void *page_address(const struct page *page);
1015void set_page_address(struct page *page, void *virtual);
1016void page_address_init(void);
1017#endif
1018
1019#if !defined(HASHED_PAGE_VIRTUAL) && !defined(WANT_PAGE_VIRTUAL)
1020#define page_address(page) lowmem_page_address(page)
1021#define set_page_address(page, address) do { } while(0)
1022#define page_address_init() do { } while(0)
1023#endif
1024
1025extern void *page_rmapping(struct page *page);
1026extern struct anon_vma *page_anon_vma(struct page *page);
1027extern struct address_space *page_mapping(struct page *page);
1028
1029extern struct address_space *__page_file_mapping(struct page *);
1030
1031static inline
1032struct address_space *page_file_mapping(struct page *page)
1033{
1034 if (unlikely(PageSwapCache(page)))
1035 return __page_file_mapping(page);
1036
1037 return page->mapping;
1038}
1039
1040extern pgoff_t __page_file_index(struct page *page);
1041
1042/*
1043 * Return the pagecache index of the passed page. Regular pagecache pages
1044 * use ->index whereas swapcache pages use swp_offset(->private)
1045 */
1046static inline pgoff_t page_index(struct page *page)
1047{
1048 if (unlikely(PageSwapCache(page)))
1049 return __page_file_index(page);
1050 return page->index;
1051}
1052
1053bool page_mapped(struct page *page);
1054struct address_space *page_mapping(struct page *page);
1055
1056/*
1057 * Return true only if the page has been allocated with
1058 * ALLOC_NO_WATERMARKS and the low watermark was not
1059 * met implying that the system is under some pressure.
1060 */
1061static inline bool page_is_pfmemalloc(struct page *page)
1062{
1063 /*
1064 * Page index cannot be this large so this must be
1065 * a pfmemalloc page.
1066 */
1067 return page->index == -1UL;
1068}
1069
1070/*
1071 * Only to be called by the page allocator on a freshly allocated
1072 * page.
1073 */
1074static inline void set_page_pfmemalloc(struct page *page)
1075{
1076 page->index = -1UL;
1077}
1078
1079static inline void clear_page_pfmemalloc(struct page *page)
1080{
1081 page->index = 0;
1082}
1083
1084/*
1085 * Different kinds of faults, as returned by handle_mm_fault().
1086 * Used to decide whether a process gets delivered SIGBUS or
1087 * just gets major/minor fault counters bumped up.
1088 */
1089
1090#define VM_FAULT_OOM 0x0001
1091#define VM_FAULT_SIGBUS 0x0002
1092#define VM_FAULT_MAJOR 0x0004
1093#define VM_FAULT_WRITE 0x0008 /* Special case for get_user_pages */
1094#define VM_FAULT_HWPOISON 0x0010 /* Hit poisoned small page */
1095#define VM_FAULT_HWPOISON_LARGE 0x0020 /* Hit poisoned large page. Index encoded in upper bits */
1096#define VM_FAULT_SIGSEGV 0x0040
1097
1098#define VM_FAULT_NOPAGE 0x0100 /* ->fault installed the pte, not return page */
1099#define VM_FAULT_LOCKED 0x0200 /* ->fault locked the returned page */
1100#define VM_FAULT_RETRY 0x0400 /* ->fault blocked, must retry */
1101#define VM_FAULT_FALLBACK 0x0800 /* huge page fault failed, fall back to small */
1102#define VM_FAULT_DONE_COW 0x1000 /* ->fault has fully handled COW */
1103
1104#define VM_FAULT_HWPOISON_LARGE_MASK 0xf000 /* encodes hpage index for large hwpoison */
1105
1106#define VM_FAULT_ERROR (VM_FAULT_OOM | VM_FAULT_SIGBUS | VM_FAULT_SIGSEGV | \
1107 VM_FAULT_HWPOISON | VM_FAULT_HWPOISON_LARGE | \
1108 VM_FAULT_FALLBACK)
1109
1110/* Encode hstate index for a hwpoisoned large page */
1111#define VM_FAULT_SET_HINDEX(x) ((x) << 12)
1112#define VM_FAULT_GET_HINDEX(x) (((x) >> 12) & 0xf)
1113
1114/*
1115 * Can be called by the pagefault handler when it gets a VM_FAULT_OOM.
1116 */
1117extern void pagefault_out_of_memory(void);
1118
1119#define offset_in_page(p) ((unsigned long)(p) & ~PAGE_MASK)
1120
1121/*
1122 * Flags passed to show_mem() and show_free_areas() to suppress output in
1123 * various contexts.
1124 */
1125#define SHOW_MEM_FILTER_NODES (0x0001u) /* disallowed nodes */
1126
1127extern void show_free_areas(unsigned int flags);
1128extern bool skip_free_areas_node(unsigned int flags, int nid);
1129
1130int shmem_zero_setup(struct vm_area_struct *);
1131#ifdef CONFIG_SHMEM
1132bool shmem_mapping(struct address_space *mapping);
1133#else
1134static inline bool shmem_mapping(struct address_space *mapping)
1135{
1136 return false;
1137}
1138#endif
1139
1140extern bool can_do_mlock(void);
1141extern int user_shm_lock(size_t, struct user_struct *);
1142extern void user_shm_unlock(size_t, struct user_struct *);
1143
1144/*
1145 * Parameter block passed down to zap_pte_range in exceptional cases.
1146 */
1147struct zap_details {
1148 struct address_space *check_mapping; /* Check page->mapping if set */
1149 pgoff_t first_index; /* Lowest page->index to unmap */
1150 pgoff_t last_index; /* Highest page->index to unmap */
1151 bool ignore_dirty; /* Ignore dirty pages */
1152 bool check_swap_entries; /* Check also swap entries */
1153};
1154
1155struct page *vm_normal_page(struct vm_area_struct *vma, unsigned long addr,
1156 pte_t pte);
1157struct page *vm_normal_page_pmd(struct vm_area_struct *vma, unsigned long addr,
1158 pmd_t pmd);
1159
1160int zap_vma_ptes(struct vm_area_struct *vma, unsigned long address,
1161 unsigned long size);
1162void zap_page_range(struct vm_area_struct *vma, unsigned long address,
1163 unsigned long size, struct zap_details *);
1164void unmap_vmas(struct mmu_gather *tlb, struct vm_area_struct *start_vma,
1165 unsigned long start, unsigned long end);
1166
1167/**
1168 * mm_walk - callbacks for walk_page_range
1169 * @pmd_entry: if set, called for each non-empty PMD (3rd-level) entry
1170 * this handler is required to be able to handle
1171 * pmd_trans_huge() pmds. They may simply choose to
1172 * split_huge_page() instead of handling it explicitly.
1173 * @pte_entry: if set, called for each non-empty PTE (4th-level) entry
1174 * @pte_hole: if set, called for each hole at all levels
1175 * @hugetlb_entry: if set, called for each hugetlb entry
1176 * @test_walk: caller specific callback function to determine whether
1177 * we walk over the current vma or not. Returning 0
1178 * value means "do page table walk over the current vma,"
1179 * and a negative one means "abort current page table walk
1180 * right now." 1 means "skip the current vma."
1181 * @mm: mm_struct representing the target process of page table walk
1182 * @vma: vma currently walked (NULL if walking outside vmas)
1183 * @private: private data for callbacks' usage
1184 *
1185 * (see the comment on walk_page_range() for more details)
1186 */
1187struct mm_walk {
1188 int (*pmd_entry)(pmd_t *pmd, unsigned long addr,
1189 unsigned long next, struct mm_walk *walk);
1190 int (*pte_entry)(pte_t *pte, unsigned long addr,
1191 unsigned long next, struct mm_walk *walk);
1192 int (*pte_hole)(unsigned long addr, unsigned long next,
1193 struct mm_walk *walk);
1194 int (*hugetlb_entry)(pte_t *pte, unsigned long hmask,
1195 unsigned long addr, unsigned long next,
1196 struct mm_walk *walk);
1197 int (*test_walk)(unsigned long addr, unsigned long next,
1198 struct mm_walk *walk);
1199 struct mm_struct *mm;
1200 struct vm_area_struct *vma;
1201 void *private;
1202};
1203
1204int walk_page_range(unsigned long addr, unsigned long end,
1205 struct mm_walk *walk);
1206int walk_page_vma(struct vm_area_struct *vma, struct mm_walk *walk);
1207void free_pgd_range(struct mmu_gather *tlb, unsigned long addr,
1208 unsigned long end, unsigned long floor, unsigned long ceiling);
1209int copy_page_range(struct mm_struct *dst, struct mm_struct *src,
1210 struct vm_area_struct *vma);
1211void unmap_mapping_range(struct address_space *mapping,
1212 loff_t const holebegin, loff_t const holelen, int even_cows);
1213int follow_pte_pmd(struct mm_struct *mm, unsigned long address,
1214 pte_t **ptepp, pmd_t **pmdpp, spinlock_t **ptlp);
1215int follow_pfn(struct vm_area_struct *vma, unsigned long address,
1216 unsigned long *pfn);
1217int follow_phys(struct vm_area_struct *vma, unsigned long address,
1218 unsigned int flags, unsigned long *prot, resource_size_t *phys);
1219int generic_access_phys(struct vm_area_struct *vma, unsigned long addr,
1220 void *buf, int len, int write);
1221
1222static inline void unmap_shared_mapping_range(struct address_space *mapping,
1223 loff_t const holebegin, loff_t const holelen)
1224{
1225 unmap_mapping_range(mapping, holebegin, holelen, 0);
1226}
1227
1228extern void truncate_pagecache(struct inode *inode, loff_t new);
1229extern void truncate_setsize(struct inode *inode, loff_t newsize);
1230void pagecache_isize_extended(struct inode *inode, loff_t from, loff_t to);
1231void truncate_pagecache_range(struct inode *inode, loff_t offset, loff_t end);
1232int truncate_inode_page(struct address_space *mapping, struct page *page);
1233int generic_error_remove_page(struct address_space *mapping, struct page *page);
1234int invalidate_inode_page(struct page *page);
1235
1236#ifdef CONFIG_MMU
1237extern int handle_mm_fault(struct vm_area_struct *vma, unsigned long address,
1238 unsigned int flags);
1239extern int fixup_user_fault(struct task_struct *tsk, struct mm_struct *mm,
1240 unsigned long address, unsigned int fault_flags,
1241 bool *unlocked);
1242#else
1243static inline int handle_mm_fault(struct vm_area_struct *vma,
1244 unsigned long address, unsigned int flags)
1245{
1246 /* should never happen if there's no MMU */
1247 BUG();
1248 return VM_FAULT_SIGBUS;
1249}
1250static inline int fixup_user_fault(struct task_struct *tsk,
1251 struct mm_struct *mm, unsigned long address,
1252 unsigned int fault_flags, bool *unlocked)
1253{
1254 /* should never happen if there's no MMU */
1255 BUG();
1256 return -EFAULT;
1257}
1258#endif
1259
1260extern int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len,
1261 unsigned int gup_flags);
1262extern int access_remote_vm(struct mm_struct *mm, unsigned long addr,
1263 void *buf, int len, unsigned int gup_flags);
1264extern int __access_remote_vm(struct task_struct *tsk, struct mm_struct *mm,
1265 unsigned long addr, void *buf, int len, unsigned int gup_flags);
1266
1267long get_user_pages_remote(struct task_struct *tsk, struct mm_struct *mm,
1268 unsigned long start, unsigned long nr_pages,
1269 unsigned int gup_flags, struct page **pages,
1270 struct vm_area_struct **vmas, int *locked);
1271long get_user_pages(unsigned long start, unsigned long nr_pages,
1272 unsigned int gup_flags, struct page **pages,
1273 struct vm_area_struct **vmas);
1274long get_user_pages_locked(unsigned long start, unsigned long nr_pages,
1275 unsigned int gup_flags, struct page **pages, int *locked);
1276long get_user_pages_unlocked(unsigned long start, unsigned long nr_pages,
1277 struct page **pages, unsigned int gup_flags);
1278int get_user_pages_fast(unsigned long start, int nr_pages, int write,
1279 struct page **pages);
1280
1281/* Container for pinned pfns / pages */
1282struct frame_vector {
1283 unsigned int nr_allocated; /* Number of frames we have space for */
1284 unsigned int nr_frames; /* Number of frames stored in ptrs array */
1285 bool got_ref; /* Did we pin pages by getting page ref? */
1286 bool is_pfns; /* Does array contain pages or pfns? */
1287 void *ptrs[0]; /* Array of pinned pfns / pages. Use
1288 * pfns_vector_pages() or pfns_vector_pfns()
1289 * for access */
1290};
1291
1292struct frame_vector *frame_vector_create(unsigned int nr_frames);
1293void frame_vector_destroy(struct frame_vector *vec);
1294int get_vaddr_frames(unsigned long start, unsigned int nr_pfns,
1295 unsigned int gup_flags, struct frame_vector *vec);
1296void put_vaddr_frames(struct frame_vector *vec);
1297int frame_vector_to_pages(struct frame_vector *vec);
1298void frame_vector_to_pfns(struct frame_vector *vec);
1299
1300static inline unsigned int frame_vector_count(struct frame_vector *vec)
1301{
1302 return vec->nr_frames;
1303}
1304
1305static inline struct page **frame_vector_pages(struct frame_vector *vec)
1306{
1307 if (vec->is_pfns) {
1308 int err = frame_vector_to_pages(vec);
1309
1310 if (err)
1311 return ERR_PTR(err);
1312 }
1313 return (struct page **)(vec->ptrs);
1314}
1315
1316static inline unsigned long *frame_vector_pfns(struct frame_vector *vec)
1317{
1318 if (!vec->is_pfns)
1319 frame_vector_to_pfns(vec);
1320 return (unsigned long *)(vec->ptrs);
1321}
1322
1323struct kvec;
1324int get_kernel_pages(const struct kvec *iov, int nr_pages, int write,
1325 struct page **pages);
1326int get_kernel_page(unsigned long start, int write, struct page **pages);
1327struct page *get_dump_page(unsigned long addr);
1328
1329extern int try_to_release_page(struct page * page, gfp_t gfp_mask);
1330extern void do_invalidatepage(struct page *page, unsigned int offset,
1331 unsigned int length);
1332
1333int __set_page_dirty_nobuffers(struct page *page);
1334int __set_page_dirty_no_writeback(struct page *page);
1335int redirty_page_for_writepage(struct writeback_control *wbc,
1336 struct page *page);
1337void account_page_dirtied(struct page *page, struct address_space *mapping);
1338void account_page_cleaned(struct page *page, struct address_space *mapping,
1339 struct bdi_writeback *wb);
1340int set_page_dirty(struct page *page);
1341int set_page_dirty_lock(struct page *page);
1342void cancel_dirty_page(struct page *page);
1343int clear_page_dirty_for_io(struct page *page);
1344
1345int get_cmdline(struct task_struct *task, char *buffer, int buflen);
1346
1347/* Is the vma a continuation of the stack vma above it? */
1348static inline int vma_growsdown(struct vm_area_struct *vma, unsigned long addr)
1349{
1350 return vma && (vma->vm_end == addr) && (vma->vm_flags & VM_GROWSDOWN);
1351}
1352
1353static inline bool vma_is_anonymous(struct vm_area_struct *vma)
1354{
1355 return !vma->vm_ops;
1356}
1357
1358static inline int stack_guard_page_start(struct vm_area_struct *vma,
1359 unsigned long addr)
1360{
1361 return (vma->vm_flags & VM_GROWSDOWN) &&
1362 (vma->vm_start == addr) &&
1363 !vma_growsdown(vma->vm_prev, addr);
1364}
1365
1366/* Is the vma a continuation of the stack vma below it? */
1367static inline int vma_growsup(struct vm_area_struct *vma, unsigned long addr)
1368{
1369 return vma && (vma->vm_start == addr) && (vma->vm_flags & VM_GROWSUP);
1370}
1371
1372static inline int stack_guard_page_end(struct vm_area_struct *vma,
1373 unsigned long addr)
1374{
1375 return (vma->vm_flags & VM_GROWSUP) &&
1376 (vma->vm_end == addr) &&
1377 !vma_growsup(vma->vm_next, addr);
1378}
1379
1380int vma_is_stack_for_current(struct vm_area_struct *vma);
1381
1382extern unsigned long move_page_tables(struct vm_area_struct *vma,
1383 unsigned long old_addr, struct vm_area_struct *new_vma,
1384 unsigned long new_addr, unsigned long len,
1385 bool need_rmap_locks);
1386extern unsigned long change_protection(struct vm_area_struct *vma, unsigned long start,
1387 unsigned long end, pgprot_t newprot,
1388 int dirty_accountable, int prot_numa);
1389extern int mprotect_fixup(struct vm_area_struct *vma,
1390 struct vm_area_struct **pprev, unsigned long start,
1391 unsigned long end, unsigned long newflags);
1392
1393/*
1394 * doesn't attempt to fault and will return short.
1395 */
1396int __get_user_pages_fast(unsigned long start, int nr_pages, int write,
1397 struct page **pages);
1398/*
1399 * per-process(per-mm_struct) statistics.
1400 */
1401static inline unsigned long get_mm_counter(struct mm_struct *mm, int member)
1402{
1403 long val = atomic_long_read(&mm->rss_stat.count[member]);
1404
1405#ifdef SPLIT_RSS_COUNTING
1406 /*
1407 * counter is updated in asynchronous manner and may go to minus.
1408 * But it's never be expected number for users.
1409 */
1410 if (val < 0)
1411 val = 0;
1412#endif
1413 return (unsigned long)val;
1414}
1415
1416static inline void add_mm_counter(struct mm_struct *mm, int member, long value)
1417{
1418 atomic_long_add(value, &mm->rss_stat.count[member]);
1419}
1420
1421static inline void inc_mm_counter(struct mm_struct *mm, int member)
1422{
1423 atomic_long_inc(&mm->rss_stat.count[member]);
1424}
1425
1426static inline void dec_mm_counter(struct mm_struct *mm, int member)
1427{
1428 atomic_long_dec(&mm->rss_stat.count[member]);
1429}
1430
1431/* Optimized variant when page is already known not to be PageAnon */
1432static inline int mm_counter_file(struct page *page)
1433{
1434 if (PageSwapBacked(page))
1435 return MM_SHMEMPAGES;
1436 return MM_FILEPAGES;
1437}
1438
1439static inline int mm_counter(struct page *page)
1440{
1441 if (PageAnon(page))
1442 return MM_ANONPAGES;
1443 return mm_counter_file(page);
1444}
1445
1446static inline unsigned long get_mm_rss(struct mm_struct *mm)
1447{
1448 return get_mm_counter(mm, MM_FILEPAGES) +
1449 get_mm_counter(mm, MM_ANONPAGES) +
1450 get_mm_counter(mm, MM_SHMEMPAGES);
1451}
1452
1453static inline unsigned long get_mm_hiwater_rss(struct mm_struct *mm)
1454{
1455 return max(mm->hiwater_rss, get_mm_rss(mm));
1456}
1457
1458static inline unsigned long get_mm_hiwater_vm(struct mm_struct *mm)
1459{
1460 return max(mm->hiwater_vm, mm->total_vm);
1461}
1462
1463static inline void update_hiwater_rss(struct mm_struct *mm)
1464{
1465 unsigned long _rss = get_mm_rss(mm);
1466
1467 if ((mm)->hiwater_rss < _rss)
1468 (mm)->hiwater_rss = _rss;
1469}
1470
1471static inline void update_hiwater_vm(struct mm_struct *mm)
1472{
1473 if (mm->hiwater_vm < mm->total_vm)
1474 mm->hiwater_vm = mm->total_vm;
1475}
1476
1477static inline void reset_mm_hiwater_rss(struct mm_struct *mm)
1478{
1479 mm->hiwater_rss = get_mm_rss(mm);
1480}
1481
1482static inline void setmax_mm_hiwater_rss(unsigned long *maxrss,
1483 struct mm_struct *mm)
1484{
1485 unsigned long hiwater_rss = get_mm_hiwater_rss(mm);
1486
1487 if (*maxrss < hiwater_rss)
1488 *maxrss = hiwater_rss;
1489}
1490
1491#if defined(SPLIT_RSS_COUNTING)
1492void sync_mm_rss(struct mm_struct *mm);
1493#else
1494static inline void sync_mm_rss(struct mm_struct *mm)
1495{
1496}
1497#endif
1498
1499#ifndef __HAVE_ARCH_PTE_DEVMAP
1500static inline int pte_devmap(pte_t pte)
1501{
1502 return 0;
1503}
1504#endif
1505
1506int vma_wants_writenotify(struct vm_area_struct *vma, pgprot_t vm_page_prot);
1507
1508extern pte_t *__get_locked_pte(struct mm_struct *mm, unsigned long addr,
1509 spinlock_t **ptl);
1510static inline pte_t *get_locked_pte(struct mm_struct *mm, unsigned long addr,
1511 spinlock_t **ptl)
1512{
1513 pte_t *ptep;
1514 __cond_lock(*ptl, ptep = __get_locked_pte(mm, addr, ptl));
1515 return ptep;
1516}
1517
1518#ifdef __PAGETABLE_PUD_FOLDED
1519static inline int __pud_alloc(struct mm_struct *mm, pgd_t *pgd,
1520 unsigned long address)
1521{
1522 return 0;
1523}
1524#else
1525int __pud_alloc(struct mm_struct *mm, pgd_t *pgd, unsigned long address);
1526#endif
1527
1528#if defined(__PAGETABLE_PMD_FOLDED) || !defined(CONFIG_MMU)
1529static inline int __pmd_alloc(struct mm_struct *mm, pud_t *pud,
1530 unsigned long address)
1531{
1532 return 0;
1533}
1534
1535static inline void mm_nr_pmds_init(struct mm_struct *mm) {}
1536
1537static inline unsigned long mm_nr_pmds(struct mm_struct *mm)
1538{
1539 return 0;
1540}
1541
1542static inline void mm_inc_nr_pmds(struct mm_struct *mm) {}
1543static inline void mm_dec_nr_pmds(struct mm_struct *mm) {}
1544
1545#else
1546int __pmd_alloc(struct mm_struct *mm, pud_t *pud, unsigned long address);
1547
1548static inline void mm_nr_pmds_init(struct mm_struct *mm)
1549{
1550 atomic_long_set(&mm->nr_pmds, 0);
1551}
1552
1553static inline unsigned long mm_nr_pmds(struct mm_struct *mm)
1554{
1555 return atomic_long_read(&mm->nr_pmds);
1556}
1557
1558static inline void mm_inc_nr_pmds(struct mm_struct *mm)
1559{
1560 atomic_long_inc(&mm->nr_pmds);
1561}
1562
1563static inline void mm_dec_nr_pmds(struct mm_struct *mm)
1564{
1565 atomic_long_dec(&mm->nr_pmds);
1566}
1567#endif
1568
1569int __pte_alloc(struct mm_struct *mm, pmd_t *pmd, unsigned long address);
1570int __pte_alloc_kernel(pmd_t *pmd, unsigned long address);
1571
1572/*
1573 * The following ifdef needed to get the 4level-fixup.h header to work.
1574 * Remove it when 4level-fixup.h has been removed.
1575 */
1576#if defined(CONFIG_MMU) && !defined(__ARCH_HAS_4LEVEL_HACK)
1577static inline pud_t *pud_alloc(struct mm_struct *mm, pgd_t *pgd, unsigned long address)
1578{
1579 return (unlikely(pgd_none(*pgd)) && __pud_alloc(mm, pgd, address))?
1580 NULL: pud_offset(pgd, address);
1581}
1582
1583static inline pmd_t *pmd_alloc(struct mm_struct *mm, pud_t *pud, unsigned long address)
1584{
1585 return (unlikely(pud_none(*pud)) && __pmd_alloc(mm, pud, address))?
1586 NULL: pmd_offset(pud, address);
1587}
1588#endif /* CONFIG_MMU && !__ARCH_HAS_4LEVEL_HACK */
1589
1590#if USE_SPLIT_PTE_PTLOCKS
1591#if ALLOC_SPLIT_PTLOCKS
1592void __init ptlock_cache_init(void);
1593extern bool ptlock_alloc(struct page *page);
1594extern void ptlock_free(struct page *page);
1595
1596static inline spinlock_t *ptlock_ptr(struct page *page)
1597{
1598 return page->ptl;
1599}
1600#else /* ALLOC_SPLIT_PTLOCKS */
1601static inline void ptlock_cache_init(void)
1602{
1603}
1604
1605static inline bool ptlock_alloc(struct page *page)
1606{
1607 return true;
1608}
1609
1610static inline void ptlock_free(struct page *page)
1611{
1612}
1613
1614static inline spinlock_t *ptlock_ptr(struct page *page)
1615{
1616 return &page->ptl;
1617}
1618#endif /* ALLOC_SPLIT_PTLOCKS */
1619
1620static inline spinlock_t *pte_lockptr(struct mm_struct *mm, pmd_t *pmd)
1621{
1622 return ptlock_ptr(pmd_page(*pmd));
1623}
1624
1625static inline bool ptlock_init(struct page *page)
1626{
1627 /*
1628 * prep_new_page() initialize page->private (and therefore page->ptl)
1629 * with 0. Make sure nobody took it in use in between.
1630 *
1631 * It can happen if arch try to use slab for page table allocation:
1632 * slab code uses page->slab_cache, which share storage with page->ptl.
1633 */
1634 VM_BUG_ON_PAGE(*(unsigned long *)&page->ptl, page);
1635 if (!ptlock_alloc(page))
1636 return false;
1637 spin_lock_init(ptlock_ptr(page));
1638 return true;
1639}
1640
1641/* Reset page->mapping so free_pages_check won't complain. */
1642static inline void pte_lock_deinit(struct page *page)
1643{
1644 page->mapping = NULL;
1645 ptlock_free(page);
1646}
1647
1648#else /* !USE_SPLIT_PTE_PTLOCKS */
1649/*
1650 * We use mm->page_table_lock to guard all pagetable pages of the mm.
1651 */
1652static inline spinlock_t *pte_lockptr(struct mm_struct *mm, pmd_t *pmd)
1653{
1654 return &mm->page_table_lock;
1655}
1656static inline void ptlock_cache_init(void) {}
1657static inline bool ptlock_init(struct page *page) { return true; }
1658static inline void pte_lock_deinit(struct page *page) {}
1659#endif /* USE_SPLIT_PTE_PTLOCKS */
1660
1661static inline void pgtable_init(void)
1662{
1663 ptlock_cache_init();
1664 pgtable_cache_init();
1665}
1666
1667static inline bool pgtable_page_ctor(struct page *page)
1668{
1669 if (!ptlock_init(page))
1670 return false;
1671 inc_zone_page_state(page, NR_PAGETABLE);
1672 return true;
1673}
1674
1675static inline void pgtable_page_dtor(struct page *page)
1676{
1677 pte_lock_deinit(page);
1678 dec_zone_page_state(page, NR_PAGETABLE);
1679}
1680
1681#define pte_offset_map_lock(mm, pmd, address, ptlp) \
1682({ \
1683 spinlock_t *__ptl = pte_lockptr(mm, pmd); \
1684 pte_t *__pte = pte_offset_map(pmd, address); \
1685 *(ptlp) = __ptl; \
1686 spin_lock(__ptl); \
1687 __pte; \
1688})
1689
1690#define pte_unmap_unlock(pte, ptl) do { \
1691 spin_unlock(ptl); \
1692 pte_unmap(pte); \
1693} while (0)
1694
1695#define pte_alloc(mm, pmd, address) \
1696 (unlikely(pmd_none(*(pmd))) && __pte_alloc(mm, pmd, address))
1697
1698#define pte_alloc_map(mm, pmd, address) \
1699 (pte_alloc(mm, pmd, address) ? NULL : pte_offset_map(pmd, address))
1700
1701#define pte_alloc_map_lock(mm, pmd, address, ptlp) \
1702 (pte_alloc(mm, pmd, address) ? \
1703 NULL : pte_offset_map_lock(mm, pmd, address, ptlp))
1704
1705#define pte_alloc_kernel(pmd, address) \
1706 ((unlikely(pmd_none(*(pmd))) && __pte_alloc_kernel(pmd, address))? \
1707 NULL: pte_offset_kernel(pmd, address))
1708
1709#if USE_SPLIT_PMD_PTLOCKS
1710
1711static struct page *pmd_to_page(pmd_t *pmd)
1712{
1713 unsigned long mask = ~(PTRS_PER_PMD * sizeof(pmd_t) - 1);
1714 return virt_to_page((void *)((unsigned long) pmd & mask));
1715}
1716
1717static inline spinlock_t *pmd_lockptr(struct mm_struct *mm, pmd_t *pmd)
1718{
1719 return ptlock_ptr(pmd_to_page(pmd));
1720}
1721
1722static inline bool pgtable_pmd_page_ctor(struct page *page)
1723{
1724#ifdef CONFIG_TRANSPARENT_HUGEPAGE
1725 page->pmd_huge_pte = NULL;
1726#endif
1727 return ptlock_init(page);
1728}
1729
1730static inline void pgtable_pmd_page_dtor(struct page *page)
1731{
1732#ifdef CONFIG_TRANSPARENT_HUGEPAGE
1733 VM_BUG_ON_PAGE(page->pmd_huge_pte, page);
1734#endif
1735 ptlock_free(page);
1736}
1737
1738#define pmd_huge_pte(mm, pmd) (pmd_to_page(pmd)->pmd_huge_pte)
1739
1740#else
1741
1742static inline spinlock_t *pmd_lockptr(struct mm_struct *mm, pmd_t *pmd)
1743{
1744 return &mm->page_table_lock;
1745}
1746
1747static inline bool pgtable_pmd_page_ctor(struct page *page) { return true; }
1748static inline void pgtable_pmd_page_dtor(struct page *page) {}
1749
1750#define pmd_huge_pte(mm, pmd) ((mm)->pmd_huge_pte)
1751
1752#endif
1753
1754static inline spinlock_t *pmd_lock(struct mm_struct *mm, pmd_t *pmd)
1755{
1756 spinlock_t *ptl = pmd_lockptr(mm, pmd);
1757 spin_lock(ptl);
1758 return ptl;
1759}
1760
1761extern void __init pagecache_init(void);
1762
1763extern void free_area_init(unsigned long * zones_size);
1764extern void free_area_init_node(int nid, unsigned long * zones_size,
1765 unsigned long zone_start_pfn, unsigned long *zholes_size);
1766extern void free_initmem(void);
1767
1768/*
1769 * Free reserved pages within range [PAGE_ALIGN(start), end & PAGE_MASK)
1770 * into the buddy system. The freed pages will be poisoned with pattern
1771 * "poison" if it's within range [0, UCHAR_MAX].
1772 * Return pages freed into the buddy system.
1773 */
1774extern unsigned long free_reserved_area(void *start, void *end,
1775 int poison, char *s);
1776
1777#ifdef CONFIG_HIGHMEM
1778/*
1779 * Free a highmem page into the buddy system, adjusting totalhigh_pages
1780 * and totalram_pages.
1781 */
1782extern void free_highmem_page(struct page *page);
1783#endif
1784
1785extern void adjust_managed_page_count(struct page *page, long count);
1786extern void mem_init_print_info(const char *str);
1787
1788extern void reserve_bootmem_region(phys_addr_t start, phys_addr_t end);
1789
1790/* Free the reserved page into the buddy system, so it gets managed. */
1791static inline void __free_reserved_page(struct page *page)
1792{
1793 ClearPageReserved(page);
1794 init_page_count(page);
1795 __free_page(page);
1796}
1797
1798static inline void free_reserved_page(struct page *page)
1799{
1800 __free_reserved_page(page);
1801 adjust_managed_page_count(page, 1);
1802}
1803
1804static inline void mark_page_reserved(struct page *page)
1805{
1806 SetPageReserved(page);
1807 adjust_managed_page_count(page, -1);
1808}
1809
1810/*
1811 * Default method to free all the __init memory into the buddy system.
1812 * The freed pages will be poisoned with pattern "poison" if it's within
1813 * range [0, UCHAR_MAX].
1814 * Return pages freed into the buddy system.
1815 */
1816static inline unsigned long free_initmem_default(int poison)
1817{
1818 extern char __init_begin[], __init_end[];
1819
1820 return free_reserved_area(&__init_begin, &__init_end,
1821 poison, "unused kernel");
1822}
1823
1824static inline unsigned long get_num_physpages(void)
1825{
1826 int nid;
1827 unsigned long phys_pages = 0;
1828
1829 for_each_online_node(nid)
1830 phys_pages += node_present_pages(nid);
1831
1832 return phys_pages;
1833}
1834
1835#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
1836/*
1837 * With CONFIG_HAVE_MEMBLOCK_NODE_MAP set, an architecture may initialise its
1838 * zones, allocate the backing mem_map and account for memory holes in a more
1839 * architecture independent manner. This is a substitute for creating the
1840 * zone_sizes[] and zholes_size[] arrays and passing them to
1841 * free_area_init_node()
1842 *
1843 * An architecture is expected to register range of page frames backed by
1844 * physical memory with memblock_add[_node]() before calling
1845 * free_area_init_nodes() passing in the PFN each zone ends at. At a basic
1846 * usage, an architecture is expected to do something like
1847 *
1848 * unsigned long max_zone_pfns[MAX_NR_ZONES] = {max_dma, max_normal_pfn,
1849 * max_highmem_pfn};
1850 * for_each_valid_physical_page_range()
1851 * memblock_add_node(base, size, nid)
1852 * free_area_init_nodes(max_zone_pfns);
1853 *
1854 * free_bootmem_with_active_regions() calls free_bootmem_node() for each
1855 * registered physical page range. Similarly
1856 * sparse_memory_present_with_active_regions() calls memory_present() for
1857 * each range when SPARSEMEM is enabled.
1858 *
1859 * See mm/page_alloc.c for more information on each function exposed by
1860 * CONFIG_HAVE_MEMBLOCK_NODE_MAP.
1861 */
1862extern void free_area_init_nodes(unsigned long *max_zone_pfn);
1863unsigned long node_map_pfn_alignment(void);
1864unsigned long __absent_pages_in_range(int nid, unsigned long start_pfn,
1865 unsigned long end_pfn);
1866extern unsigned long absent_pages_in_range(unsigned long start_pfn,
1867 unsigned long end_pfn);
1868extern void get_pfn_range_for_nid(unsigned int nid,
1869 unsigned long *start_pfn, unsigned long *end_pfn);
1870extern unsigned long find_min_pfn_with_active_regions(void);
1871extern void free_bootmem_with_active_regions(int nid,
1872 unsigned long max_low_pfn);
1873extern void sparse_memory_present_with_active_regions(int nid);
1874
1875#endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
1876
1877#if !defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP) && \
1878 !defined(CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID)
1879static inline int __early_pfn_to_nid(unsigned long pfn,
1880 struct mminit_pfnnid_cache *state)
1881{
1882 return 0;
1883}
1884#else
1885/* please see mm/page_alloc.c */
1886extern int __meminit early_pfn_to_nid(unsigned long pfn);
1887/* there is a per-arch backend function. */
1888extern int __meminit __early_pfn_to_nid(unsigned long pfn,
1889 struct mminit_pfnnid_cache *state);
1890#endif
1891
1892extern void set_dma_reserve(unsigned long new_dma_reserve);
1893extern void memmap_init_zone(unsigned long, int, unsigned long,
1894 unsigned long, enum memmap_context);
1895extern void setup_per_zone_wmarks(void);
1896extern int __meminit init_per_zone_wmark_min(void);
1897extern void mem_init(void);
1898extern void __init mmap_init(void);
1899extern void show_mem(unsigned int flags);
1900extern long si_mem_available(void);
1901extern void si_meminfo(struct sysinfo * val);
1902extern void si_meminfo_node(struct sysinfo *val, int nid);
1903#ifdef __HAVE_ARCH_RESERVED_KERNEL_PAGES
1904extern unsigned long arch_reserved_kernel_pages(void);
1905#endif
1906
1907extern __printf(2, 3)
1908void warn_alloc(gfp_t gfp_mask, const char *fmt, ...);
1909
1910extern void setup_per_cpu_pageset(void);
1911
1912extern void zone_pcp_update(struct zone *zone);
1913extern void zone_pcp_reset(struct zone *zone);
1914
1915/* page_alloc.c */
1916extern int min_free_kbytes;
1917extern int watermark_scale_factor;
1918
1919/* nommu.c */
1920extern atomic_long_t mmap_pages_allocated;
1921extern int nommu_shrink_inode_mappings(struct inode *, size_t, size_t);
1922
1923/* interval_tree.c */
1924void vma_interval_tree_insert(struct vm_area_struct *node,
1925 struct rb_root *root);
1926void vma_interval_tree_insert_after(struct vm_area_struct *node,
1927 struct vm_area_struct *prev,
1928 struct rb_root *root);
1929void vma_interval_tree_remove(struct vm_area_struct *node,
1930 struct rb_root *root);
1931struct vm_area_struct *vma_interval_tree_iter_first(struct rb_root *root,
1932 unsigned long start, unsigned long last);
1933struct vm_area_struct *vma_interval_tree_iter_next(struct vm_area_struct *node,
1934 unsigned long start, unsigned long last);
1935
1936#define vma_interval_tree_foreach(vma, root, start, last) \
1937 for (vma = vma_interval_tree_iter_first(root, start, last); \
1938 vma; vma = vma_interval_tree_iter_next(vma, start, last))
1939
1940void anon_vma_interval_tree_insert(struct anon_vma_chain *node,
1941 struct rb_root *root);
1942void anon_vma_interval_tree_remove(struct anon_vma_chain *node,
1943 struct rb_root *root);
1944struct anon_vma_chain *anon_vma_interval_tree_iter_first(
1945 struct rb_root *root, unsigned long start, unsigned long last);
1946struct anon_vma_chain *anon_vma_interval_tree_iter_next(
1947 struct anon_vma_chain *node, unsigned long start, unsigned long last);
1948#ifdef CONFIG_DEBUG_VM_RB
1949void anon_vma_interval_tree_verify(struct anon_vma_chain *node);
1950#endif
1951
1952#define anon_vma_interval_tree_foreach(avc, root, start, last) \
1953 for (avc = anon_vma_interval_tree_iter_first(root, start, last); \
1954 avc; avc = anon_vma_interval_tree_iter_next(avc, start, last))
1955
1956/* mmap.c */
1957extern int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin);
1958extern int __vma_adjust(struct vm_area_struct *vma, unsigned long start,
1959 unsigned long end, pgoff_t pgoff, struct vm_area_struct *insert,
1960 struct vm_area_struct *expand);
1961static inline int vma_adjust(struct vm_area_struct *vma, unsigned long start,
1962 unsigned long end, pgoff_t pgoff, struct vm_area_struct *insert)
1963{
1964 return __vma_adjust(vma, start, end, pgoff, insert, NULL);
1965}
1966extern struct vm_area_struct *vma_merge(struct mm_struct *,
1967 struct vm_area_struct *prev, unsigned long addr, unsigned long end,
1968 unsigned long vm_flags, struct anon_vma *, struct file *, pgoff_t,
1969 struct mempolicy *, struct vm_userfaultfd_ctx);
1970extern struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *);
1971extern int split_vma(struct mm_struct *,
1972 struct vm_area_struct *, unsigned long addr, int new_below);
1973extern int insert_vm_struct(struct mm_struct *, struct vm_area_struct *);
1974extern void __vma_link_rb(struct mm_struct *, struct vm_area_struct *,
1975 struct rb_node **, struct rb_node *);
1976extern void unlink_file_vma(struct vm_area_struct *);
1977extern struct vm_area_struct *copy_vma(struct vm_area_struct **,
1978 unsigned long addr, unsigned long len, pgoff_t pgoff,
1979 bool *need_rmap_locks);
1980extern void exit_mmap(struct mm_struct *);
1981
1982static inline int check_data_rlimit(unsigned long rlim,
1983 unsigned long new,
1984 unsigned long start,
1985 unsigned long end_data,
1986 unsigned long start_data)
1987{
1988 if (rlim < RLIM_INFINITY) {
1989 if (((new - start) + (end_data - start_data)) > rlim)
1990 return -ENOSPC;
1991 }
1992
1993 return 0;
1994}
1995
1996extern int mm_take_all_locks(struct mm_struct *mm);
1997extern void mm_drop_all_locks(struct mm_struct *mm);
1998
1999extern void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file);
2000extern struct file *get_mm_exe_file(struct mm_struct *mm);
2001extern struct file *get_task_exe_file(struct task_struct *task);
2002
2003extern bool may_expand_vm(struct mm_struct *, vm_flags_t, unsigned long npages);
2004extern void vm_stat_account(struct mm_struct *, vm_flags_t, long npages);
2005
2006extern bool vma_is_special_mapping(const struct vm_area_struct *vma,
2007 const struct vm_special_mapping *sm);
2008extern struct vm_area_struct *_install_special_mapping(struct mm_struct *mm,
2009 unsigned long addr, unsigned long len,
2010 unsigned long flags,
2011 const struct vm_special_mapping *spec);
2012/* This is an obsolete alternative to _install_special_mapping. */
2013extern int install_special_mapping(struct mm_struct *mm,
2014 unsigned long addr, unsigned long len,
2015 unsigned long flags, struct page **pages);
2016
2017extern unsigned long get_unmapped_area(struct file *, unsigned long, unsigned long, unsigned long, unsigned long);
2018
2019extern unsigned long mmap_region(struct file *file, unsigned long addr,
2020 unsigned long len, vm_flags_t vm_flags, unsigned long pgoff);
2021extern unsigned long do_mmap(struct file *file, unsigned long addr,
2022 unsigned long len, unsigned long prot, unsigned long flags,
2023 vm_flags_t vm_flags, unsigned long pgoff, unsigned long *populate);
2024extern int do_munmap(struct mm_struct *, unsigned long, size_t);
2025
2026static inline unsigned long
2027do_mmap_pgoff(struct file *file, unsigned long addr,
2028 unsigned long len, unsigned long prot, unsigned long flags,
2029 unsigned long pgoff, unsigned long *populate)
2030{
2031 return do_mmap(file, addr, len, prot, flags, 0, pgoff, populate);
2032}
2033
2034#ifdef CONFIG_MMU
2035extern int __mm_populate(unsigned long addr, unsigned long len,
2036 int ignore_errors);
2037static inline void mm_populate(unsigned long addr, unsigned long len)
2038{
2039 /* Ignore errors */
2040 (void) __mm_populate(addr, len, 1);
2041}
2042#else
2043static inline void mm_populate(unsigned long addr, unsigned long len) {}
2044#endif
2045
2046/* These take the mm semaphore themselves */
2047extern int __must_check vm_brk(unsigned long, unsigned long);
2048extern int vm_munmap(unsigned long, size_t);
2049extern unsigned long __must_check vm_mmap(struct file *, unsigned long,
2050 unsigned long, unsigned long,
2051 unsigned long, unsigned long);
2052
2053struct vm_unmapped_area_info {
2054#define VM_UNMAPPED_AREA_TOPDOWN 1
2055 unsigned long flags;
2056 unsigned long length;
2057 unsigned long low_limit;
2058 unsigned long high_limit;
2059 unsigned long align_mask;
2060 unsigned long align_offset;
2061};
2062
2063extern unsigned long unmapped_area(struct vm_unmapped_area_info *info);
2064extern unsigned long unmapped_area_topdown(struct vm_unmapped_area_info *info);
2065
2066/*
2067 * Search for an unmapped address range.
2068 *
2069 * We are looking for a range that:
2070 * - does not intersect with any VMA;
2071 * - is contained within the [low_limit, high_limit) interval;
2072 * - is at least the desired size.
2073 * - satisfies (begin_addr & align_mask) == (align_offset & align_mask)
2074 */
2075static inline unsigned long
2076vm_unmapped_area(struct vm_unmapped_area_info *info)
2077{
2078 if (info->flags & VM_UNMAPPED_AREA_TOPDOWN)
2079 return unmapped_area_topdown(info);
2080 else
2081 return unmapped_area(info);
2082}
2083
2084/* truncate.c */
2085extern void truncate_inode_pages(struct address_space *, loff_t);
2086extern void truncate_inode_pages_range(struct address_space *,
2087 loff_t lstart, loff_t lend);
2088extern void truncate_inode_pages_final(struct address_space *);
2089
2090/* generic vm_area_ops exported for stackable file systems */
2091extern int filemap_fault(struct vm_area_struct *, struct vm_fault *);
2092extern void filemap_map_pages(struct vm_fault *vmf,
2093 pgoff_t start_pgoff, pgoff_t end_pgoff);
2094extern int filemap_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf);
2095
2096/* mm/page-writeback.c */
2097int write_one_page(struct page *page, int wait);
2098void task_dirty_inc(struct task_struct *tsk);
2099
2100/* readahead.c */
2101#define VM_MAX_READAHEAD 128 /* kbytes */
2102#define VM_MIN_READAHEAD 16 /* kbytes (includes current page) */
2103
2104int force_page_cache_readahead(struct address_space *mapping, struct file *filp,
2105 pgoff_t offset, unsigned long nr_to_read);
2106
2107void page_cache_sync_readahead(struct address_space *mapping,
2108 struct file_ra_state *ra,
2109 struct file *filp,
2110 pgoff_t offset,
2111 unsigned long size);
2112
2113void page_cache_async_readahead(struct address_space *mapping,
2114 struct file_ra_state *ra,
2115 struct file *filp,
2116 struct page *pg,
2117 pgoff_t offset,
2118 unsigned long size);
2119
2120/* Generic expand stack which grows the stack according to GROWS{UP,DOWN} */
2121extern int expand_stack(struct vm_area_struct *vma, unsigned long address);
2122
2123/* CONFIG_STACK_GROWSUP still needs to to grow downwards at some places */
2124extern int expand_downwards(struct vm_area_struct *vma,
2125 unsigned long address);
2126#if VM_GROWSUP
2127extern int expand_upwards(struct vm_area_struct *vma, unsigned long address);
2128#else
2129 #define expand_upwards(vma, address) (0)
2130#endif
2131
2132/* Look up the first VMA which satisfies addr < vm_end, NULL if none. */
2133extern struct vm_area_struct * find_vma(struct mm_struct * mm, unsigned long addr);
2134extern struct vm_area_struct * find_vma_prev(struct mm_struct * mm, unsigned long addr,
2135 struct vm_area_struct **pprev);
2136
2137/* Look up the first VMA which intersects the interval start_addr..end_addr-1,
2138 NULL if none. Assume start_addr < end_addr. */
2139static inline struct vm_area_struct * find_vma_intersection(struct mm_struct * mm, unsigned long start_addr, unsigned long end_addr)
2140{
2141 struct vm_area_struct * vma = find_vma(mm,start_addr);
2142
2143 if (vma && end_addr <= vma->vm_start)
2144 vma = NULL;
2145 return vma;
2146}
2147
2148static inline unsigned long vma_pages(struct vm_area_struct *vma)
2149{
2150 return (vma->vm_end - vma->vm_start) >> PAGE_SHIFT;
2151}
2152
2153/* Look up the first VMA which exactly match the interval vm_start ... vm_end */
2154static inline struct vm_area_struct *find_exact_vma(struct mm_struct *mm,
2155 unsigned long vm_start, unsigned long vm_end)
2156{
2157 struct vm_area_struct *vma = find_vma(mm, vm_start);
2158
2159 if (vma && (vma->vm_start != vm_start || vma->vm_end != vm_end))
2160 vma = NULL;
2161
2162 return vma;
2163}
2164
2165#ifdef CONFIG_MMU
2166pgprot_t vm_get_page_prot(unsigned long vm_flags);
2167void vma_set_page_prot(struct vm_area_struct *vma);
2168#else
2169static inline pgprot_t vm_get_page_prot(unsigned long vm_flags)
2170{
2171 return __pgprot(0);
2172}
2173static inline void vma_set_page_prot(struct vm_area_struct *vma)
2174{
2175 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
2176}
2177#endif
2178
2179#ifdef CONFIG_NUMA_BALANCING
2180unsigned long change_prot_numa(struct vm_area_struct *vma,
2181 unsigned long start, unsigned long end);
2182#endif
2183
2184struct vm_area_struct *find_extend_vma(struct mm_struct *, unsigned long addr);
2185int remap_pfn_range(struct vm_area_struct *, unsigned long addr,
2186 unsigned long pfn, unsigned long size, pgprot_t);
2187int vm_insert_page(struct vm_area_struct *, unsigned long addr, struct page *);
2188int vm_insert_pfn(struct vm_area_struct *vma, unsigned long addr,
2189 unsigned long pfn);
2190int vm_insert_pfn_prot(struct vm_area_struct *vma, unsigned long addr,
2191 unsigned long pfn, pgprot_t pgprot);
2192int vm_insert_mixed(struct vm_area_struct *vma, unsigned long addr,
2193 pfn_t pfn);
2194int vm_iomap_memory(struct vm_area_struct *vma, phys_addr_t start, unsigned long len);
2195
2196
2197struct page *follow_page_mask(struct vm_area_struct *vma,
2198 unsigned long address, unsigned int foll_flags,
2199 unsigned int *page_mask);
2200
2201static inline struct page *follow_page(struct vm_area_struct *vma,
2202 unsigned long address, unsigned int foll_flags)
2203{
2204 unsigned int unused_page_mask;
2205 return follow_page_mask(vma, address, foll_flags, &unused_page_mask);
2206}
2207
2208#define FOLL_WRITE 0x01 /* check pte is writable */
2209#define FOLL_TOUCH 0x02 /* mark page accessed */
2210#define FOLL_GET 0x04 /* do get_page on page */
2211#define FOLL_DUMP 0x08 /* give error on hole if it would be zero */
2212#define FOLL_FORCE 0x10 /* get_user_pages read/write w/o permission */
2213#define FOLL_NOWAIT 0x20 /* if a disk transfer is needed, start the IO
2214 * and return without waiting upon it */
2215#define FOLL_POPULATE 0x40 /* fault in page */
2216#define FOLL_SPLIT 0x80 /* don't return transhuge pages, split them */
2217#define FOLL_HWPOISON 0x100 /* check page is hwpoisoned */
2218#define FOLL_NUMA 0x200 /* force NUMA hinting page fault */
2219#define FOLL_MIGRATION 0x400 /* wait for page to replace migration entry */
2220#define FOLL_TRIED 0x800 /* a retry, previous pass started an IO */
2221#define FOLL_MLOCK 0x1000 /* lock present pages */
2222#define FOLL_REMOTE 0x2000 /* we are working on non-current tsk/mm */
2223#define FOLL_COW 0x4000 /* internal GUP flag */
2224
2225typedef int (*pte_fn_t)(pte_t *pte, pgtable_t token, unsigned long addr,
2226 void *data);
2227extern int apply_to_page_range(struct mm_struct *mm, unsigned long address,
2228 unsigned long size, pte_fn_t fn, void *data);
2229
2230
2231#ifdef CONFIG_PAGE_POISONING
2232extern bool page_poisoning_enabled(void);
2233extern void kernel_poison_pages(struct page *page, int numpages, int enable);
2234extern bool page_is_poisoned(struct page *page);
2235#else
2236static inline bool page_poisoning_enabled(void) { return false; }
2237static inline void kernel_poison_pages(struct page *page, int numpages,
2238 int enable) { }
2239static inline bool page_is_poisoned(struct page *page) { return false; }
2240#endif
2241
2242#ifdef CONFIG_DEBUG_PAGEALLOC
2243extern bool _debug_pagealloc_enabled;
2244extern void __kernel_map_pages(struct page *page, int numpages, int enable);
2245
2246static inline bool debug_pagealloc_enabled(void)
2247{
2248 return _debug_pagealloc_enabled;
2249}
2250
2251static inline void
2252kernel_map_pages(struct page *page, int numpages, int enable)
2253{
2254 if (!debug_pagealloc_enabled())
2255 return;
2256
2257 __kernel_map_pages(page, numpages, enable);
2258}
2259#ifdef CONFIG_HIBERNATION
2260extern bool kernel_page_present(struct page *page);
2261#endif /* CONFIG_HIBERNATION */
2262#else /* CONFIG_DEBUG_PAGEALLOC */
2263static inline void
2264kernel_map_pages(struct page *page, int numpages, int enable) {}
2265#ifdef CONFIG_HIBERNATION
2266static inline bool kernel_page_present(struct page *page) { return true; }
2267#endif /* CONFIG_HIBERNATION */
2268static inline bool debug_pagealloc_enabled(void)
2269{
2270 return false;
2271}
2272#endif /* CONFIG_DEBUG_PAGEALLOC */
2273
2274#ifdef __HAVE_ARCH_GATE_AREA
2275extern struct vm_area_struct *get_gate_vma(struct mm_struct *mm);
2276extern int in_gate_area_no_mm(unsigned long addr);
2277extern int in_gate_area(struct mm_struct *mm, unsigned long addr);
2278#else
2279static inline struct vm_area_struct *get_gate_vma(struct mm_struct *mm)
2280{
2281 return NULL;
2282}
2283static inline int in_gate_area_no_mm(unsigned long addr) { return 0; }
2284static inline int in_gate_area(struct mm_struct *mm, unsigned long addr)
2285{
2286 return 0;
2287}
2288#endif /* __HAVE_ARCH_GATE_AREA */
2289
2290extern bool process_shares_mm(struct task_struct *p, struct mm_struct *mm);
2291
2292#ifdef CONFIG_SYSCTL
2293extern int sysctl_drop_caches;
2294int drop_caches_sysctl_handler(struct ctl_table *, int,
2295 void __user *, size_t *, loff_t *);
2296#endif
2297
2298void drop_slab(void);
2299void drop_slab_node(int nid);
2300
2301#ifndef CONFIG_MMU
2302#define randomize_va_space 0
2303#else
2304extern int randomize_va_space;
2305#endif
2306
2307const char * arch_vma_name(struct vm_area_struct *vma);
2308void print_vma_addr(char *prefix, unsigned long rip);
2309
2310void sparse_mem_maps_populate_node(struct page **map_map,
2311 unsigned long pnum_begin,
2312 unsigned long pnum_end,
2313 unsigned long map_count,
2314 int nodeid);
2315
2316struct page *sparse_mem_map_populate(unsigned long pnum, int nid);
2317pgd_t *vmemmap_pgd_populate(unsigned long addr, int node);
2318pud_t *vmemmap_pud_populate(pgd_t *pgd, unsigned long addr, int node);
2319pmd_t *vmemmap_pmd_populate(pud_t *pud, unsigned long addr, int node);
2320pte_t *vmemmap_pte_populate(pmd_t *pmd, unsigned long addr, int node);
2321void *vmemmap_alloc_block(unsigned long size, int node);
2322struct vmem_altmap;
2323void *__vmemmap_alloc_block_buf(unsigned long size, int node,
2324 struct vmem_altmap *altmap);
2325static inline void *vmemmap_alloc_block_buf(unsigned long size, int node)
2326{
2327 return __vmemmap_alloc_block_buf(size, node, NULL);
2328}
2329
2330void vmemmap_verify(pte_t *, int, unsigned long, unsigned long);
2331int vmemmap_populate_basepages(unsigned long start, unsigned long end,
2332 int node);
2333int vmemmap_populate(unsigned long start, unsigned long end, int node);
2334void vmemmap_populate_print_last(void);
2335#ifdef CONFIG_MEMORY_HOTPLUG
2336void vmemmap_free(unsigned long start, unsigned long end);
2337#endif
2338void register_page_bootmem_memmap(unsigned long section_nr, struct page *map,
2339 unsigned long size);
2340
2341enum mf_flags {
2342 MF_COUNT_INCREASED = 1 << 0,
2343 MF_ACTION_REQUIRED = 1 << 1,
2344 MF_MUST_KILL = 1 << 2,
2345 MF_SOFT_OFFLINE = 1 << 3,
2346};
2347extern int memory_failure(unsigned long pfn, int trapno, int flags);
2348extern void memory_failure_queue(unsigned long pfn, int trapno, int flags);
2349extern int unpoison_memory(unsigned long pfn);
2350extern int get_hwpoison_page(struct page *page);
2351#define put_hwpoison_page(page) put_page(page)
2352extern int sysctl_memory_failure_early_kill;
2353extern int sysctl_memory_failure_recovery;
2354extern void shake_page(struct page *p, int access);
2355extern atomic_long_t num_poisoned_pages;
2356extern int soft_offline_page(struct page *page, int flags);
2357
2358
2359/*
2360 * Error handlers for various types of pages.
2361 */
2362enum mf_result {
2363 MF_IGNORED, /* Error: cannot be handled */
2364 MF_FAILED, /* Error: handling failed */
2365 MF_DELAYED, /* Will be handled later */
2366 MF_RECOVERED, /* Successfully recovered */
2367};
2368
2369enum mf_action_page_type {
2370 MF_MSG_KERNEL,
2371 MF_MSG_KERNEL_HIGH_ORDER,
2372 MF_MSG_SLAB,
2373 MF_MSG_DIFFERENT_COMPOUND,
2374 MF_MSG_POISONED_HUGE,
2375 MF_MSG_HUGE,
2376 MF_MSG_FREE_HUGE,
2377 MF_MSG_UNMAP_FAILED,
2378 MF_MSG_DIRTY_SWAPCACHE,
2379 MF_MSG_CLEAN_SWAPCACHE,
2380 MF_MSG_DIRTY_MLOCKED_LRU,
2381 MF_MSG_CLEAN_MLOCKED_LRU,
2382 MF_MSG_DIRTY_UNEVICTABLE_LRU,
2383 MF_MSG_CLEAN_UNEVICTABLE_LRU,
2384 MF_MSG_DIRTY_LRU,
2385 MF_MSG_CLEAN_LRU,
2386 MF_MSG_TRUNCATED_LRU,
2387 MF_MSG_BUDDY,
2388 MF_MSG_BUDDY_2ND,
2389 MF_MSG_UNKNOWN,
2390};
2391
2392#if defined(CONFIG_TRANSPARENT_HUGEPAGE) || defined(CONFIG_HUGETLBFS)
2393extern void clear_huge_page(struct page *page,
2394 unsigned long addr,
2395 unsigned int pages_per_huge_page);
2396extern void copy_user_huge_page(struct page *dst, struct page *src,
2397 unsigned long addr, struct vm_area_struct *vma,
2398 unsigned int pages_per_huge_page);
2399#endif /* CONFIG_TRANSPARENT_HUGEPAGE || CONFIG_HUGETLBFS */
2400
2401extern struct page_ext_operations debug_guardpage_ops;
2402extern struct page_ext_operations page_poisoning_ops;
2403
2404#ifdef CONFIG_DEBUG_PAGEALLOC
2405extern unsigned int _debug_guardpage_minorder;
2406extern bool _debug_guardpage_enabled;
2407
2408static inline unsigned int debug_guardpage_minorder(void)
2409{
2410 return _debug_guardpage_minorder;
2411}
2412
2413static inline bool debug_guardpage_enabled(void)
2414{
2415 return _debug_guardpage_enabled;
2416}
2417
2418static inline bool page_is_guard(struct page *page)
2419{
2420 struct page_ext *page_ext;
2421
2422 if (!debug_guardpage_enabled())
2423 return false;
2424
2425 page_ext = lookup_page_ext(page);
2426 if (unlikely(!page_ext))
2427 return false;
2428
2429 return test_bit(PAGE_EXT_DEBUG_GUARD, &page_ext->flags);
2430}
2431#else
2432static inline unsigned int debug_guardpage_minorder(void) { return 0; }
2433static inline bool debug_guardpage_enabled(void) { return false; }
2434static inline bool page_is_guard(struct page *page) { return false; }
2435#endif /* CONFIG_DEBUG_PAGEALLOC */
2436
2437#if MAX_NUMNODES > 1
2438void __init setup_nr_node_ids(void);
2439#else
2440static inline void setup_nr_node_ids(void) {}
2441#endif
2442
2443#endif /* __KERNEL__ */
2444#endif /* _LINUX_MM_H */
1/* SPDX-License-Identifier: GPL-2.0 */
2#ifndef _LINUX_MM_H
3#define _LINUX_MM_H
4
5#include <linux/errno.h>
6#include <linux/mmdebug.h>
7#include <linux/gfp.h>
8#include <linux/bug.h>
9#include <linux/list.h>
10#include <linux/mmzone.h>
11#include <linux/rbtree.h>
12#include <linux/atomic.h>
13#include <linux/debug_locks.h>
14#include <linux/mm_types.h>
15#include <linux/mmap_lock.h>
16#include <linux/range.h>
17#include <linux/pfn.h>
18#include <linux/percpu-refcount.h>
19#include <linux/bit_spinlock.h>
20#include <linux/shrinker.h>
21#include <linux/resource.h>
22#include <linux/page_ext.h>
23#include <linux/err.h>
24#include <linux/page-flags.h>
25#include <linux/page_ref.h>
26#include <linux/overflow.h>
27#include <linux/sizes.h>
28#include <linux/sched.h>
29#include <linux/pgtable.h>
30#include <linux/kasan.h>
31#include <linux/memremap.h>
32#include <linux/slab.h>
33
34struct mempolicy;
35struct anon_vma;
36struct anon_vma_chain;
37struct user_struct;
38struct pt_regs;
39
40extern int sysctl_page_lock_unfairness;
41
42void mm_core_init(void);
43void init_mm_internals(void);
44
45#ifndef CONFIG_NUMA /* Don't use mapnrs, do it properly */
46extern unsigned long max_mapnr;
47
48static inline void set_max_mapnr(unsigned long limit)
49{
50 max_mapnr = limit;
51}
52#else
53static inline void set_max_mapnr(unsigned long limit) { }
54#endif
55
56extern atomic_long_t _totalram_pages;
57static inline unsigned long totalram_pages(void)
58{
59 return (unsigned long)atomic_long_read(&_totalram_pages);
60}
61
62static inline void totalram_pages_inc(void)
63{
64 atomic_long_inc(&_totalram_pages);
65}
66
67static inline void totalram_pages_dec(void)
68{
69 atomic_long_dec(&_totalram_pages);
70}
71
72static inline void totalram_pages_add(long count)
73{
74 atomic_long_add(count, &_totalram_pages);
75}
76
77extern void * high_memory;
78extern int page_cluster;
79extern const int page_cluster_max;
80
81#ifdef CONFIG_SYSCTL
82extern int sysctl_legacy_va_layout;
83#else
84#define sysctl_legacy_va_layout 0
85#endif
86
87#ifdef CONFIG_HAVE_ARCH_MMAP_RND_BITS
88extern const int mmap_rnd_bits_min;
89extern const int mmap_rnd_bits_max;
90extern int mmap_rnd_bits __read_mostly;
91#endif
92#ifdef CONFIG_HAVE_ARCH_MMAP_RND_COMPAT_BITS
93extern const int mmap_rnd_compat_bits_min;
94extern const int mmap_rnd_compat_bits_max;
95extern int mmap_rnd_compat_bits __read_mostly;
96#endif
97
98#include <asm/page.h>
99#include <asm/processor.h>
100
101#ifndef __pa_symbol
102#define __pa_symbol(x) __pa(RELOC_HIDE((unsigned long)(x), 0))
103#endif
104
105#ifndef page_to_virt
106#define page_to_virt(x) __va(PFN_PHYS(page_to_pfn(x)))
107#endif
108
109#ifndef lm_alias
110#define lm_alias(x) __va(__pa_symbol(x))
111#endif
112
113/*
114 * To prevent common memory management code establishing
115 * a zero page mapping on a read fault.
116 * This macro should be defined within <asm/pgtable.h>.
117 * s390 does this to prevent multiplexing of hardware bits
118 * related to the physical page in case of virtualization.
119 */
120#ifndef mm_forbids_zeropage
121#define mm_forbids_zeropage(X) (0)
122#endif
123
124/*
125 * On some architectures it is expensive to call memset() for small sizes.
126 * If an architecture decides to implement their own version of
127 * mm_zero_struct_page they should wrap the defines below in a #ifndef and
128 * define their own version of this macro in <asm/pgtable.h>
129 */
130#if BITS_PER_LONG == 64
131/* This function must be updated when the size of struct page grows above 96
132 * or reduces below 56. The idea that compiler optimizes out switch()
133 * statement, and only leaves move/store instructions. Also the compiler can
134 * combine write statements if they are both assignments and can be reordered,
135 * this can result in several of the writes here being dropped.
136 */
137#define mm_zero_struct_page(pp) __mm_zero_struct_page(pp)
138static inline void __mm_zero_struct_page(struct page *page)
139{
140 unsigned long *_pp = (void *)page;
141
142 /* Check that struct page is either 56, 64, 72, 80, 88 or 96 bytes */
143 BUILD_BUG_ON(sizeof(struct page) & 7);
144 BUILD_BUG_ON(sizeof(struct page) < 56);
145 BUILD_BUG_ON(sizeof(struct page) > 96);
146
147 switch (sizeof(struct page)) {
148 case 96:
149 _pp[11] = 0;
150 fallthrough;
151 case 88:
152 _pp[10] = 0;
153 fallthrough;
154 case 80:
155 _pp[9] = 0;
156 fallthrough;
157 case 72:
158 _pp[8] = 0;
159 fallthrough;
160 case 64:
161 _pp[7] = 0;
162 fallthrough;
163 case 56:
164 _pp[6] = 0;
165 _pp[5] = 0;
166 _pp[4] = 0;
167 _pp[3] = 0;
168 _pp[2] = 0;
169 _pp[1] = 0;
170 _pp[0] = 0;
171 }
172}
173#else
174#define mm_zero_struct_page(pp) ((void)memset((pp), 0, sizeof(struct page)))
175#endif
176
177/*
178 * Default maximum number of active map areas, this limits the number of vmas
179 * per mm struct. Users can overwrite this number by sysctl but there is a
180 * problem.
181 *
182 * When a program's coredump is generated as ELF format, a section is created
183 * per a vma. In ELF, the number of sections is represented in unsigned short.
184 * This means the number of sections should be smaller than 65535 at coredump.
185 * Because the kernel adds some informative sections to a image of program at
186 * generating coredump, we need some margin. The number of extra sections is
187 * 1-3 now and depends on arch. We use "5" as safe margin, here.
188 *
189 * ELF extended numbering allows more than 65535 sections, so 16-bit bound is
190 * not a hard limit any more. Although some userspace tools can be surprised by
191 * that.
192 */
193#define MAPCOUNT_ELF_CORE_MARGIN (5)
194#define DEFAULT_MAX_MAP_COUNT (USHRT_MAX - MAPCOUNT_ELF_CORE_MARGIN)
195
196extern int sysctl_max_map_count;
197
198extern unsigned long sysctl_user_reserve_kbytes;
199extern unsigned long sysctl_admin_reserve_kbytes;
200
201extern int sysctl_overcommit_memory;
202extern int sysctl_overcommit_ratio;
203extern unsigned long sysctl_overcommit_kbytes;
204
205int overcommit_ratio_handler(struct ctl_table *, int, void *, size_t *,
206 loff_t *);
207int overcommit_kbytes_handler(struct ctl_table *, int, void *, size_t *,
208 loff_t *);
209int overcommit_policy_handler(struct ctl_table *, int, void *, size_t *,
210 loff_t *);
211
212#if defined(CONFIG_SPARSEMEM) && !defined(CONFIG_SPARSEMEM_VMEMMAP)
213#define nth_page(page,n) pfn_to_page(page_to_pfn((page)) + (n))
214#define folio_page_idx(folio, p) (page_to_pfn(p) - folio_pfn(folio))
215#else
216#define nth_page(page,n) ((page) + (n))
217#define folio_page_idx(folio, p) ((p) - &(folio)->page)
218#endif
219
220/* to align the pointer to the (next) page boundary */
221#define PAGE_ALIGN(addr) ALIGN(addr, PAGE_SIZE)
222
223/* to align the pointer to the (prev) page boundary */
224#define PAGE_ALIGN_DOWN(addr) ALIGN_DOWN(addr, PAGE_SIZE)
225
226/* test whether an address (unsigned long or pointer) is aligned to PAGE_SIZE */
227#define PAGE_ALIGNED(addr) IS_ALIGNED((unsigned long)(addr), PAGE_SIZE)
228
229#define lru_to_page(head) (list_entry((head)->prev, struct page, lru))
230static inline struct folio *lru_to_folio(struct list_head *head)
231{
232 return list_entry((head)->prev, struct folio, lru);
233}
234
235void setup_initial_init_mm(void *start_code, void *end_code,
236 void *end_data, void *brk);
237
238/*
239 * Linux kernel virtual memory manager primitives.
240 * The idea being to have a "virtual" mm in the same way
241 * we have a virtual fs - giving a cleaner interface to the
242 * mm details, and allowing different kinds of memory mappings
243 * (from shared memory to executable loading to arbitrary
244 * mmap() functions).
245 */
246
247struct vm_area_struct *vm_area_alloc(struct mm_struct *);
248struct vm_area_struct *vm_area_dup(struct vm_area_struct *);
249void vm_area_free(struct vm_area_struct *);
250/* Use only if VMA has no other users */
251void __vm_area_free(struct vm_area_struct *vma);
252
253#ifndef CONFIG_MMU
254extern struct rb_root nommu_region_tree;
255extern struct rw_semaphore nommu_region_sem;
256
257extern unsigned int kobjsize(const void *objp);
258#endif
259
260/*
261 * vm_flags in vm_area_struct, see mm_types.h.
262 * When changing, update also include/trace/events/mmflags.h
263 */
264#define VM_NONE 0x00000000
265
266#define VM_READ 0x00000001 /* currently active flags */
267#define VM_WRITE 0x00000002
268#define VM_EXEC 0x00000004
269#define VM_SHARED 0x00000008
270
271/* mprotect() hardcodes VM_MAYREAD >> 4 == VM_READ, and so for r/w/x bits. */
272#define VM_MAYREAD 0x00000010 /* limits for mprotect() etc */
273#define VM_MAYWRITE 0x00000020
274#define VM_MAYEXEC 0x00000040
275#define VM_MAYSHARE 0x00000080
276
277#define VM_GROWSDOWN 0x00000100 /* general info on the segment */
278#ifdef CONFIG_MMU
279#define VM_UFFD_MISSING 0x00000200 /* missing pages tracking */
280#else /* CONFIG_MMU */
281#define VM_MAYOVERLAY 0x00000200 /* nommu: R/O MAP_PRIVATE mapping that might overlay a file mapping */
282#define VM_UFFD_MISSING 0
283#endif /* CONFIG_MMU */
284#define VM_PFNMAP 0x00000400 /* Page-ranges managed without "struct page", just pure PFN */
285#define VM_UFFD_WP 0x00001000 /* wrprotect pages tracking */
286
287#define VM_LOCKED 0x00002000
288#define VM_IO 0x00004000 /* Memory mapped I/O or similar */
289
290 /* Used by sys_madvise() */
291#define VM_SEQ_READ 0x00008000 /* App will access data sequentially */
292#define VM_RAND_READ 0x00010000 /* App will not benefit from clustered reads */
293
294#define VM_DONTCOPY 0x00020000 /* Do not copy this vma on fork */
295#define VM_DONTEXPAND 0x00040000 /* Cannot expand with mremap() */
296#define VM_LOCKONFAULT 0x00080000 /* Lock the pages covered when they are faulted in */
297#define VM_ACCOUNT 0x00100000 /* Is a VM accounted object */
298#define VM_NORESERVE 0x00200000 /* should the VM suppress accounting */
299#define VM_HUGETLB 0x00400000 /* Huge TLB Page VM */
300#define VM_SYNC 0x00800000 /* Synchronous page faults */
301#define VM_ARCH_1 0x01000000 /* Architecture-specific flag */
302#define VM_WIPEONFORK 0x02000000 /* Wipe VMA contents in child. */
303#define VM_DONTDUMP 0x04000000 /* Do not include in the core dump */
304
305#ifdef CONFIG_MEM_SOFT_DIRTY
306# define VM_SOFTDIRTY 0x08000000 /* Not soft dirty clean area */
307#else
308# define VM_SOFTDIRTY 0
309#endif
310
311#define VM_MIXEDMAP 0x10000000 /* Can contain "struct page" and pure PFN pages */
312#define VM_HUGEPAGE 0x20000000 /* MADV_HUGEPAGE marked this vma */
313#define VM_NOHUGEPAGE 0x40000000 /* MADV_NOHUGEPAGE marked this vma */
314#define VM_MERGEABLE 0x80000000 /* KSM may merge identical pages */
315
316#ifdef CONFIG_ARCH_USES_HIGH_VMA_FLAGS
317#define VM_HIGH_ARCH_BIT_0 32 /* bit only usable on 64-bit architectures */
318#define VM_HIGH_ARCH_BIT_1 33 /* bit only usable on 64-bit architectures */
319#define VM_HIGH_ARCH_BIT_2 34 /* bit only usable on 64-bit architectures */
320#define VM_HIGH_ARCH_BIT_3 35 /* bit only usable on 64-bit architectures */
321#define VM_HIGH_ARCH_BIT_4 36 /* bit only usable on 64-bit architectures */
322#define VM_HIGH_ARCH_BIT_5 37 /* bit only usable on 64-bit architectures */
323#define VM_HIGH_ARCH_0 BIT(VM_HIGH_ARCH_BIT_0)
324#define VM_HIGH_ARCH_1 BIT(VM_HIGH_ARCH_BIT_1)
325#define VM_HIGH_ARCH_2 BIT(VM_HIGH_ARCH_BIT_2)
326#define VM_HIGH_ARCH_3 BIT(VM_HIGH_ARCH_BIT_3)
327#define VM_HIGH_ARCH_4 BIT(VM_HIGH_ARCH_BIT_4)
328#define VM_HIGH_ARCH_5 BIT(VM_HIGH_ARCH_BIT_5)
329#endif /* CONFIG_ARCH_USES_HIGH_VMA_FLAGS */
330
331#ifdef CONFIG_ARCH_HAS_PKEYS
332# define VM_PKEY_SHIFT VM_HIGH_ARCH_BIT_0
333# define VM_PKEY_BIT0 VM_HIGH_ARCH_0 /* A protection key is a 4-bit value */
334# define VM_PKEY_BIT1 VM_HIGH_ARCH_1 /* on x86 and 5-bit value on ppc64 */
335# define VM_PKEY_BIT2 VM_HIGH_ARCH_2
336# define VM_PKEY_BIT3 VM_HIGH_ARCH_3
337#ifdef CONFIG_PPC
338# define VM_PKEY_BIT4 VM_HIGH_ARCH_4
339#else
340# define VM_PKEY_BIT4 0
341#endif
342#endif /* CONFIG_ARCH_HAS_PKEYS */
343
344#ifdef CONFIG_X86_USER_SHADOW_STACK
345/*
346 * VM_SHADOW_STACK should not be set with VM_SHARED because of lack of
347 * support core mm.
348 *
349 * These VMAs will get a single end guard page. This helps userspace protect
350 * itself from attacks. A single page is enough for current shadow stack archs
351 * (x86). See the comments near alloc_shstk() in arch/x86/kernel/shstk.c
352 * for more details on the guard size.
353 */
354# define VM_SHADOW_STACK VM_HIGH_ARCH_5
355#else
356# define VM_SHADOW_STACK VM_NONE
357#endif
358
359#if defined(CONFIG_X86)
360# define VM_PAT VM_ARCH_1 /* PAT reserves whole VMA at once (x86) */
361#elif defined(CONFIG_PPC)
362# define VM_SAO VM_ARCH_1 /* Strong Access Ordering (powerpc) */
363#elif defined(CONFIG_PARISC)
364# define VM_GROWSUP VM_ARCH_1
365#elif defined(CONFIG_SPARC64)
366# define VM_SPARC_ADI VM_ARCH_1 /* Uses ADI tag for access control */
367# define VM_ARCH_CLEAR VM_SPARC_ADI
368#elif defined(CONFIG_ARM64)
369# define VM_ARM64_BTI VM_ARCH_1 /* BTI guarded page, a.k.a. GP bit */
370# define VM_ARCH_CLEAR VM_ARM64_BTI
371#elif !defined(CONFIG_MMU)
372# define VM_MAPPED_COPY VM_ARCH_1 /* T if mapped copy of data (nommu mmap) */
373#endif
374
375#if defined(CONFIG_ARM64_MTE)
376# define VM_MTE VM_HIGH_ARCH_0 /* Use Tagged memory for access control */
377# define VM_MTE_ALLOWED VM_HIGH_ARCH_1 /* Tagged memory permitted */
378#else
379# define VM_MTE VM_NONE
380# define VM_MTE_ALLOWED VM_NONE
381#endif
382
383#ifndef VM_GROWSUP
384# define VM_GROWSUP VM_NONE
385#endif
386
387#ifdef CONFIG_HAVE_ARCH_USERFAULTFD_MINOR
388# define VM_UFFD_MINOR_BIT 38
389# define VM_UFFD_MINOR BIT(VM_UFFD_MINOR_BIT) /* UFFD minor faults */
390#else /* !CONFIG_HAVE_ARCH_USERFAULTFD_MINOR */
391# define VM_UFFD_MINOR VM_NONE
392#endif /* CONFIG_HAVE_ARCH_USERFAULTFD_MINOR */
393
394/* Bits set in the VMA until the stack is in its final location */
395#define VM_STACK_INCOMPLETE_SETUP (VM_RAND_READ | VM_SEQ_READ | VM_STACK_EARLY)
396
397#define TASK_EXEC ((current->personality & READ_IMPLIES_EXEC) ? VM_EXEC : 0)
398
399/* Common data flag combinations */
400#define VM_DATA_FLAGS_TSK_EXEC (VM_READ | VM_WRITE | TASK_EXEC | \
401 VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC)
402#define VM_DATA_FLAGS_NON_EXEC (VM_READ | VM_WRITE | VM_MAYREAD | \
403 VM_MAYWRITE | VM_MAYEXEC)
404#define VM_DATA_FLAGS_EXEC (VM_READ | VM_WRITE | VM_EXEC | \
405 VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC)
406
407#ifndef VM_DATA_DEFAULT_FLAGS /* arch can override this */
408#define VM_DATA_DEFAULT_FLAGS VM_DATA_FLAGS_EXEC
409#endif
410
411#ifndef VM_STACK_DEFAULT_FLAGS /* arch can override this */
412#define VM_STACK_DEFAULT_FLAGS VM_DATA_DEFAULT_FLAGS
413#endif
414
415#define VM_STARTGAP_FLAGS (VM_GROWSDOWN | VM_SHADOW_STACK)
416
417#ifdef CONFIG_STACK_GROWSUP
418#define VM_STACK VM_GROWSUP
419#define VM_STACK_EARLY VM_GROWSDOWN
420#else
421#define VM_STACK VM_GROWSDOWN
422#define VM_STACK_EARLY 0
423#endif
424
425#define VM_STACK_FLAGS (VM_STACK | VM_STACK_DEFAULT_FLAGS | VM_ACCOUNT)
426
427/* VMA basic access permission flags */
428#define VM_ACCESS_FLAGS (VM_READ | VM_WRITE | VM_EXEC)
429
430
431/*
432 * Special vmas that are non-mergable, non-mlock()able.
433 */
434#define VM_SPECIAL (VM_IO | VM_DONTEXPAND | VM_PFNMAP | VM_MIXEDMAP)
435
436/* This mask prevents VMA from being scanned with khugepaged */
437#define VM_NO_KHUGEPAGED (VM_SPECIAL | VM_HUGETLB)
438
439/* This mask defines which mm->def_flags a process can inherit its parent */
440#define VM_INIT_DEF_MASK VM_NOHUGEPAGE
441
442/* This mask represents all the VMA flag bits used by mlock */
443#define VM_LOCKED_MASK (VM_LOCKED | VM_LOCKONFAULT)
444
445/* Arch-specific flags to clear when updating VM flags on protection change */
446#ifndef VM_ARCH_CLEAR
447# define VM_ARCH_CLEAR VM_NONE
448#endif
449#define VM_FLAGS_CLEAR (ARCH_VM_PKEY_FLAGS | VM_ARCH_CLEAR)
450
451/*
452 * mapping from the currently active vm_flags protection bits (the
453 * low four bits) to a page protection mask..
454 */
455
456/*
457 * The default fault flags that should be used by most of the
458 * arch-specific page fault handlers.
459 */
460#define FAULT_FLAG_DEFAULT (FAULT_FLAG_ALLOW_RETRY | \
461 FAULT_FLAG_KILLABLE | \
462 FAULT_FLAG_INTERRUPTIBLE)
463
464/**
465 * fault_flag_allow_retry_first - check ALLOW_RETRY the first time
466 * @flags: Fault flags.
467 *
468 * This is mostly used for places where we want to try to avoid taking
469 * the mmap_lock for too long a time when waiting for another condition
470 * to change, in which case we can try to be polite to release the
471 * mmap_lock in the first round to avoid potential starvation of other
472 * processes that would also want the mmap_lock.
473 *
474 * Return: true if the page fault allows retry and this is the first
475 * attempt of the fault handling; false otherwise.
476 */
477static inline bool fault_flag_allow_retry_first(enum fault_flag flags)
478{
479 return (flags & FAULT_FLAG_ALLOW_RETRY) &&
480 (!(flags & FAULT_FLAG_TRIED));
481}
482
483#define FAULT_FLAG_TRACE \
484 { FAULT_FLAG_WRITE, "WRITE" }, \
485 { FAULT_FLAG_MKWRITE, "MKWRITE" }, \
486 { FAULT_FLAG_ALLOW_RETRY, "ALLOW_RETRY" }, \
487 { FAULT_FLAG_RETRY_NOWAIT, "RETRY_NOWAIT" }, \
488 { FAULT_FLAG_KILLABLE, "KILLABLE" }, \
489 { FAULT_FLAG_TRIED, "TRIED" }, \
490 { FAULT_FLAG_USER, "USER" }, \
491 { FAULT_FLAG_REMOTE, "REMOTE" }, \
492 { FAULT_FLAG_INSTRUCTION, "INSTRUCTION" }, \
493 { FAULT_FLAG_INTERRUPTIBLE, "INTERRUPTIBLE" }, \
494 { FAULT_FLAG_VMA_LOCK, "VMA_LOCK" }
495
496/*
497 * vm_fault is filled by the pagefault handler and passed to the vma's
498 * ->fault function. The vma's ->fault is responsible for returning a bitmask
499 * of VM_FAULT_xxx flags that give details about how the fault was handled.
500 *
501 * MM layer fills up gfp_mask for page allocations but fault handler might
502 * alter it if its implementation requires a different allocation context.
503 *
504 * pgoff should be used in favour of virtual_address, if possible.
505 */
506struct vm_fault {
507 const struct {
508 struct vm_area_struct *vma; /* Target VMA */
509 gfp_t gfp_mask; /* gfp mask to be used for allocations */
510 pgoff_t pgoff; /* Logical page offset based on vma */
511 unsigned long address; /* Faulting virtual address - masked */
512 unsigned long real_address; /* Faulting virtual address - unmasked */
513 };
514 enum fault_flag flags; /* FAULT_FLAG_xxx flags
515 * XXX: should really be 'const' */
516 pmd_t *pmd; /* Pointer to pmd entry matching
517 * the 'address' */
518 pud_t *pud; /* Pointer to pud entry matching
519 * the 'address'
520 */
521 union {
522 pte_t orig_pte; /* Value of PTE at the time of fault */
523 pmd_t orig_pmd; /* Value of PMD at the time of fault,
524 * used by PMD fault only.
525 */
526 };
527
528 struct page *cow_page; /* Page handler may use for COW fault */
529 struct page *page; /* ->fault handlers should return a
530 * page here, unless VM_FAULT_NOPAGE
531 * is set (which is also implied by
532 * VM_FAULT_ERROR).
533 */
534 /* These three entries are valid only while holding ptl lock */
535 pte_t *pte; /* Pointer to pte entry matching
536 * the 'address'. NULL if the page
537 * table hasn't been allocated.
538 */
539 spinlock_t *ptl; /* Page table lock.
540 * Protects pte page table if 'pte'
541 * is not NULL, otherwise pmd.
542 */
543 pgtable_t prealloc_pte; /* Pre-allocated pte page table.
544 * vm_ops->map_pages() sets up a page
545 * table from atomic context.
546 * do_fault_around() pre-allocates
547 * page table to avoid allocation from
548 * atomic context.
549 */
550};
551
552/*
553 * These are the virtual MM functions - opening of an area, closing and
554 * unmapping it (needed to keep files on disk up-to-date etc), pointer
555 * to the functions called when a no-page or a wp-page exception occurs.
556 */
557struct vm_operations_struct {
558 void (*open)(struct vm_area_struct * area);
559 /**
560 * @close: Called when the VMA is being removed from the MM.
561 * Context: User context. May sleep. Caller holds mmap_lock.
562 */
563 void (*close)(struct vm_area_struct * area);
564 /* Called any time before splitting to check if it's allowed */
565 int (*may_split)(struct vm_area_struct *area, unsigned long addr);
566 int (*mremap)(struct vm_area_struct *area);
567 /*
568 * Called by mprotect() to make driver-specific permission
569 * checks before mprotect() is finalised. The VMA must not
570 * be modified. Returns 0 if mprotect() can proceed.
571 */
572 int (*mprotect)(struct vm_area_struct *vma, unsigned long start,
573 unsigned long end, unsigned long newflags);
574 vm_fault_t (*fault)(struct vm_fault *vmf);
575 vm_fault_t (*huge_fault)(struct vm_fault *vmf, unsigned int order);
576 vm_fault_t (*map_pages)(struct vm_fault *vmf,
577 pgoff_t start_pgoff, pgoff_t end_pgoff);
578 unsigned long (*pagesize)(struct vm_area_struct * area);
579
580 /* notification that a previously read-only page is about to become
581 * writable, if an error is returned it will cause a SIGBUS */
582 vm_fault_t (*page_mkwrite)(struct vm_fault *vmf);
583
584 /* same as page_mkwrite when using VM_PFNMAP|VM_MIXEDMAP */
585 vm_fault_t (*pfn_mkwrite)(struct vm_fault *vmf);
586
587 /* called by access_process_vm when get_user_pages() fails, typically
588 * for use by special VMAs. See also generic_access_phys() for a generic
589 * implementation useful for any iomem mapping.
590 */
591 int (*access)(struct vm_area_struct *vma, unsigned long addr,
592 void *buf, int len, int write);
593
594 /* Called by the /proc/PID/maps code to ask the vma whether it
595 * has a special name. Returning non-NULL will also cause this
596 * vma to be dumped unconditionally. */
597 const char *(*name)(struct vm_area_struct *vma);
598
599#ifdef CONFIG_NUMA
600 /*
601 * set_policy() op must add a reference to any non-NULL @new mempolicy
602 * to hold the policy upon return. Caller should pass NULL @new to
603 * remove a policy and fall back to surrounding context--i.e. do not
604 * install a MPOL_DEFAULT policy, nor the task or system default
605 * mempolicy.
606 */
607 int (*set_policy)(struct vm_area_struct *vma, struct mempolicy *new);
608
609 /*
610 * get_policy() op must add reference [mpol_get()] to any policy at
611 * (vma,addr) marked as MPOL_SHARED. The shared policy infrastructure
612 * in mm/mempolicy.c will do this automatically.
613 * get_policy() must NOT add a ref if the policy at (vma,addr) is not
614 * marked as MPOL_SHARED. vma policies are protected by the mmap_lock.
615 * If no [shared/vma] mempolicy exists at the addr, get_policy() op
616 * must return NULL--i.e., do not "fallback" to task or system default
617 * policy.
618 */
619 struct mempolicy *(*get_policy)(struct vm_area_struct *vma,
620 unsigned long addr, pgoff_t *ilx);
621#endif
622 /*
623 * Called by vm_normal_page() for special PTEs to find the
624 * page for @addr. This is useful if the default behavior
625 * (using pte_page()) would not find the correct page.
626 */
627 struct page *(*find_special_page)(struct vm_area_struct *vma,
628 unsigned long addr);
629};
630
631#ifdef CONFIG_NUMA_BALANCING
632static inline void vma_numab_state_init(struct vm_area_struct *vma)
633{
634 vma->numab_state = NULL;
635}
636static inline void vma_numab_state_free(struct vm_area_struct *vma)
637{
638 kfree(vma->numab_state);
639}
640#else
641static inline void vma_numab_state_init(struct vm_area_struct *vma) {}
642static inline void vma_numab_state_free(struct vm_area_struct *vma) {}
643#endif /* CONFIG_NUMA_BALANCING */
644
645#ifdef CONFIG_PER_VMA_LOCK
646/*
647 * Try to read-lock a vma. The function is allowed to occasionally yield false
648 * locked result to avoid performance overhead, in which case we fall back to
649 * using mmap_lock. The function should never yield false unlocked result.
650 */
651static inline bool vma_start_read(struct vm_area_struct *vma)
652{
653 /*
654 * Check before locking. A race might cause false locked result.
655 * We can use READ_ONCE() for the mm_lock_seq here, and don't need
656 * ACQUIRE semantics, because this is just a lockless check whose result
657 * we don't rely on for anything - the mm_lock_seq read against which we
658 * need ordering is below.
659 */
660 if (READ_ONCE(vma->vm_lock_seq) == READ_ONCE(vma->vm_mm->mm_lock_seq))
661 return false;
662
663 if (unlikely(down_read_trylock(&vma->vm_lock->lock) == 0))
664 return false;
665
666 /*
667 * Overflow might produce false locked result.
668 * False unlocked result is impossible because we modify and check
669 * vma->vm_lock_seq under vma->vm_lock protection and mm->mm_lock_seq
670 * modification invalidates all existing locks.
671 *
672 * We must use ACQUIRE semantics for the mm_lock_seq so that if we are
673 * racing with vma_end_write_all(), we only start reading from the VMA
674 * after it has been unlocked.
675 * This pairs with RELEASE semantics in vma_end_write_all().
676 */
677 if (unlikely(vma->vm_lock_seq == smp_load_acquire(&vma->vm_mm->mm_lock_seq))) {
678 up_read(&vma->vm_lock->lock);
679 return false;
680 }
681 return true;
682}
683
684static inline void vma_end_read(struct vm_area_struct *vma)
685{
686 rcu_read_lock(); /* keeps vma alive till the end of up_read */
687 up_read(&vma->vm_lock->lock);
688 rcu_read_unlock();
689}
690
691/* WARNING! Can only be used if mmap_lock is expected to be write-locked */
692static bool __is_vma_write_locked(struct vm_area_struct *vma, int *mm_lock_seq)
693{
694 mmap_assert_write_locked(vma->vm_mm);
695
696 /*
697 * current task is holding mmap_write_lock, both vma->vm_lock_seq and
698 * mm->mm_lock_seq can't be concurrently modified.
699 */
700 *mm_lock_seq = vma->vm_mm->mm_lock_seq;
701 return (vma->vm_lock_seq == *mm_lock_seq);
702}
703
704/*
705 * Begin writing to a VMA.
706 * Exclude concurrent readers under the per-VMA lock until the currently
707 * write-locked mmap_lock is dropped or downgraded.
708 */
709static inline void vma_start_write(struct vm_area_struct *vma)
710{
711 int mm_lock_seq;
712
713 if (__is_vma_write_locked(vma, &mm_lock_seq))
714 return;
715
716 down_write(&vma->vm_lock->lock);
717 /*
718 * We should use WRITE_ONCE() here because we can have concurrent reads
719 * from the early lockless pessimistic check in vma_start_read().
720 * We don't really care about the correctness of that early check, but
721 * we should use WRITE_ONCE() for cleanliness and to keep KCSAN happy.
722 */
723 WRITE_ONCE(vma->vm_lock_seq, mm_lock_seq);
724 up_write(&vma->vm_lock->lock);
725}
726
727static inline void vma_assert_write_locked(struct vm_area_struct *vma)
728{
729 int mm_lock_seq;
730
731 VM_BUG_ON_VMA(!__is_vma_write_locked(vma, &mm_lock_seq), vma);
732}
733
734static inline void vma_assert_locked(struct vm_area_struct *vma)
735{
736 if (!rwsem_is_locked(&vma->vm_lock->lock))
737 vma_assert_write_locked(vma);
738}
739
740static inline void vma_mark_detached(struct vm_area_struct *vma, bool detached)
741{
742 /* When detaching vma should be write-locked */
743 if (detached)
744 vma_assert_write_locked(vma);
745 vma->detached = detached;
746}
747
748static inline void release_fault_lock(struct vm_fault *vmf)
749{
750 if (vmf->flags & FAULT_FLAG_VMA_LOCK)
751 vma_end_read(vmf->vma);
752 else
753 mmap_read_unlock(vmf->vma->vm_mm);
754}
755
756static inline void assert_fault_locked(struct vm_fault *vmf)
757{
758 if (vmf->flags & FAULT_FLAG_VMA_LOCK)
759 vma_assert_locked(vmf->vma);
760 else
761 mmap_assert_locked(vmf->vma->vm_mm);
762}
763
764struct vm_area_struct *lock_vma_under_rcu(struct mm_struct *mm,
765 unsigned long address);
766
767#else /* CONFIG_PER_VMA_LOCK */
768
769static inline bool vma_start_read(struct vm_area_struct *vma)
770 { return false; }
771static inline void vma_end_read(struct vm_area_struct *vma) {}
772static inline void vma_start_write(struct vm_area_struct *vma) {}
773static inline void vma_assert_write_locked(struct vm_area_struct *vma)
774 { mmap_assert_write_locked(vma->vm_mm); }
775static inline void vma_mark_detached(struct vm_area_struct *vma,
776 bool detached) {}
777
778static inline struct vm_area_struct *lock_vma_under_rcu(struct mm_struct *mm,
779 unsigned long address)
780{
781 return NULL;
782}
783
784static inline void release_fault_lock(struct vm_fault *vmf)
785{
786 mmap_read_unlock(vmf->vma->vm_mm);
787}
788
789static inline void assert_fault_locked(struct vm_fault *vmf)
790{
791 mmap_assert_locked(vmf->vma->vm_mm);
792}
793
794#endif /* CONFIG_PER_VMA_LOCK */
795
796extern const struct vm_operations_struct vma_dummy_vm_ops;
797
798/*
799 * WARNING: vma_init does not initialize vma->vm_lock.
800 * Use vm_area_alloc()/vm_area_free() if vma needs locking.
801 */
802static inline void vma_init(struct vm_area_struct *vma, struct mm_struct *mm)
803{
804 memset(vma, 0, sizeof(*vma));
805 vma->vm_mm = mm;
806 vma->vm_ops = &vma_dummy_vm_ops;
807 INIT_LIST_HEAD(&vma->anon_vma_chain);
808 vma_mark_detached(vma, false);
809 vma_numab_state_init(vma);
810}
811
812/* Use when VMA is not part of the VMA tree and needs no locking */
813static inline void vm_flags_init(struct vm_area_struct *vma,
814 vm_flags_t flags)
815{
816 ACCESS_PRIVATE(vma, __vm_flags) = flags;
817}
818
819/*
820 * Use when VMA is part of the VMA tree and modifications need coordination
821 * Note: vm_flags_reset and vm_flags_reset_once do not lock the vma and
822 * it should be locked explicitly beforehand.
823 */
824static inline void vm_flags_reset(struct vm_area_struct *vma,
825 vm_flags_t flags)
826{
827 vma_assert_write_locked(vma);
828 vm_flags_init(vma, flags);
829}
830
831static inline void vm_flags_reset_once(struct vm_area_struct *vma,
832 vm_flags_t flags)
833{
834 vma_assert_write_locked(vma);
835 WRITE_ONCE(ACCESS_PRIVATE(vma, __vm_flags), flags);
836}
837
838static inline void vm_flags_set(struct vm_area_struct *vma,
839 vm_flags_t flags)
840{
841 vma_start_write(vma);
842 ACCESS_PRIVATE(vma, __vm_flags) |= flags;
843}
844
845static inline void vm_flags_clear(struct vm_area_struct *vma,
846 vm_flags_t flags)
847{
848 vma_start_write(vma);
849 ACCESS_PRIVATE(vma, __vm_flags) &= ~flags;
850}
851
852/*
853 * Use only if VMA is not part of the VMA tree or has no other users and
854 * therefore needs no locking.
855 */
856static inline void __vm_flags_mod(struct vm_area_struct *vma,
857 vm_flags_t set, vm_flags_t clear)
858{
859 vm_flags_init(vma, (vma->vm_flags | set) & ~clear);
860}
861
862/*
863 * Use only when the order of set/clear operations is unimportant, otherwise
864 * use vm_flags_{set|clear} explicitly.
865 */
866static inline void vm_flags_mod(struct vm_area_struct *vma,
867 vm_flags_t set, vm_flags_t clear)
868{
869 vma_start_write(vma);
870 __vm_flags_mod(vma, set, clear);
871}
872
873static inline void vma_set_anonymous(struct vm_area_struct *vma)
874{
875 vma->vm_ops = NULL;
876}
877
878static inline bool vma_is_anonymous(struct vm_area_struct *vma)
879{
880 return !vma->vm_ops;
881}
882
883/*
884 * Indicate if the VMA is a heap for the given task; for
885 * /proc/PID/maps that is the heap of the main task.
886 */
887static inline bool vma_is_initial_heap(const struct vm_area_struct *vma)
888{
889 return vma->vm_start < vma->vm_mm->brk &&
890 vma->vm_end > vma->vm_mm->start_brk;
891}
892
893/*
894 * Indicate if the VMA is a stack for the given task; for
895 * /proc/PID/maps that is the stack of the main task.
896 */
897static inline bool vma_is_initial_stack(const struct vm_area_struct *vma)
898{
899 /*
900 * We make no effort to guess what a given thread considers to be
901 * its "stack". It's not even well-defined for programs written
902 * languages like Go.
903 */
904 return vma->vm_start <= vma->vm_mm->start_stack &&
905 vma->vm_end >= vma->vm_mm->start_stack;
906}
907
908static inline bool vma_is_temporary_stack(struct vm_area_struct *vma)
909{
910 int maybe_stack = vma->vm_flags & (VM_GROWSDOWN | VM_GROWSUP);
911
912 if (!maybe_stack)
913 return false;
914
915 if ((vma->vm_flags & VM_STACK_INCOMPLETE_SETUP) ==
916 VM_STACK_INCOMPLETE_SETUP)
917 return true;
918
919 return false;
920}
921
922static inline bool vma_is_foreign(struct vm_area_struct *vma)
923{
924 if (!current->mm)
925 return true;
926
927 if (current->mm != vma->vm_mm)
928 return true;
929
930 return false;
931}
932
933static inline bool vma_is_accessible(struct vm_area_struct *vma)
934{
935 return vma->vm_flags & VM_ACCESS_FLAGS;
936}
937
938static inline bool is_shared_maywrite(vm_flags_t vm_flags)
939{
940 return (vm_flags & (VM_SHARED | VM_MAYWRITE)) ==
941 (VM_SHARED | VM_MAYWRITE);
942}
943
944static inline bool vma_is_shared_maywrite(struct vm_area_struct *vma)
945{
946 return is_shared_maywrite(vma->vm_flags);
947}
948
949static inline
950struct vm_area_struct *vma_find(struct vma_iterator *vmi, unsigned long max)
951{
952 return mas_find(&vmi->mas, max - 1);
953}
954
955static inline struct vm_area_struct *vma_next(struct vma_iterator *vmi)
956{
957 /*
958 * Uses mas_find() to get the first VMA when the iterator starts.
959 * Calling mas_next() could skip the first entry.
960 */
961 return mas_find(&vmi->mas, ULONG_MAX);
962}
963
964static inline
965struct vm_area_struct *vma_iter_next_range(struct vma_iterator *vmi)
966{
967 return mas_next_range(&vmi->mas, ULONG_MAX);
968}
969
970
971static inline struct vm_area_struct *vma_prev(struct vma_iterator *vmi)
972{
973 return mas_prev(&vmi->mas, 0);
974}
975
976static inline
977struct vm_area_struct *vma_iter_prev_range(struct vma_iterator *vmi)
978{
979 return mas_prev_range(&vmi->mas, 0);
980}
981
982static inline unsigned long vma_iter_addr(struct vma_iterator *vmi)
983{
984 return vmi->mas.index;
985}
986
987static inline unsigned long vma_iter_end(struct vma_iterator *vmi)
988{
989 return vmi->mas.last + 1;
990}
991static inline int vma_iter_bulk_alloc(struct vma_iterator *vmi,
992 unsigned long count)
993{
994 return mas_expected_entries(&vmi->mas, count);
995}
996
997static inline int vma_iter_clear_gfp(struct vma_iterator *vmi,
998 unsigned long start, unsigned long end, gfp_t gfp)
999{
1000 __mas_set_range(&vmi->mas, start, end - 1);
1001 mas_store_gfp(&vmi->mas, NULL, gfp);
1002 if (unlikely(mas_is_err(&vmi->mas)))
1003 return -ENOMEM;
1004
1005 return 0;
1006}
1007
1008/* Free any unused preallocations */
1009static inline void vma_iter_free(struct vma_iterator *vmi)
1010{
1011 mas_destroy(&vmi->mas);
1012}
1013
1014static inline int vma_iter_bulk_store(struct vma_iterator *vmi,
1015 struct vm_area_struct *vma)
1016{
1017 vmi->mas.index = vma->vm_start;
1018 vmi->mas.last = vma->vm_end - 1;
1019 mas_store(&vmi->mas, vma);
1020 if (unlikely(mas_is_err(&vmi->mas)))
1021 return -ENOMEM;
1022
1023 return 0;
1024}
1025
1026static inline void vma_iter_invalidate(struct vma_iterator *vmi)
1027{
1028 mas_pause(&vmi->mas);
1029}
1030
1031static inline void vma_iter_set(struct vma_iterator *vmi, unsigned long addr)
1032{
1033 mas_set(&vmi->mas, addr);
1034}
1035
1036#define for_each_vma(__vmi, __vma) \
1037 while (((__vma) = vma_next(&(__vmi))) != NULL)
1038
1039/* The MM code likes to work with exclusive end addresses */
1040#define for_each_vma_range(__vmi, __vma, __end) \
1041 while (((__vma) = vma_find(&(__vmi), (__end))) != NULL)
1042
1043#ifdef CONFIG_SHMEM
1044/*
1045 * The vma_is_shmem is not inline because it is used only by slow
1046 * paths in userfault.
1047 */
1048bool vma_is_shmem(struct vm_area_struct *vma);
1049bool vma_is_anon_shmem(struct vm_area_struct *vma);
1050#else
1051static inline bool vma_is_shmem(struct vm_area_struct *vma) { return false; }
1052static inline bool vma_is_anon_shmem(struct vm_area_struct *vma) { return false; }
1053#endif
1054
1055int vma_is_stack_for_current(struct vm_area_struct *vma);
1056
1057/* flush_tlb_range() takes a vma, not a mm, and can care about flags */
1058#define TLB_FLUSH_VMA(mm,flags) { .vm_mm = (mm), .vm_flags = (flags) }
1059
1060struct mmu_gather;
1061struct inode;
1062
1063/*
1064 * compound_order() can be called without holding a reference, which means
1065 * that niceties like page_folio() don't work. These callers should be
1066 * prepared to handle wild return values. For example, PG_head may be
1067 * set before the order is initialised, or this may be a tail page.
1068 * See compaction.c for some good examples.
1069 */
1070static inline unsigned int compound_order(struct page *page)
1071{
1072 struct folio *folio = (struct folio *)page;
1073
1074 if (!test_bit(PG_head, &folio->flags))
1075 return 0;
1076 return folio->_flags_1 & 0xff;
1077}
1078
1079/**
1080 * folio_order - The allocation order of a folio.
1081 * @folio: The folio.
1082 *
1083 * A folio is composed of 2^order pages. See get_order() for the definition
1084 * of order.
1085 *
1086 * Return: The order of the folio.
1087 */
1088static inline unsigned int folio_order(struct folio *folio)
1089{
1090 if (!folio_test_large(folio))
1091 return 0;
1092 return folio->_flags_1 & 0xff;
1093}
1094
1095#include <linux/huge_mm.h>
1096
1097/*
1098 * Methods to modify the page usage count.
1099 *
1100 * What counts for a page usage:
1101 * - cache mapping (page->mapping)
1102 * - private data (page->private)
1103 * - page mapped in a task's page tables, each mapping
1104 * is counted separately
1105 *
1106 * Also, many kernel routines increase the page count before a critical
1107 * routine so they can be sure the page doesn't go away from under them.
1108 */
1109
1110/*
1111 * Drop a ref, return true if the refcount fell to zero (the page has no users)
1112 */
1113static inline int put_page_testzero(struct page *page)
1114{
1115 VM_BUG_ON_PAGE(page_ref_count(page) == 0, page);
1116 return page_ref_dec_and_test(page);
1117}
1118
1119static inline int folio_put_testzero(struct folio *folio)
1120{
1121 return put_page_testzero(&folio->page);
1122}
1123
1124/*
1125 * Try to grab a ref unless the page has a refcount of zero, return false if
1126 * that is the case.
1127 * This can be called when MMU is off so it must not access
1128 * any of the virtual mappings.
1129 */
1130static inline bool get_page_unless_zero(struct page *page)
1131{
1132 return page_ref_add_unless(page, 1, 0);
1133}
1134
1135static inline struct folio *folio_get_nontail_page(struct page *page)
1136{
1137 if (unlikely(!get_page_unless_zero(page)))
1138 return NULL;
1139 return (struct folio *)page;
1140}
1141
1142extern int page_is_ram(unsigned long pfn);
1143
1144enum {
1145 REGION_INTERSECTS,
1146 REGION_DISJOINT,
1147 REGION_MIXED,
1148};
1149
1150int region_intersects(resource_size_t offset, size_t size, unsigned long flags,
1151 unsigned long desc);
1152
1153/* Support for virtually mapped pages */
1154struct page *vmalloc_to_page(const void *addr);
1155unsigned long vmalloc_to_pfn(const void *addr);
1156
1157/*
1158 * Determine if an address is within the vmalloc range
1159 *
1160 * On nommu, vmalloc/vfree wrap through kmalloc/kfree directly, so there
1161 * is no special casing required.
1162 */
1163#ifdef CONFIG_MMU
1164extern bool is_vmalloc_addr(const void *x);
1165extern int is_vmalloc_or_module_addr(const void *x);
1166#else
1167static inline bool is_vmalloc_addr(const void *x)
1168{
1169 return false;
1170}
1171static inline int is_vmalloc_or_module_addr(const void *x)
1172{
1173 return 0;
1174}
1175#endif
1176
1177/*
1178 * How many times the entire folio is mapped as a single unit (eg by a
1179 * PMD or PUD entry). This is probably not what you want, except for
1180 * debugging purposes - it does not include PTE-mapped sub-pages; look
1181 * at folio_mapcount() or page_mapcount() or total_mapcount() instead.
1182 */
1183static inline int folio_entire_mapcount(struct folio *folio)
1184{
1185 VM_BUG_ON_FOLIO(!folio_test_large(folio), folio);
1186 return atomic_read(&folio->_entire_mapcount) + 1;
1187}
1188
1189/*
1190 * The atomic page->_mapcount, starts from -1: so that transitions
1191 * both from it and to it can be tracked, using atomic_inc_and_test
1192 * and atomic_add_negative(-1).
1193 */
1194static inline void page_mapcount_reset(struct page *page)
1195{
1196 atomic_set(&(page)->_mapcount, -1);
1197}
1198
1199/**
1200 * page_mapcount() - Number of times this precise page is mapped.
1201 * @page: The page.
1202 *
1203 * The number of times this page is mapped. If this page is part of
1204 * a large folio, it includes the number of times this page is mapped
1205 * as part of that folio.
1206 *
1207 * The result is undefined for pages which cannot be mapped into userspace.
1208 * For example SLAB or special types of pages. See function page_has_type().
1209 * They use this field in struct page differently.
1210 */
1211static inline int page_mapcount(struct page *page)
1212{
1213 int mapcount = atomic_read(&page->_mapcount) + 1;
1214
1215 if (unlikely(PageCompound(page)))
1216 mapcount += folio_entire_mapcount(page_folio(page));
1217
1218 return mapcount;
1219}
1220
1221int folio_total_mapcount(struct folio *folio);
1222
1223/**
1224 * folio_mapcount() - Calculate the number of mappings of this folio.
1225 * @folio: The folio.
1226 *
1227 * A large folio tracks both how many times the entire folio is mapped,
1228 * and how many times each individual page in the folio is mapped.
1229 * This function calculates the total number of times the folio is
1230 * mapped.
1231 *
1232 * Return: The number of times this folio is mapped.
1233 */
1234static inline int folio_mapcount(struct folio *folio)
1235{
1236 if (likely(!folio_test_large(folio)))
1237 return atomic_read(&folio->_mapcount) + 1;
1238 return folio_total_mapcount(folio);
1239}
1240
1241static inline int total_mapcount(struct page *page)
1242{
1243 if (likely(!PageCompound(page)))
1244 return atomic_read(&page->_mapcount) + 1;
1245 return folio_total_mapcount(page_folio(page));
1246}
1247
1248static inline bool folio_large_is_mapped(struct folio *folio)
1249{
1250 /*
1251 * Reading _entire_mapcount below could be omitted if hugetlb
1252 * participated in incrementing nr_pages_mapped when compound mapped.
1253 */
1254 return atomic_read(&folio->_nr_pages_mapped) > 0 ||
1255 atomic_read(&folio->_entire_mapcount) >= 0;
1256}
1257
1258/**
1259 * folio_mapped - Is this folio mapped into userspace?
1260 * @folio: The folio.
1261 *
1262 * Return: True if any page in this folio is referenced by user page tables.
1263 */
1264static inline bool folio_mapped(struct folio *folio)
1265{
1266 if (likely(!folio_test_large(folio)))
1267 return atomic_read(&folio->_mapcount) >= 0;
1268 return folio_large_is_mapped(folio);
1269}
1270
1271/*
1272 * Return true if this page is mapped into pagetables.
1273 * For compound page it returns true if any sub-page of compound page is mapped,
1274 * even if this particular sub-page is not itself mapped by any PTE or PMD.
1275 */
1276static inline bool page_mapped(struct page *page)
1277{
1278 if (likely(!PageCompound(page)))
1279 return atomic_read(&page->_mapcount) >= 0;
1280 return folio_large_is_mapped(page_folio(page));
1281}
1282
1283static inline struct page *virt_to_head_page(const void *x)
1284{
1285 struct page *page = virt_to_page(x);
1286
1287 return compound_head(page);
1288}
1289
1290static inline struct folio *virt_to_folio(const void *x)
1291{
1292 struct page *page = virt_to_page(x);
1293
1294 return page_folio(page);
1295}
1296
1297void __folio_put(struct folio *folio);
1298
1299void put_pages_list(struct list_head *pages);
1300
1301void split_page(struct page *page, unsigned int order);
1302void folio_copy(struct folio *dst, struct folio *src);
1303
1304unsigned long nr_free_buffer_pages(void);
1305
1306void destroy_large_folio(struct folio *folio);
1307
1308/* Returns the number of bytes in this potentially compound page. */
1309static inline unsigned long page_size(struct page *page)
1310{
1311 return PAGE_SIZE << compound_order(page);
1312}
1313
1314/* Returns the number of bits needed for the number of bytes in a page */
1315static inline unsigned int page_shift(struct page *page)
1316{
1317 return PAGE_SHIFT + compound_order(page);
1318}
1319
1320/**
1321 * thp_order - Order of a transparent huge page.
1322 * @page: Head page of a transparent huge page.
1323 */
1324static inline unsigned int thp_order(struct page *page)
1325{
1326 VM_BUG_ON_PGFLAGS(PageTail(page), page);
1327 return compound_order(page);
1328}
1329
1330/**
1331 * thp_size - Size of a transparent huge page.
1332 * @page: Head page of a transparent huge page.
1333 *
1334 * Return: Number of bytes in this page.
1335 */
1336static inline unsigned long thp_size(struct page *page)
1337{
1338 return PAGE_SIZE << thp_order(page);
1339}
1340
1341#ifdef CONFIG_MMU
1342/*
1343 * Do pte_mkwrite, but only if the vma says VM_WRITE. We do this when
1344 * servicing faults for write access. In the normal case, do always want
1345 * pte_mkwrite. But get_user_pages can cause write faults for mappings
1346 * that do not have writing enabled, when used by access_process_vm.
1347 */
1348static inline pte_t maybe_mkwrite(pte_t pte, struct vm_area_struct *vma)
1349{
1350 if (likely(vma->vm_flags & VM_WRITE))
1351 pte = pte_mkwrite(pte, vma);
1352 return pte;
1353}
1354
1355vm_fault_t do_set_pmd(struct vm_fault *vmf, struct page *page);
1356void set_pte_range(struct vm_fault *vmf, struct folio *folio,
1357 struct page *page, unsigned int nr, unsigned long addr);
1358
1359vm_fault_t finish_fault(struct vm_fault *vmf);
1360#endif
1361
1362/*
1363 * Multiple processes may "see" the same page. E.g. for untouched
1364 * mappings of /dev/null, all processes see the same page full of
1365 * zeroes, and text pages of executables and shared libraries have
1366 * only one copy in memory, at most, normally.
1367 *
1368 * For the non-reserved pages, page_count(page) denotes a reference count.
1369 * page_count() == 0 means the page is free. page->lru is then used for
1370 * freelist management in the buddy allocator.
1371 * page_count() > 0 means the page has been allocated.
1372 *
1373 * Pages are allocated by the slab allocator in order to provide memory
1374 * to kmalloc and kmem_cache_alloc. In this case, the management of the
1375 * page, and the fields in 'struct page' are the responsibility of mm/slab.c
1376 * unless a particular usage is carefully commented. (the responsibility of
1377 * freeing the kmalloc memory is the caller's, of course).
1378 *
1379 * A page may be used by anyone else who does a __get_free_page().
1380 * In this case, page_count still tracks the references, and should only
1381 * be used through the normal accessor functions. The top bits of page->flags
1382 * and page->virtual store page management information, but all other fields
1383 * are unused and could be used privately, carefully. The management of this
1384 * page is the responsibility of the one who allocated it, and those who have
1385 * subsequently been given references to it.
1386 *
1387 * The other pages (we may call them "pagecache pages") are completely
1388 * managed by the Linux memory manager: I/O, buffers, swapping etc.
1389 * The following discussion applies only to them.
1390 *
1391 * A pagecache page contains an opaque `private' member, which belongs to the
1392 * page's address_space. Usually, this is the address of a circular list of
1393 * the page's disk buffers. PG_private must be set to tell the VM to call
1394 * into the filesystem to release these pages.
1395 *
1396 * A page may belong to an inode's memory mapping. In this case, page->mapping
1397 * is the pointer to the inode, and page->index is the file offset of the page,
1398 * in units of PAGE_SIZE.
1399 *
1400 * If pagecache pages are not associated with an inode, they are said to be
1401 * anonymous pages. These may become associated with the swapcache, and in that
1402 * case PG_swapcache is set, and page->private is an offset into the swapcache.
1403 *
1404 * In either case (swapcache or inode backed), the pagecache itself holds one
1405 * reference to the page. Setting PG_private should also increment the
1406 * refcount. The each user mapping also has a reference to the page.
1407 *
1408 * The pagecache pages are stored in a per-mapping radix tree, which is
1409 * rooted at mapping->i_pages, and indexed by offset.
1410 * Where 2.4 and early 2.6 kernels kept dirty/clean pages in per-address_space
1411 * lists, we instead now tag pages as dirty/writeback in the radix tree.
1412 *
1413 * All pagecache pages may be subject to I/O:
1414 * - inode pages may need to be read from disk,
1415 * - inode pages which have been modified and are MAP_SHARED may need
1416 * to be written back to the inode on disk,
1417 * - anonymous pages (including MAP_PRIVATE file mappings) which have been
1418 * modified may need to be swapped out to swap space and (later) to be read
1419 * back into memory.
1420 */
1421
1422#if defined(CONFIG_ZONE_DEVICE) && defined(CONFIG_FS_DAX)
1423DECLARE_STATIC_KEY_FALSE(devmap_managed_key);
1424
1425bool __put_devmap_managed_page_refs(struct page *page, int refs);
1426static inline bool put_devmap_managed_page_refs(struct page *page, int refs)
1427{
1428 if (!static_branch_unlikely(&devmap_managed_key))
1429 return false;
1430 if (!is_zone_device_page(page))
1431 return false;
1432 return __put_devmap_managed_page_refs(page, refs);
1433}
1434#else /* CONFIG_ZONE_DEVICE && CONFIG_FS_DAX */
1435static inline bool put_devmap_managed_page_refs(struct page *page, int refs)
1436{
1437 return false;
1438}
1439#endif /* CONFIG_ZONE_DEVICE && CONFIG_FS_DAX */
1440
1441static inline bool put_devmap_managed_page(struct page *page)
1442{
1443 return put_devmap_managed_page_refs(page, 1);
1444}
1445
1446/* 127: arbitrary random number, small enough to assemble well */
1447#define folio_ref_zero_or_close_to_overflow(folio) \
1448 ((unsigned int) folio_ref_count(folio) + 127u <= 127u)
1449
1450/**
1451 * folio_get - Increment the reference count on a folio.
1452 * @folio: The folio.
1453 *
1454 * Context: May be called in any context, as long as you know that
1455 * you have a refcount on the folio. If you do not already have one,
1456 * folio_try_get() may be the right interface for you to use.
1457 */
1458static inline void folio_get(struct folio *folio)
1459{
1460 VM_BUG_ON_FOLIO(folio_ref_zero_or_close_to_overflow(folio), folio);
1461 folio_ref_inc(folio);
1462}
1463
1464static inline void get_page(struct page *page)
1465{
1466 folio_get(page_folio(page));
1467}
1468
1469static inline __must_check bool try_get_page(struct page *page)
1470{
1471 page = compound_head(page);
1472 if (WARN_ON_ONCE(page_ref_count(page) <= 0))
1473 return false;
1474 page_ref_inc(page);
1475 return true;
1476}
1477
1478/**
1479 * folio_put - Decrement the reference count on a folio.
1480 * @folio: The folio.
1481 *
1482 * If the folio's reference count reaches zero, the memory will be
1483 * released back to the page allocator and may be used by another
1484 * allocation immediately. Do not access the memory or the struct folio
1485 * after calling folio_put() unless you can be sure that it wasn't the
1486 * last reference.
1487 *
1488 * Context: May be called in process or interrupt context, but not in NMI
1489 * context. May be called while holding a spinlock.
1490 */
1491static inline void folio_put(struct folio *folio)
1492{
1493 if (folio_put_testzero(folio))
1494 __folio_put(folio);
1495}
1496
1497/**
1498 * folio_put_refs - Reduce the reference count on a folio.
1499 * @folio: The folio.
1500 * @refs: The amount to subtract from the folio's reference count.
1501 *
1502 * If the folio's reference count reaches zero, the memory will be
1503 * released back to the page allocator and may be used by another
1504 * allocation immediately. Do not access the memory or the struct folio
1505 * after calling folio_put_refs() unless you can be sure that these weren't
1506 * the last references.
1507 *
1508 * Context: May be called in process or interrupt context, but not in NMI
1509 * context. May be called while holding a spinlock.
1510 */
1511static inline void folio_put_refs(struct folio *folio, int refs)
1512{
1513 if (folio_ref_sub_and_test(folio, refs))
1514 __folio_put(folio);
1515}
1516
1517/*
1518 * union release_pages_arg - an array of pages or folios
1519 *
1520 * release_pages() releases a simple array of multiple pages, and
1521 * accepts various different forms of said page array: either
1522 * a regular old boring array of pages, an array of folios, or
1523 * an array of encoded page pointers.
1524 *
1525 * The transparent union syntax for this kind of "any of these
1526 * argument types" is all kinds of ugly, so look away.
1527 */
1528typedef union {
1529 struct page **pages;
1530 struct folio **folios;
1531 struct encoded_page **encoded_pages;
1532} release_pages_arg __attribute__ ((__transparent_union__));
1533
1534void release_pages(release_pages_arg, int nr);
1535
1536/**
1537 * folios_put - Decrement the reference count on an array of folios.
1538 * @folios: The folios.
1539 * @nr: How many folios there are.
1540 *
1541 * Like folio_put(), but for an array of folios. This is more efficient
1542 * than writing the loop yourself as it will optimise the locks which
1543 * need to be taken if the folios are freed.
1544 *
1545 * Context: May be called in process or interrupt context, but not in NMI
1546 * context. May be called while holding a spinlock.
1547 */
1548static inline void folios_put(struct folio **folios, unsigned int nr)
1549{
1550 release_pages(folios, nr);
1551}
1552
1553static inline void put_page(struct page *page)
1554{
1555 struct folio *folio = page_folio(page);
1556
1557 /*
1558 * For some devmap managed pages we need to catch refcount transition
1559 * from 2 to 1:
1560 */
1561 if (put_devmap_managed_page(&folio->page))
1562 return;
1563 folio_put(folio);
1564}
1565
1566/*
1567 * GUP_PIN_COUNTING_BIAS, and the associated functions that use it, overload
1568 * the page's refcount so that two separate items are tracked: the original page
1569 * reference count, and also a new count of how many pin_user_pages() calls were
1570 * made against the page. ("gup-pinned" is another term for the latter).
1571 *
1572 * With this scheme, pin_user_pages() becomes special: such pages are marked as
1573 * distinct from normal pages. As such, the unpin_user_page() call (and its
1574 * variants) must be used in order to release gup-pinned pages.
1575 *
1576 * Choice of value:
1577 *
1578 * By making GUP_PIN_COUNTING_BIAS a power of two, debugging of page reference
1579 * counts with respect to pin_user_pages() and unpin_user_page() becomes
1580 * simpler, due to the fact that adding an even power of two to the page
1581 * refcount has the effect of using only the upper N bits, for the code that
1582 * counts up using the bias value. This means that the lower bits are left for
1583 * the exclusive use of the original code that increments and decrements by one
1584 * (or at least, by much smaller values than the bias value).
1585 *
1586 * Of course, once the lower bits overflow into the upper bits (and this is
1587 * OK, because subtraction recovers the original values), then visual inspection
1588 * no longer suffices to directly view the separate counts. However, for normal
1589 * applications that don't have huge page reference counts, this won't be an
1590 * issue.
1591 *
1592 * Locking: the lockless algorithm described in folio_try_get_rcu()
1593 * provides safe operation for get_user_pages(), page_mkclean() and
1594 * other calls that race to set up page table entries.
1595 */
1596#define GUP_PIN_COUNTING_BIAS (1U << 10)
1597
1598void unpin_user_page(struct page *page);
1599void unpin_user_pages_dirty_lock(struct page **pages, unsigned long npages,
1600 bool make_dirty);
1601void unpin_user_page_range_dirty_lock(struct page *page, unsigned long npages,
1602 bool make_dirty);
1603void unpin_user_pages(struct page **pages, unsigned long npages);
1604
1605static inline bool is_cow_mapping(vm_flags_t flags)
1606{
1607 return (flags & (VM_SHARED | VM_MAYWRITE)) == VM_MAYWRITE;
1608}
1609
1610#ifndef CONFIG_MMU
1611static inline bool is_nommu_shared_mapping(vm_flags_t flags)
1612{
1613 /*
1614 * NOMMU shared mappings are ordinary MAP_SHARED mappings and selected
1615 * R/O MAP_PRIVATE file mappings that are an effective R/O overlay of
1616 * a file mapping. R/O MAP_PRIVATE mappings might still modify
1617 * underlying memory if ptrace is active, so this is only possible if
1618 * ptrace does not apply. Note that there is no mprotect() to upgrade
1619 * write permissions later.
1620 */
1621 return flags & (VM_MAYSHARE | VM_MAYOVERLAY);
1622}
1623#endif
1624
1625#if defined(CONFIG_SPARSEMEM) && !defined(CONFIG_SPARSEMEM_VMEMMAP)
1626#define SECTION_IN_PAGE_FLAGS
1627#endif
1628
1629/*
1630 * The identification function is mainly used by the buddy allocator for
1631 * determining if two pages could be buddies. We are not really identifying
1632 * the zone since we could be using the section number id if we do not have
1633 * node id available in page flags.
1634 * We only guarantee that it will return the same value for two combinable
1635 * pages in a zone.
1636 */
1637static inline int page_zone_id(struct page *page)
1638{
1639 return (page->flags >> ZONEID_PGSHIFT) & ZONEID_MASK;
1640}
1641
1642#ifdef NODE_NOT_IN_PAGE_FLAGS
1643extern int page_to_nid(const struct page *page);
1644#else
1645static inline int page_to_nid(const struct page *page)
1646{
1647 struct page *p = (struct page *)page;
1648
1649 return (PF_POISONED_CHECK(p)->flags >> NODES_PGSHIFT) & NODES_MASK;
1650}
1651#endif
1652
1653static inline int folio_nid(const struct folio *folio)
1654{
1655 return page_to_nid(&folio->page);
1656}
1657
1658#ifdef CONFIG_NUMA_BALANCING
1659/* page access time bits needs to hold at least 4 seconds */
1660#define PAGE_ACCESS_TIME_MIN_BITS 12
1661#if LAST_CPUPID_SHIFT < PAGE_ACCESS_TIME_MIN_BITS
1662#define PAGE_ACCESS_TIME_BUCKETS \
1663 (PAGE_ACCESS_TIME_MIN_BITS - LAST_CPUPID_SHIFT)
1664#else
1665#define PAGE_ACCESS_TIME_BUCKETS 0
1666#endif
1667
1668#define PAGE_ACCESS_TIME_MASK \
1669 (LAST_CPUPID_MASK << PAGE_ACCESS_TIME_BUCKETS)
1670
1671static inline int cpu_pid_to_cpupid(int cpu, int pid)
1672{
1673 return ((cpu & LAST__CPU_MASK) << LAST__PID_SHIFT) | (pid & LAST__PID_MASK);
1674}
1675
1676static inline int cpupid_to_pid(int cpupid)
1677{
1678 return cpupid & LAST__PID_MASK;
1679}
1680
1681static inline int cpupid_to_cpu(int cpupid)
1682{
1683 return (cpupid >> LAST__PID_SHIFT) & LAST__CPU_MASK;
1684}
1685
1686static inline int cpupid_to_nid(int cpupid)
1687{
1688 return cpu_to_node(cpupid_to_cpu(cpupid));
1689}
1690
1691static inline bool cpupid_pid_unset(int cpupid)
1692{
1693 return cpupid_to_pid(cpupid) == (-1 & LAST__PID_MASK);
1694}
1695
1696static inline bool cpupid_cpu_unset(int cpupid)
1697{
1698 return cpupid_to_cpu(cpupid) == (-1 & LAST__CPU_MASK);
1699}
1700
1701static inline bool __cpupid_match_pid(pid_t task_pid, int cpupid)
1702{
1703 return (task_pid & LAST__PID_MASK) == cpupid_to_pid(cpupid);
1704}
1705
1706#define cpupid_match_pid(task, cpupid) __cpupid_match_pid(task->pid, cpupid)
1707#ifdef LAST_CPUPID_NOT_IN_PAGE_FLAGS
1708static inline int folio_xchg_last_cpupid(struct folio *folio, int cpupid)
1709{
1710 return xchg(&folio->_last_cpupid, cpupid & LAST_CPUPID_MASK);
1711}
1712
1713static inline int folio_last_cpupid(struct folio *folio)
1714{
1715 return folio->_last_cpupid;
1716}
1717static inline void page_cpupid_reset_last(struct page *page)
1718{
1719 page->_last_cpupid = -1 & LAST_CPUPID_MASK;
1720}
1721#else
1722static inline int folio_last_cpupid(struct folio *folio)
1723{
1724 return (folio->flags >> LAST_CPUPID_PGSHIFT) & LAST_CPUPID_MASK;
1725}
1726
1727int folio_xchg_last_cpupid(struct folio *folio, int cpupid);
1728
1729static inline void page_cpupid_reset_last(struct page *page)
1730{
1731 page->flags |= LAST_CPUPID_MASK << LAST_CPUPID_PGSHIFT;
1732}
1733#endif /* LAST_CPUPID_NOT_IN_PAGE_FLAGS */
1734
1735static inline int folio_xchg_access_time(struct folio *folio, int time)
1736{
1737 int last_time;
1738
1739 last_time = folio_xchg_last_cpupid(folio,
1740 time >> PAGE_ACCESS_TIME_BUCKETS);
1741 return last_time << PAGE_ACCESS_TIME_BUCKETS;
1742}
1743
1744static inline void vma_set_access_pid_bit(struct vm_area_struct *vma)
1745{
1746 unsigned int pid_bit;
1747
1748 pid_bit = hash_32(current->pid, ilog2(BITS_PER_LONG));
1749 if (vma->numab_state && !test_bit(pid_bit, &vma->numab_state->pids_active[1])) {
1750 __set_bit(pid_bit, &vma->numab_state->pids_active[1]);
1751 }
1752}
1753#else /* !CONFIG_NUMA_BALANCING */
1754static inline int folio_xchg_last_cpupid(struct folio *folio, int cpupid)
1755{
1756 return folio_nid(folio); /* XXX */
1757}
1758
1759static inline int folio_xchg_access_time(struct folio *folio, int time)
1760{
1761 return 0;
1762}
1763
1764static inline int folio_last_cpupid(struct folio *folio)
1765{
1766 return folio_nid(folio); /* XXX */
1767}
1768
1769static inline int cpupid_to_nid(int cpupid)
1770{
1771 return -1;
1772}
1773
1774static inline int cpupid_to_pid(int cpupid)
1775{
1776 return -1;
1777}
1778
1779static inline int cpupid_to_cpu(int cpupid)
1780{
1781 return -1;
1782}
1783
1784static inline int cpu_pid_to_cpupid(int nid, int pid)
1785{
1786 return -1;
1787}
1788
1789static inline bool cpupid_pid_unset(int cpupid)
1790{
1791 return true;
1792}
1793
1794static inline void page_cpupid_reset_last(struct page *page)
1795{
1796}
1797
1798static inline bool cpupid_match_pid(struct task_struct *task, int cpupid)
1799{
1800 return false;
1801}
1802
1803static inline void vma_set_access_pid_bit(struct vm_area_struct *vma)
1804{
1805}
1806#endif /* CONFIG_NUMA_BALANCING */
1807
1808#if defined(CONFIG_KASAN_SW_TAGS) || defined(CONFIG_KASAN_HW_TAGS)
1809
1810/*
1811 * KASAN per-page tags are stored xor'ed with 0xff. This allows to avoid
1812 * setting tags for all pages to native kernel tag value 0xff, as the default
1813 * value 0x00 maps to 0xff.
1814 */
1815
1816static inline u8 page_kasan_tag(const struct page *page)
1817{
1818 u8 tag = KASAN_TAG_KERNEL;
1819
1820 if (kasan_enabled()) {
1821 tag = (page->flags >> KASAN_TAG_PGSHIFT) & KASAN_TAG_MASK;
1822 tag ^= 0xff;
1823 }
1824
1825 return tag;
1826}
1827
1828static inline void page_kasan_tag_set(struct page *page, u8 tag)
1829{
1830 unsigned long old_flags, flags;
1831
1832 if (!kasan_enabled())
1833 return;
1834
1835 tag ^= 0xff;
1836 old_flags = READ_ONCE(page->flags);
1837 do {
1838 flags = old_flags;
1839 flags &= ~(KASAN_TAG_MASK << KASAN_TAG_PGSHIFT);
1840 flags |= (tag & KASAN_TAG_MASK) << KASAN_TAG_PGSHIFT;
1841 } while (unlikely(!try_cmpxchg(&page->flags, &old_flags, flags)));
1842}
1843
1844static inline void page_kasan_tag_reset(struct page *page)
1845{
1846 if (kasan_enabled())
1847 page_kasan_tag_set(page, KASAN_TAG_KERNEL);
1848}
1849
1850#else /* CONFIG_KASAN_SW_TAGS || CONFIG_KASAN_HW_TAGS */
1851
1852static inline u8 page_kasan_tag(const struct page *page)
1853{
1854 return 0xff;
1855}
1856
1857static inline void page_kasan_tag_set(struct page *page, u8 tag) { }
1858static inline void page_kasan_tag_reset(struct page *page) { }
1859
1860#endif /* CONFIG_KASAN_SW_TAGS || CONFIG_KASAN_HW_TAGS */
1861
1862static inline struct zone *page_zone(const struct page *page)
1863{
1864 return &NODE_DATA(page_to_nid(page))->node_zones[page_zonenum(page)];
1865}
1866
1867static inline pg_data_t *page_pgdat(const struct page *page)
1868{
1869 return NODE_DATA(page_to_nid(page));
1870}
1871
1872static inline struct zone *folio_zone(const struct folio *folio)
1873{
1874 return page_zone(&folio->page);
1875}
1876
1877static inline pg_data_t *folio_pgdat(const struct folio *folio)
1878{
1879 return page_pgdat(&folio->page);
1880}
1881
1882#ifdef SECTION_IN_PAGE_FLAGS
1883static inline void set_page_section(struct page *page, unsigned long section)
1884{
1885 page->flags &= ~(SECTIONS_MASK << SECTIONS_PGSHIFT);
1886 page->flags |= (section & SECTIONS_MASK) << SECTIONS_PGSHIFT;
1887}
1888
1889static inline unsigned long page_to_section(const struct page *page)
1890{
1891 return (page->flags >> SECTIONS_PGSHIFT) & SECTIONS_MASK;
1892}
1893#endif
1894
1895/**
1896 * folio_pfn - Return the Page Frame Number of a folio.
1897 * @folio: The folio.
1898 *
1899 * A folio may contain multiple pages. The pages have consecutive
1900 * Page Frame Numbers.
1901 *
1902 * Return: The Page Frame Number of the first page in the folio.
1903 */
1904static inline unsigned long folio_pfn(struct folio *folio)
1905{
1906 return page_to_pfn(&folio->page);
1907}
1908
1909static inline struct folio *pfn_folio(unsigned long pfn)
1910{
1911 return page_folio(pfn_to_page(pfn));
1912}
1913
1914/**
1915 * folio_maybe_dma_pinned - Report if a folio may be pinned for DMA.
1916 * @folio: The folio.
1917 *
1918 * This function checks if a folio has been pinned via a call to
1919 * a function in the pin_user_pages() family.
1920 *
1921 * For small folios, the return value is partially fuzzy: false is not fuzzy,
1922 * because it means "definitely not pinned for DMA", but true means "probably
1923 * pinned for DMA, but possibly a false positive due to having at least
1924 * GUP_PIN_COUNTING_BIAS worth of normal folio references".
1925 *
1926 * False positives are OK, because: a) it's unlikely for a folio to
1927 * get that many refcounts, and b) all the callers of this routine are
1928 * expected to be able to deal gracefully with a false positive.
1929 *
1930 * For large folios, the result will be exactly correct. That's because
1931 * we have more tracking data available: the _pincount field is used
1932 * instead of the GUP_PIN_COUNTING_BIAS scheme.
1933 *
1934 * For more information, please see Documentation/core-api/pin_user_pages.rst.
1935 *
1936 * Return: True, if it is likely that the page has been "dma-pinned".
1937 * False, if the page is definitely not dma-pinned.
1938 */
1939static inline bool folio_maybe_dma_pinned(struct folio *folio)
1940{
1941 if (folio_test_large(folio))
1942 return atomic_read(&folio->_pincount) > 0;
1943
1944 /*
1945 * folio_ref_count() is signed. If that refcount overflows, then
1946 * folio_ref_count() returns a negative value, and callers will avoid
1947 * further incrementing the refcount.
1948 *
1949 * Here, for that overflow case, use the sign bit to count a little
1950 * bit higher via unsigned math, and thus still get an accurate result.
1951 */
1952 return ((unsigned int)folio_ref_count(folio)) >=
1953 GUP_PIN_COUNTING_BIAS;
1954}
1955
1956static inline bool page_maybe_dma_pinned(struct page *page)
1957{
1958 return folio_maybe_dma_pinned(page_folio(page));
1959}
1960
1961/*
1962 * This should most likely only be called during fork() to see whether we
1963 * should break the cow immediately for an anon page on the src mm.
1964 *
1965 * The caller has to hold the PT lock and the vma->vm_mm->->write_protect_seq.
1966 */
1967static inline bool folio_needs_cow_for_dma(struct vm_area_struct *vma,
1968 struct folio *folio)
1969{
1970 VM_BUG_ON(!(raw_read_seqcount(&vma->vm_mm->write_protect_seq) & 1));
1971
1972 if (!test_bit(MMF_HAS_PINNED, &vma->vm_mm->flags))
1973 return false;
1974
1975 return folio_maybe_dma_pinned(folio);
1976}
1977
1978/**
1979 * is_zero_page - Query if a page is a zero page
1980 * @page: The page to query
1981 *
1982 * This returns true if @page is one of the permanent zero pages.
1983 */
1984static inline bool is_zero_page(const struct page *page)
1985{
1986 return is_zero_pfn(page_to_pfn(page));
1987}
1988
1989/**
1990 * is_zero_folio - Query if a folio is a zero page
1991 * @folio: The folio to query
1992 *
1993 * This returns true if @folio is one of the permanent zero pages.
1994 */
1995static inline bool is_zero_folio(const struct folio *folio)
1996{
1997 return is_zero_page(&folio->page);
1998}
1999
2000/* MIGRATE_CMA and ZONE_MOVABLE do not allow pin folios */
2001#ifdef CONFIG_MIGRATION
2002static inline bool folio_is_longterm_pinnable(struct folio *folio)
2003{
2004#ifdef CONFIG_CMA
2005 int mt = folio_migratetype(folio);
2006
2007 if (mt == MIGRATE_CMA || mt == MIGRATE_ISOLATE)
2008 return false;
2009#endif
2010 /* The zero page can be "pinned" but gets special handling. */
2011 if (is_zero_folio(folio))
2012 return true;
2013
2014 /* Coherent device memory must always allow eviction. */
2015 if (folio_is_device_coherent(folio))
2016 return false;
2017
2018 /* Otherwise, non-movable zone folios can be pinned. */
2019 return !folio_is_zone_movable(folio);
2020
2021}
2022#else
2023static inline bool folio_is_longterm_pinnable(struct folio *folio)
2024{
2025 return true;
2026}
2027#endif
2028
2029static inline void set_page_zone(struct page *page, enum zone_type zone)
2030{
2031 page->flags &= ~(ZONES_MASK << ZONES_PGSHIFT);
2032 page->flags |= (zone & ZONES_MASK) << ZONES_PGSHIFT;
2033}
2034
2035static inline void set_page_node(struct page *page, unsigned long node)
2036{
2037 page->flags &= ~(NODES_MASK << NODES_PGSHIFT);
2038 page->flags |= (node & NODES_MASK) << NODES_PGSHIFT;
2039}
2040
2041static inline void set_page_links(struct page *page, enum zone_type zone,
2042 unsigned long node, unsigned long pfn)
2043{
2044 set_page_zone(page, zone);
2045 set_page_node(page, node);
2046#ifdef SECTION_IN_PAGE_FLAGS
2047 set_page_section(page, pfn_to_section_nr(pfn));
2048#endif
2049}
2050
2051/**
2052 * folio_nr_pages - The number of pages in the folio.
2053 * @folio: The folio.
2054 *
2055 * Return: A positive power of two.
2056 */
2057static inline long folio_nr_pages(struct folio *folio)
2058{
2059 if (!folio_test_large(folio))
2060 return 1;
2061#ifdef CONFIG_64BIT
2062 return folio->_folio_nr_pages;
2063#else
2064 return 1L << (folio->_flags_1 & 0xff);
2065#endif
2066}
2067
2068/*
2069 * compound_nr() returns the number of pages in this potentially compound
2070 * page. compound_nr() can be called on a tail page, and is defined to
2071 * return 1 in that case.
2072 */
2073static inline unsigned long compound_nr(struct page *page)
2074{
2075 struct folio *folio = (struct folio *)page;
2076
2077 if (!test_bit(PG_head, &folio->flags))
2078 return 1;
2079#ifdef CONFIG_64BIT
2080 return folio->_folio_nr_pages;
2081#else
2082 return 1L << (folio->_flags_1 & 0xff);
2083#endif
2084}
2085
2086/**
2087 * thp_nr_pages - The number of regular pages in this huge page.
2088 * @page: The head page of a huge page.
2089 */
2090static inline int thp_nr_pages(struct page *page)
2091{
2092 return folio_nr_pages((struct folio *)page);
2093}
2094
2095/**
2096 * folio_next - Move to the next physical folio.
2097 * @folio: The folio we're currently operating on.
2098 *
2099 * If you have physically contiguous memory which may span more than
2100 * one folio (eg a &struct bio_vec), use this function to move from one
2101 * folio to the next. Do not use it if the memory is only virtually
2102 * contiguous as the folios are almost certainly not adjacent to each
2103 * other. This is the folio equivalent to writing ``page++``.
2104 *
2105 * Context: We assume that the folios are refcounted and/or locked at a
2106 * higher level and do not adjust the reference counts.
2107 * Return: The next struct folio.
2108 */
2109static inline struct folio *folio_next(struct folio *folio)
2110{
2111 return (struct folio *)folio_page(folio, folio_nr_pages(folio));
2112}
2113
2114/**
2115 * folio_shift - The size of the memory described by this folio.
2116 * @folio: The folio.
2117 *
2118 * A folio represents a number of bytes which is a power-of-two in size.
2119 * This function tells you which power-of-two the folio is. See also
2120 * folio_size() and folio_order().
2121 *
2122 * Context: The caller should have a reference on the folio to prevent
2123 * it from being split. It is not necessary for the folio to be locked.
2124 * Return: The base-2 logarithm of the size of this folio.
2125 */
2126static inline unsigned int folio_shift(struct folio *folio)
2127{
2128 return PAGE_SHIFT + folio_order(folio);
2129}
2130
2131/**
2132 * folio_size - The number of bytes in a folio.
2133 * @folio: The folio.
2134 *
2135 * Context: The caller should have a reference on the folio to prevent
2136 * it from being split. It is not necessary for the folio to be locked.
2137 * Return: The number of bytes in this folio.
2138 */
2139static inline size_t folio_size(struct folio *folio)
2140{
2141 return PAGE_SIZE << folio_order(folio);
2142}
2143
2144/**
2145 * folio_estimated_sharers - Estimate the number of sharers of a folio.
2146 * @folio: The folio.
2147 *
2148 * folio_estimated_sharers() aims to serve as a function to efficiently
2149 * estimate the number of processes sharing a folio. This is done by
2150 * looking at the precise mapcount of the first subpage in the folio, and
2151 * assuming the other subpages are the same. This may not be true for large
2152 * folios. If you want exact mapcounts for exact calculations, look at
2153 * page_mapcount() or folio_total_mapcount().
2154 *
2155 * Return: The estimated number of processes sharing a folio.
2156 */
2157static inline int folio_estimated_sharers(struct folio *folio)
2158{
2159 return page_mapcount(folio_page(folio, 0));
2160}
2161
2162#ifndef HAVE_ARCH_MAKE_PAGE_ACCESSIBLE
2163static inline int arch_make_page_accessible(struct page *page)
2164{
2165 return 0;
2166}
2167#endif
2168
2169#ifndef HAVE_ARCH_MAKE_FOLIO_ACCESSIBLE
2170static inline int arch_make_folio_accessible(struct folio *folio)
2171{
2172 int ret;
2173 long i, nr = folio_nr_pages(folio);
2174
2175 for (i = 0; i < nr; i++) {
2176 ret = arch_make_page_accessible(folio_page(folio, i));
2177 if (ret)
2178 break;
2179 }
2180
2181 return ret;
2182}
2183#endif
2184
2185/*
2186 * Some inline functions in vmstat.h depend on page_zone()
2187 */
2188#include <linux/vmstat.h>
2189
2190static __always_inline void *lowmem_page_address(const struct page *page)
2191{
2192 return page_to_virt(page);
2193}
2194
2195#if defined(CONFIG_HIGHMEM) && !defined(WANT_PAGE_VIRTUAL)
2196#define HASHED_PAGE_VIRTUAL
2197#endif
2198
2199#if defined(WANT_PAGE_VIRTUAL)
2200static inline void *page_address(const struct page *page)
2201{
2202 return page->virtual;
2203}
2204static inline void set_page_address(struct page *page, void *address)
2205{
2206 page->virtual = address;
2207}
2208#define page_address_init() do { } while(0)
2209#endif
2210
2211#if defined(HASHED_PAGE_VIRTUAL)
2212void *page_address(const struct page *page);
2213void set_page_address(struct page *page, void *virtual);
2214void page_address_init(void);
2215#endif
2216
2217#if !defined(HASHED_PAGE_VIRTUAL) && !defined(WANT_PAGE_VIRTUAL)
2218#define page_address(page) lowmem_page_address(page)
2219#define set_page_address(page, address) do { } while(0)
2220#define page_address_init() do { } while(0)
2221#endif
2222
2223static inline void *folio_address(const struct folio *folio)
2224{
2225 return page_address(&folio->page);
2226}
2227
2228extern pgoff_t __page_file_index(struct page *page);
2229
2230/*
2231 * Return the pagecache index of the passed page. Regular pagecache pages
2232 * use ->index whereas swapcache pages use swp_offset(->private)
2233 */
2234static inline pgoff_t page_index(struct page *page)
2235{
2236 if (unlikely(PageSwapCache(page)))
2237 return __page_file_index(page);
2238 return page->index;
2239}
2240
2241/*
2242 * Return true only if the page has been allocated with
2243 * ALLOC_NO_WATERMARKS and the low watermark was not
2244 * met implying that the system is under some pressure.
2245 */
2246static inline bool page_is_pfmemalloc(const struct page *page)
2247{
2248 /*
2249 * lru.next has bit 1 set if the page is allocated from the
2250 * pfmemalloc reserves. Callers may simply overwrite it if
2251 * they do not need to preserve that information.
2252 */
2253 return (uintptr_t)page->lru.next & BIT(1);
2254}
2255
2256/*
2257 * Return true only if the folio has been allocated with
2258 * ALLOC_NO_WATERMARKS and the low watermark was not
2259 * met implying that the system is under some pressure.
2260 */
2261static inline bool folio_is_pfmemalloc(const struct folio *folio)
2262{
2263 /*
2264 * lru.next has bit 1 set if the page is allocated from the
2265 * pfmemalloc reserves. Callers may simply overwrite it if
2266 * they do not need to preserve that information.
2267 */
2268 return (uintptr_t)folio->lru.next & BIT(1);
2269}
2270
2271/*
2272 * Only to be called by the page allocator on a freshly allocated
2273 * page.
2274 */
2275static inline void set_page_pfmemalloc(struct page *page)
2276{
2277 page->lru.next = (void *)BIT(1);
2278}
2279
2280static inline void clear_page_pfmemalloc(struct page *page)
2281{
2282 page->lru.next = NULL;
2283}
2284
2285/*
2286 * Can be called by the pagefault handler when it gets a VM_FAULT_OOM.
2287 */
2288extern void pagefault_out_of_memory(void);
2289
2290#define offset_in_page(p) ((unsigned long)(p) & ~PAGE_MASK)
2291#define offset_in_thp(page, p) ((unsigned long)(p) & (thp_size(page) - 1))
2292#define offset_in_folio(folio, p) ((unsigned long)(p) & (folio_size(folio) - 1))
2293
2294/*
2295 * Parameter block passed down to zap_pte_range in exceptional cases.
2296 */
2297struct zap_details {
2298 struct folio *single_folio; /* Locked folio to be unmapped */
2299 bool even_cows; /* Zap COWed private pages too? */
2300 zap_flags_t zap_flags; /* Extra flags for zapping */
2301};
2302
2303/*
2304 * Whether to drop the pte markers, for example, the uffd-wp information for
2305 * file-backed memory. This should only be specified when we will completely
2306 * drop the page in the mm, either by truncation or unmapping of the vma. By
2307 * default, the flag is not set.
2308 */
2309#define ZAP_FLAG_DROP_MARKER ((__force zap_flags_t) BIT(0))
2310/* Set in unmap_vmas() to indicate a final unmap call. Only used by hugetlb */
2311#define ZAP_FLAG_UNMAP ((__force zap_flags_t) BIT(1))
2312
2313#ifdef CONFIG_SCHED_MM_CID
2314void sched_mm_cid_before_execve(struct task_struct *t);
2315void sched_mm_cid_after_execve(struct task_struct *t);
2316void sched_mm_cid_fork(struct task_struct *t);
2317void sched_mm_cid_exit_signals(struct task_struct *t);
2318static inline int task_mm_cid(struct task_struct *t)
2319{
2320 return t->mm_cid;
2321}
2322#else
2323static inline void sched_mm_cid_before_execve(struct task_struct *t) { }
2324static inline void sched_mm_cid_after_execve(struct task_struct *t) { }
2325static inline void sched_mm_cid_fork(struct task_struct *t) { }
2326static inline void sched_mm_cid_exit_signals(struct task_struct *t) { }
2327static inline int task_mm_cid(struct task_struct *t)
2328{
2329 /*
2330 * Use the processor id as a fall-back when the mm cid feature is
2331 * disabled. This provides functional per-cpu data structure accesses
2332 * in user-space, althrough it won't provide the memory usage benefits.
2333 */
2334 return raw_smp_processor_id();
2335}
2336#endif
2337
2338#ifdef CONFIG_MMU
2339extern bool can_do_mlock(void);
2340#else
2341static inline bool can_do_mlock(void) { return false; }
2342#endif
2343extern int user_shm_lock(size_t, struct ucounts *);
2344extern void user_shm_unlock(size_t, struct ucounts *);
2345
2346struct folio *vm_normal_folio(struct vm_area_struct *vma, unsigned long addr,
2347 pte_t pte);
2348struct page *vm_normal_page(struct vm_area_struct *vma, unsigned long addr,
2349 pte_t pte);
2350struct folio *vm_normal_folio_pmd(struct vm_area_struct *vma,
2351 unsigned long addr, pmd_t pmd);
2352struct page *vm_normal_page_pmd(struct vm_area_struct *vma, unsigned long addr,
2353 pmd_t pmd);
2354
2355void zap_vma_ptes(struct vm_area_struct *vma, unsigned long address,
2356 unsigned long size);
2357void zap_page_range_single(struct vm_area_struct *vma, unsigned long address,
2358 unsigned long size, struct zap_details *details);
2359static inline void zap_vma_pages(struct vm_area_struct *vma)
2360{
2361 zap_page_range_single(vma, vma->vm_start,
2362 vma->vm_end - vma->vm_start, NULL);
2363}
2364void unmap_vmas(struct mmu_gather *tlb, struct ma_state *mas,
2365 struct vm_area_struct *start_vma, unsigned long start,
2366 unsigned long end, unsigned long tree_end, bool mm_wr_locked);
2367
2368struct mmu_notifier_range;
2369
2370void free_pgd_range(struct mmu_gather *tlb, unsigned long addr,
2371 unsigned long end, unsigned long floor, unsigned long ceiling);
2372int
2373copy_page_range(struct vm_area_struct *dst_vma, struct vm_area_struct *src_vma);
2374int follow_pte(struct mm_struct *mm, unsigned long address,
2375 pte_t **ptepp, spinlock_t **ptlp);
2376int follow_pfn(struct vm_area_struct *vma, unsigned long address,
2377 unsigned long *pfn);
2378int follow_phys(struct vm_area_struct *vma, unsigned long address,
2379 unsigned int flags, unsigned long *prot, resource_size_t *phys);
2380int generic_access_phys(struct vm_area_struct *vma, unsigned long addr,
2381 void *buf, int len, int write);
2382
2383extern void truncate_pagecache(struct inode *inode, loff_t new);
2384extern void truncate_setsize(struct inode *inode, loff_t newsize);
2385void pagecache_isize_extended(struct inode *inode, loff_t from, loff_t to);
2386void truncate_pagecache_range(struct inode *inode, loff_t offset, loff_t end);
2387int generic_error_remove_folio(struct address_space *mapping,
2388 struct folio *folio);
2389
2390struct vm_area_struct *lock_mm_and_find_vma(struct mm_struct *mm,
2391 unsigned long address, struct pt_regs *regs);
2392
2393#ifdef CONFIG_MMU
2394extern vm_fault_t handle_mm_fault(struct vm_area_struct *vma,
2395 unsigned long address, unsigned int flags,
2396 struct pt_regs *regs);
2397extern int fixup_user_fault(struct mm_struct *mm,
2398 unsigned long address, unsigned int fault_flags,
2399 bool *unlocked);
2400void unmap_mapping_pages(struct address_space *mapping,
2401 pgoff_t start, pgoff_t nr, bool even_cows);
2402void unmap_mapping_range(struct address_space *mapping,
2403 loff_t const holebegin, loff_t const holelen, int even_cows);
2404#else
2405static inline vm_fault_t handle_mm_fault(struct vm_area_struct *vma,
2406 unsigned long address, unsigned int flags,
2407 struct pt_regs *regs)
2408{
2409 /* should never happen if there's no MMU */
2410 BUG();
2411 return VM_FAULT_SIGBUS;
2412}
2413static inline int fixup_user_fault(struct mm_struct *mm, unsigned long address,
2414 unsigned int fault_flags, bool *unlocked)
2415{
2416 /* should never happen if there's no MMU */
2417 BUG();
2418 return -EFAULT;
2419}
2420static inline void unmap_mapping_pages(struct address_space *mapping,
2421 pgoff_t start, pgoff_t nr, bool even_cows) { }
2422static inline void unmap_mapping_range(struct address_space *mapping,
2423 loff_t const holebegin, loff_t const holelen, int even_cows) { }
2424#endif
2425
2426static inline void unmap_shared_mapping_range(struct address_space *mapping,
2427 loff_t const holebegin, loff_t const holelen)
2428{
2429 unmap_mapping_range(mapping, holebegin, holelen, 0);
2430}
2431
2432static inline struct vm_area_struct *vma_lookup(struct mm_struct *mm,
2433 unsigned long addr);
2434
2435extern int access_process_vm(struct task_struct *tsk, unsigned long addr,
2436 void *buf, int len, unsigned int gup_flags);
2437extern int access_remote_vm(struct mm_struct *mm, unsigned long addr,
2438 void *buf, int len, unsigned int gup_flags);
2439
2440long get_user_pages_remote(struct mm_struct *mm,
2441 unsigned long start, unsigned long nr_pages,
2442 unsigned int gup_flags, struct page **pages,
2443 int *locked);
2444long pin_user_pages_remote(struct mm_struct *mm,
2445 unsigned long start, unsigned long nr_pages,
2446 unsigned int gup_flags, struct page **pages,
2447 int *locked);
2448
2449/*
2450 * Retrieves a single page alongside its VMA. Does not support FOLL_NOWAIT.
2451 */
2452static inline struct page *get_user_page_vma_remote(struct mm_struct *mm,
2453 unsigned long addr,
2454 int gup_flags,
2455 struct vm_area_struct **vmap)
2456{
2457 struct page *page;
2458 struct vm_area_struct *vma;
2459 int got;
2460
2461 if (WARN_ON_ONCE(unlikely(gup_flags & FOLL_NOWAIT)))
2462 return ERR_PTR(-EINVAL);
2463
2464 got = get_user_pages_remote(mm, addr, 1, gup_flags, &page, NULL);
2465
2466 if (got < 0)
2467 return ERR_PTR(got);
2468
2469 vma = vma_lookup(mm, addr);
2470 if (WARN_ON_ONCE(!vma)) {
2471 put_page(page);
2472 return ERR_PTR(-EINVAL);
2473 }
2474
2475 *vmap = vma;
2476 return page;
2477}
2478
2479long get_user_pages(unsigned long start, unsigned long nr_pages,
2480 unsigned int gup_flags, struct page **pages);
2481long pin_user_pages(unsigned long start, unsigned long nr_pages,
2482 unsigned int gup_flags, struct page **pages);
2483long get_user_pages_unlocked(unsigned long start, unsigned long nr_pages,
2484 struct page **pages, unsigned int gup_flags);
2485long pin_user_pages_unlocked(unsigned long start, unsigned long nr_pages,
2486 struct page **pages, unsigned int gup_flags);
2487
2488int get_user_pages_fast(unsigned long start, int nr_pages,
2489 unsigned int gup_flags, struct page **pages);
2490int pin_user_pages_fast(unsigned long start, int nr_pages,
2491 unsigned int gup_flags, struct page **pages);
2492void folio_add_pin(struct folio *folio);
2493
2494int account_locked_vm(struct mm_struct *mm, unsigned long pages, bool inc);
2495int __account_locked_vm(struct mm_struct *mm, unsigned long pages, bool inc,
2496 struct task_struct *task, bool bypass_rlim);
2497
2498struct kvec;
2499struct page *get_dump_page(unsigned long addr);
2500
2501bool folio_mark_dirty(struct folio *folio);
2502bool set_page_dirty(struct page *page);
2503int set_page_dirty_lock(struct page *page);
2504
2505int get_cmdline(struct task_struct *task, char *buffer, int buflen);
2506
2507extern unsigned long move_page_tables(struct vm_area_struct *vma,
2508 unsigned long old_addr, struct vm_area_struct *new_vma,
2509 unsigned long new_addr, unsigned long len,
2510 bool need_rmap_locks, bool for_stack);
2511
2512/*
2513 * Flags used by change_protection(). For now we make it a bitmap so
2514 * that we can pass in multiple flags just like parameters. However
2515 * for now all the callers are only use one of the flags at the same
2516 * time.
2517 */
2518/*
2519 * Whether we should manually check if we can map individual PTEs writable,
2520 * because something (e.g., COW, uffd-wp) blocks that from happening for all
2521 * PTEs automatically in a writable mapping.
2522 */
2523#define MM_CP_TRY_CHANGE_WRITABLE (1UL << 0)
2524/* Whether this protection change is for NUMA hints */
2525#define MM_CP_PROT_NUMA (1UL << 1)
2526/* Whether this change is for write protecting */
2527#define MM_CP_UFFD_WP (1UL << 2) /* do wp */
2528#define MM_CP_UFFD_WP_RESOLVE (1UL << 3) /* Resolve wp */
2529#define MM_CP_UFFD_WP_ALL (MM_CP_UFFD_WP | \
2530 MM_CP_UFFD_WP_RESOLVE)
2531
2532bool vma_needs_dirty_tracking(struct vm_area_struct *vma);
2533int vma_wants_writenotify(struct vm_area_struct *vma, pgprot_t vm_page_prot);
2534static inline bool vma_wants_manual_pte_write_upgrade(struct vm_area_struct *vma)
2535{
2536 /*
2537 * We want to check manually if we can change individual PTEs writable
2538 * if we can't do that automatically for all PTEs in a mapping. For
2539 * private mappings, that's always the case when we have write
2540 * permissions as we properly have to handle COW.
2541 */
2542 if (vma->vm_flags & VM_SHARED)
2543 return vma_wants_writenotify(vma, vma->vm_page_prot);
2544 return !!(vma->vm_flags & VM_WRITE);
2545
2546}
2547bool can_change_pte_writable(struct vm_area_struct *vma, unsigned long addr,
2548 pte_t pte);
2549extern long change_protection(struct mmu_gather *tlb,
2550 struct vm_area_struct *vma, unsigned long start,
2551 unsigned long end, unsigned long cp_flags);
2552extern int mprotect_fixup(struct vma_iterator *vmi, struct mmu_gather *tlb,
2553 struct vm_area_struct *vma, struct vm_area_struct **pprev,
2554 unsigned long start, unsigned long end, unsigned long newflags);
2555
2556/*
2557 * doesn't attempt to fault and will return short.
2558 */
2559int get_user_pages_fast_only(unsigned long start, int nr_pages,
2560 unsigned int gup_flags, struct page **pages);
2561
2562static inline bool get_user_page_fast_only(unsigned long addr,
2563 unsigned int gup_flags, struct page **pagep)
2564{
2565 return get_user_pages_fast_only(addr, 1, gup_flags, pagep) == 1;
2566}
2567/*
2568 * per-process(per-mm_struct) statistics.
2569 */
2570static inline unsigned long get_mm_counter(struct mm_struct *mm, int member)
2571{
2572 return percpu_counter_read_positive(&mm->rss_stat[member]);
2573}
2574
2575void mm_trace_rss_stat(struct mm_struct *mm, int member);
2576
2577static inline void add_mm_counter(struct mm_struct *mm, int member, long value)
2578{
2579 percpu_counter_add(&mm->rss_stat[member], value);
2580
2581 mm_trace_rss_stat(mm, member);
2582}
2583
2584static inline void inc_mm_counter(struct mm_struct *mm, int member)
2585{
2586 percpu_counter_inc(&mm->rss_stat[member]);
2587
2588 mm_trace_rss_stat(mm, member);
2589}
2590
2591static inline void dec_mm_counter(struct mm_struct *mm, int member)
2592{
2593 percpu_counter_dec(&mm->rss_stat[member]);
2594
2595 mm_trace_rss_stat(mm, member);
2596}
2597
2598/* Optimized variant when page is already known not to be PageAnon */
2599static inline int mm_counter_file(struct page *page)
2600{
2601 if (PageSwapBacked(page))
2602 return MM_SHMEMPAGES;
2603 return MM_FILEPAGES;
2604}
2605
2606static inline int mm_counter(struct page *page)
2607{
2608 if (PageAnon(page))
2609 return MM_ANONPAGES;
2610 return mm_counter_file(page);
2611}
2612
2613static inline unsigned long get_mm_rss(struct mm_struct *mm)
2614{
2615 return get_mm_counter(mm, MM_FILEPAGES) +
2616 get_mm_counter(mm, MM_ANONPAGES) +
2617 get_mm_counter(mm, MM_SHMEMPAGES);
2618}
2619
2620static inline unsigned long get_mm_hiwater_rss(struct mm_struct *mm)
2621{
2622 return max(mm->hiwater_rss, get_mm_rss(mm));
2623}
2624
2625static inline unsigned long get_mm_hiwater_vm(struct mm_struct *mm)
2626{
2627 return max(mm->hiwater_vm, mm->total_vm);
2628}
2629
2630static inline void update_hiwater_rss(struct mm_struct *mm)
2631{
2632 unsigned long _rss = get_mm_rss(mm);
2633
2634 if ((mm)->hiwater_rss < _rss)
2635 (mm)->hiwater_rss = _rss;
2636}
2637
2638static inline void update_hiwater_vm(struct mm_struct *mm)
2639{
2640 if (mm->hiwater_vm < mm->total_vm)
2641 mm->hiwater_vm = mm->total_vm;
2642}
2643
2644static inline void reset_mm_hiwater_rss(struct mm_struct *mm)
2645{
2646 mm->hiwater_rss = get_mm_rss(mm);
2647}
2648
2649static inline void setmax_mm_hiwater_rss(unsigned long *maxrss,
2650 struct mm_struct *mm)
2651{
2652 unsigned long hiwater_rss = get_mm_hiwater_rss(mm);
2653
2654 if (*maxrss < hiwater_rss)
2655 *maxrss = hiwater_rss;
2656}
2657
2658#ifndef CONFIG_ARCH_HAS_PTE_SPECIAL
2659static inline int pte_special(pte_t pte)
2660{
2661 return 0;
2662}
2663
2664static inline pte_t pte_mkspecial(pte_t pte)
2665{
2666 return pte;
2667}
2668#endif
2669
2670#ifndef CONFIG_ARCH_HAS_PTE_DEVMAP
2671static inline int pte_devmap(pte_t pte)
2672{
2673 return 0;
2674}
2675#endif
2676
2677extern pte_t *__get_locked_pte(struct mm_struct *mm, unsigned long addr,
2678 spinlock_t **ptl);
2679static inline pte_t *get_locked_pte(struct mm_struct *mm, unsigned long addr,
2680 spinlock_t **ptl)
2681{
2682 pte_t *ptep;
2683 __cond_lock(*ptl, ptep = __get_locked_pte(mm, addr, ptl));
2684 return ptep;
2685}
2686
2687#ifdef __PAGETABLE_P4D_FOLDED
2688static inline int __p4d_alloc(struct mm_struct *mm, pgd_t *pgd,
2689 unsigned long address)
2690{
2691 return 0;
2692}
2693#else
2694int __p4d_alloc(struct mm_struct *mm, pgd_t *pgd, unsigned long address);
2695#endif
2696
2697#if defined(__PAGETABLE_PUD_FOLDED) || !defined(CONFIG_MMU)
2698static inline int __pud_alloc(struct mm_struct *mm, p4d_t *p4d,
2699 unsigned long address)
2700{
2701 return 0;
2702}
2703static inline void mm_inc_nr_puds(struct mm_struct *mm) {}
2704static inline void mm_dec_nr_puds(struct mm_struct *mm) {}
2705
2706#else
2707int __pud_alloc(struct mm_struct *mm, p4d_t *p4d, unsigned long address);
2708
2709static inline void mm_inc_nr_puds(struct mm_struct *mm)
2710{
2711 if (mm_pud_folded(mm))
2712 return;
2713 atomic_long_add(PTRS_PER_PUD * sizeof(pud_t), &mm->pgtables_bytes);
2714}
2715
2716static inline void mm_dec_nr_puds(struct mm_struct *mm)
2717{
2718 if (mm_pud_folded(mm))
2719 return;
2720 atomic_long_sub(PTRS_PER_PUD * sizeof(pud_t), &mm->pgtables_bytes);
2721}
2722#endif
2723
2724#if defined(__PAGETABLE_PMD_FOLDED) || !defined(CONFIG_MMU)
2725static inline int __pmd_alloc(struct mm_struct *mm, pud_t *pud,
2726 unsigned long address)
2727{
2728 return 0;
2729}
2730
2731static inline void mm_inc_nr_pmds(struct mm_struct *mm) {}
2732static inline void mm_dec_nr_pmds(struct mm_struct *mm) {}
2733
2734#else
2735int __pmd_alloc(struct mm_struct *mm, pud_t *pud, unsigned long address);
2736
2737static inline void mm_inc_nr_pmds(struct mm_struct *mm)
2738{
2739 if (mm_pmd_folded(mm))
2740 return;
2741 atomic_long_add(PTRS_PER_PMD * sizeof(pmd_t), &mm->pgtables_bytes);
2742}
2743
2744static inline void mm_dec_nr_pmds(struct mm_struct *mm)
2745{
2746 if (mm_pmd_folded(mm))
2747 return;
2748 atomic_long_sub(PTRS_PER_PMD * sizeof(pmd_t), &mm->pgtables_bytes);
2749}
2750#endif
2751
2752#ifdef CONFIG_MMU
2753static inline void mm_pgtables_bytes_init(struct mm_struct *mm)
2754{
2755 atomic_long_set(&mm->pgtables_bytes, 0);
2756}
2757
2758static inline unsigned long mm_pgtables_bytes(const struct mm_struct *mm)
2759{
2760 return atomic_long_read(&mm->pgtables_bytes);
2761}
2762
2763static inline void mm_inc_nr_ptes(struct mm_struct *mm)
2764{
2765 atomic_long_add(PTRS_PER_PTE * sizeof(pte_t), &mm->pgtables_bytes);
2766}
2767
2768static inline void mm_dec_nr_ptes(struct mm_struct *mm)
2769{
2770 atomic_long_sub(PTRS_PER_PTE * sizeof(pte_t), &mm->pgtables_bytes);
2771}
2772#else
2773
2774static inline void mm_pgtables_bytes_init(struct mm_struct *mm) {}
2775static inline unsigned long mm_pgtables_bytes(const struct mm_struct *mm)
2776{
2777 return 0;
2778}
2779
2780static inline void mm_inc_nr_ptes(struct mm_struct *mm) {}
2781static inline void mm_dec_nr_ptes(struct mm_struct *mm) {}
2782#endif
2783
2784int __pte_alloc(struct mm_struct *mm, pmd_t *pmd);
2785int __pte_alloc_kernel(pmd_t *pmd);
2786
2787#if defined(CONFIG_MMU)
2788
2789static inline p4d_t *p4d_alloc(struct mm_struct *mm, pgd_t *pgd,
2790 unsigned long address)
2791{
2792 return (unlikely(pgd_none(*pgd)) && __p4d_alloc(mm, pgd, address)) ?
2793 NULL : p4d_offset(pgd, address);
2794}
2795
2796static inline pud_t *pud_alloc(struct mm_struct *mm, p4d_t *p4d,
2797 unsigned long address)
2798{
2799 return (unlikely(p4d_none(*p4d)) && __pud_alloc(mm, p4d, address)) ?
2800 NULL : pud_offset(p4d, address);
2801}
2802
2803static inline pmd_t *pmd_alloc(struct mm_struct *mm, pud_t *pud, unsigned long address)
2804{
2805 return (unlikely(pud_none(*pud)) && __pmd_alloc(mm, pud, address))?
2806 NULL: pmd_offset(pud, address);
2807}
2808#endif /* CONFIG_MMU */
2809
2810static inline struct ptdesc *virt_to_ptdesc(const void *x)
2811{
2812 return page_ptdesc(virt_to_page(x));
2813}
2814
2815static inline void *ptdesc_to_virt(const struct ptdesc *pt)
2816{
2817 return page_to_virt(ptdesc_page(pt));
2818}
2819
2820static inline void *ptdesc_address(const struct ptdesc *pt)
2821{
2822 return folio_address(ptdesc_folio(pt));
2823}
2824
2825static inline bool pagetable_is_reserved(struct ptdesc *pt)
2826{
2827 return folio_test_reserved(ptdesc_folio(pt));
2828}
2829
2830/**
2831 * pagetable_alloc - Allocate pagetables
2832 * @gfp: GFP flags
2833 * @order: desired pagetable order
2834 *
2835 * pagetable_alloc allocates memory for page tables as well as a page table
2836 * descriptor to describe that memory.
2837 *
2838 * Return: The ptdesc describing the allocated page tables.
2839 */
2840static inline struct ptdesc *pagetable_alloc(gfp_t gfp, unsigned int order)
2841{
2842 struct page *page = alloc_pages(gfp | __GFP_COMP, order);
2843
2844 return page_ptdesc(page);
2845}
2846
2847/**
2848 * pagetable_free - Free pagetables
2849 * @pt: The page table descriptor
2850 *
2851 * pagetable_free frees the memory of all page tables described by a page
2852 * table descriptor and the memory for the descriptor itself.
2853 */
2854static inline void pagetable_free(struct ptdesc *pt)
2855{
2856 struct page *page = ptdesc_page(pt);
2857
2858 __free_pages(page, compound_order(page));
2859}
2860
2861#if USE_SPLIT_PTE_PTLOCKS
2862#if ALLOC_SPLIT_PTLOCKS
2863void __init ptlock_cache_init(void);
2864bool ptlock_alloc(struct ptdesc *ptdesc);
2865void ptlock_free(struct ptdesc *ptdesc);
2866
2867static inline spinlock_t *ptlock_ptr(struct ptdesc *ptdesc)
2868{
2869 return ptdesc->ptl;
2870}
2871#else /* ALLOC_SPLIT_PTLOCKS */
2872static inline void ptlock_cache_init(void)
2873{
2874}
2875
2876static inline bool ptlock_alloc(struct ptdesc *ptdesc)
2877{
2878 return true;
2879}
2880
2881static inline void ptlock_free(struct ptdesc *ptdesc)
2882{
2883}
2884
2885static inline spinlock_t *ptlock_ptr(struct ptdesc *ptdesc)
2886{
2887 return &ptdesc->ptl;
2888}
2889#endif /* ALLOC_SPLIT_PTLOCKS */
2890
2891static inline spinlock_t *pte_lockptr(struct mm_struct *mm, pmd_t *pmd)
2892{
2893 return ptlock_ptr(page_ptdesc(pmd_page(*pmd)));
2894}
2895
2896static inline bool ptlock_init(struct ptdesc *ptdesc)
2897{
2898 /*
2899 * prep_new_page() initialize page->private (and therefore page->ptl)
2900 * with 0. Make sure nobody took it in use in between.
2901 *
2902 * It can happen if arch try to use slab for page table allocation:
2903 * slab code uses page->slab_cache, which share storage with page->ptl.
2904 */
2905 VM_BUG_ON_PAGE(*(unsigned long *)&ptdesc->ptl, ptdesc_page(ptdesc));
2906 if (!ptlock_alloc(ptdesc))
2907 return false;
2908 spin_lock_init(ptlock_ptr(ptdesc));
2909 return true;
2910}
2911
2912#else /* !USE_SPLIT_PTE_PTLOCKS */
2913/*
2914 * We use mm->page_table_lock to guard all pagetable pages of the mm.
2915 */
2916static inline spinlock_t *pte_lockptr(struct mm_struct *mm, pmd_t *pmd)
2917{
2918 return &mm->page_table_lock;
2919}
2920static inline void ptlock_cache_init(void) {}
2921static inline bool ptlock_init(struct ptdesc *ptdesc) { return true; }
2922static inline void ptlock_free(struct ptdesc *ptdesc) {}
2923#endif /* USE_SPLIT_PTE_PTLOCKS */
2924
2925static inline bool pagetable_pte_ctor(struct ptdesc *ptdesc)
2926{
2927 struct folio *folio = ptdesc_folio(ptdesc);
2928
2929 if (!ptlock_init(ptdesc))
2930 return false;
2931 __folio_set_pgtable(folio);
2932 lruvec_stat_add_folio(folio, NR_PAGETABLE);
2933 return true;
2934}
2935
2936static inline void pagetable_pte_dtor(struct ptdesc *ptdesc)
2937{
2938 struct folio *folio = ptdesc_folio(ptdesc);
2939
2940 ptlock_free(ptdesc);
2941 __folio_clear_pgtable(folio);
2942 lruvec_stat_sub_folio(folio, NR_PAGETABLE);
2943}
2944
2945pte_t *__pte_offset_map(pmd_t *pmd, unsigned long addr, pmd_t *pmdvalp);
2946static inline pte_t *pte_offset_map(pmd_t *pmd, unsigned long addr)
2947{
2948 return __pte_offset_map(pmd, addr, NULL);
2949}
2950
2951pte_t *__pte_offset_map_lock(struct mm_struct *mm, pmd_t *pmd,
2952 unsigned long addr, spinlock_t **ptlp);
2953static inline pte_t *pte_offset_map_lock(struct mm_struct *mm, pmd_t *pmd,
2954 unsigned long addr, spinlock_t **ptlp)
2955{
2956 pte_t *pte;
2957
2958 __cond_lock(*ptlp, pte = __pte_offset_map_lock(mm, pmd, addr, ptlp));
2959 return pte;
2960}
2961
2962pte_t *pte_offset_map_nolock(struct mm_struct *mm, pmd_t *pmd,
2963 unsigned long addr, spinlock_t **ptlp);
2964
2965#define pte_unmap_unlock(pte, ptl) do { \
2966 spin_unlock(ptl); \
2967 pte_unmap(pte); \
2968} while (0)
2969
2970#define pte_alloc(mm, pmd) (unlikely(pmd_none(*(pmd))) && __pte_alloc(mm, pmd))
2971
2972#define pte_alloc_map(mm, pmd, address) \
2973 (pte_alloc(mm, pmd) ? NULL : pte_offset_map(pmd, address))
2974
2975#define pte_alloc_map_lock(mm, pmd, address, ptlp) \
2976 (pte_alloc(mm, pmd) ? \
2977 NULL : pte_offset_map_lock(mm, pmd, address, ptlp))
2978
2979#define pte_alloc_kernel(pmd, address) \
2980 ((unlikely(pmd_none(*(pmd))) && __pte_alloc_kernel(pmd))? \
2981 NULL: pte_offset_kernel(pmd, address))
2982
2983#if USE_SPLIT_PMD_PTLOCKS
2984
2985static inline struct page *pmd_pgtable_page(pmd_t *pmd)
2986{
2987 unsigned long mask = ~(PTRS_PER_PMD * sizeof(pmd_t) - 1);
2988 return virt_to_page((void *)((unsigned long) pmd & mask));
2989}
2990
2991static inline struct ptdesc *pmd_ptdesc(pmd_t *pmd)
2992{
2993 return page_ptdesc(pmd_pgtable_page(pmd));
2994}
2995
2996static inline spinlock_t *pmd_lockptr(struct mm_struct *mm, pmd_t *pmd)
2997{
2998 return ptlock_ptr(pmd_ptdesc(pmd));
2999}
3000
3001static inline bool pmd_ptlock_init(struct ptdesc *ptdesc)
3002{
3003#ifdef CONFIG_TRANSPARENT_HUGEPAGE
3004 ptdesc->pmd_huge_pte = NULL;
3005#endif
3006 return ptlock_init(ptdesc);
3007}
3008
3009static inline void pmd_ptlock_free(struct ptdesc *ptdesc)
3010{
3011#ifdef CONFIG_TRANSPARENT_HUGEPAGE
3012 VM_BUG_ON_PAGE(ptdesc->pmd_huge_pte, ptdesc_page(ptdesc));
3013#endif
3014 ptlock_free(ptdesc);
3015}
3016
3017#define pmd_huge_pte(mm, pmd) (pmd_ptdesc(pmd)->pmd_huge_pte)
3018
3019#else
3020
3021static inline spinlock_t *pmd_lockptr(struct mm_struct *mm, pmd_t *pmd)
3022{
3023 return &mm->page_table_lock;
3024}
3025
3026static inline bool pmd_ptlock_init(struct ptdesc *ptdesc) { return true; }
3027static inline void pmd_ptlock_free(struct ptdesc *ptdesc) {}
3028
3029#define pmd_huge_pte(mm, pmd) ((mm)->pmd_huge_pte)
3030
3031#endif
3032
3033static inline spinlock_t *pmd_lock(struct mm_struct *mm, pmd_t *pmd)
3034{
3035 spinlock_t *ptl = pmd_lockptr(mm, pmd);
3036 spin_lock(ptl);
3037 return ptl;
3038}
3039
3040static inline bool pagetable_pmd_ctor(struct ptdesc *ptdesc)
3041{
3042 struct folio *folio = ptdesc_folio(ptdesc);
3043
3044 if (!pmd_ptlock_init(ptdesc))
3045 return false;
3046 __folio_set_pgtable(folio);
3047 lruvec_stat_add_folio(folio, NR_PAGETABLE);
3048 return true;
3049}
3050
3051static inline void pagetable_pmd_dtor(struct ptdesc *ptdesc)
3052{
3053 struct folio *folio = ptdesc_folio(ptdesc);
3054
3055 pmd_ptlock_free(ptdesc);
3056 __folio_clear_pgtable(folio);
3057 lruvec_stat_sub_folio(folio, NR_PAGETABLE);
3058}
3059
3060/*
3061 * No scalability reason to split PUD locks yet, but follow the same pattern
3062 * as the PMD locks to make it easier if we decide to. The VM should not be
3063 * considered ready to switch to split PUD locks yet; there may be places
3064 * which need to be converted from page_table_lock.
3065 */
3066static inline spinlock_t *pud_lockptr(struct mm_struct *mm, pud_t *pud)
3067{
3068 return &mm->page_table_lock;
3069}
3070
3071static inline spinlock_t *pud_lock(struct mm_struct *mm, pud_t *pud)
3072{
3073 spinlock_t *ptl = pud_lockptr(mm, pud);
3074
3075 spin_lock(ptl);
3076 return ptl;
3077}
3078
3079static inline void pagetable_pud_ctor(struct ptdesc *ptdesc)
3080{
3081 struct folio *folio = ptdesc_folio(ptdesc);
3082
3083 __folio_set_pgtable(folio);
3084 lruvec_stat_add_folio(folio, NR_PAGETABLE);
3085}
3086
3087static inline void pagetable_pud_dtor(struct ptdesc *ptdesc)
3088{
3089 struct folio *folio = ptdesc_folio(ptdesc);
3090
3091 __folio_clear_pgtable(folio);
3092 lruvec_stat_sub_folio(folio, NR_PAGETABLE);
3093}
3094
3095extern void __init pagecache_init(void);
3096extern void free_initmem(void);
3097
3098/*
3099 * Free reserved pages within range [PAGE_ALIGN(start), end & PAGE_MASK)
3100 * into the buddy system. The freed pages will be poisoned with pattern
3101 * "poison" if it's within range [0, UCHAR_MAX].
3102 * Return pages freed into the buddy system.
3103 */
3104extern unsigned long free_reserved_area(void *start, void *end,
3105 int poison, const char *s);
3106
3107extern void adjust_managed_page_count(struct page *page, long count);
3108
3109extern void reserve_bootmem_region(phys_addr_t start,
3110 phys_addr_t end, int nid);
3111
3112/* Free the reserved page into the buddy system, so it gets managed. */
3113static inline void free_reserved_page(struct page *page)
3114{
3115 ClearPageReserved(page);
3116 init_page_count(page);
3117 __free_page(page);
3118 adjust_managed_page_count(page, 1);
3119}
3120#define free_highmem_page(page) free_reserved_page(page)
3121
3122static inline void mark_page_reserved(struct page *page)
3123{
3124 SetPageReserved(page);
3125 adjust_managed_page_count(page, -1);
3126}
3127
3128static inline void free_reserved_ptdesc(struct ptdesc *pt)
3129{
3130 free_reserved_page(ptdesc_page(pt));
3131}
3132
3133/*
3134 * Default method to free all the __init memory into the buddy system.
3135 * The freed pages will be poisoned with pattern "poison" if it's within
3136 * range [0, UCHAR_MAX].
3137 * Return pages freed into the buddy system.
3138 */
3139static inline unsigned long free_initmem_default(int poison)
3140{
3141 extern char __init_begin[], __init_end[];
3142
3143 return free_reserved_area(&__init_begin, &__init_end,
3144 poison, "unused kernel image (initmem)");
3145}
3146
3147static inline unsigned long get_num_physpages(void)
3148{
3149 int nid;
3150 unsigned long phys_pages = 0;
3151
3152 for_each_online_node(nid)
3153 phys_pages += node_present_pages(nid);
3154
3155 return phys_pages;
3156}
3157
3158/*
3159 * Using memblock node mappings, an architecture may initialise its
3160 * zones, allocate the backing mem_map and account for memory holes in an
3161 * architecture independent manner.
3162 *
3163 * An architecture is expected to register range of page frames backed by
3164 * physical memory with memblock_add[_node]() before calling
3165 * free_area_init() passing in the PFN each zone ends at. At a basic
3166 * usage, an architecture is expected to do something like
3167 *
3168 * unsigned long max_zone_pfns[MAX_NR_ZONES] = {max_dma, max_normal_pfn,
3169 * max_highmem_pfn};
3170 * for_each_valid_physical_page_range()
3171 * memblock_add_node(base, size, nid, MEMBLOCK_NONE)
3172 * free_area_init(max_zone_pfns);
3173 */
3174void free_area_init(unsigned long *max_zone_pfn);
3175unsigned long node_map_pfn_alignment(void);
3176unsigned long __absent_pages_in_range(int nid, unsigned long start_pfn,
3177 unsigned long end_pfn);
3178extern unsigned long absent_pages_in_range(unsigned long start_pfn,
3179 unsigned long end_pfn);
3180extern void get_pfn_range_for_nid(unsigned int nid,
3181 unsigned long *start_pfn, unsigned long *end_pfn);
3182
3183#ifndef CONFIG_NUMA
3184static inline int early_pfn_to_nid(unsigned long pfn)
3185{
3186 return 0;
3187}
3188#else
3189/* please see mm/page_alloc.c */
3190extern int __meminit early_pfn_to_nid(unsigned long pfn);
3191#endif
3192
3193extern void set_dma_reserve(unsigned long new_dma_reserve);
3194extern void mem_init(void);
3195extern void __init mmap_init(void);
3196
3197extern void __show_mem(unsigned int flags, nodemask_t *nodemask, int max_zone_idx);
3198static inline void show_mem(void)
3199{
3200 __show_mem(0, NULL, MAX_NR_ZONES - 1);
3201}
3202extern long si_mem_available(void);
3203extern void si_meminfo(struct sysinfo * val);
3204extern void si_meminfo_node(struct sysinfo *val, int nid);
3205#ifdef __HAVE_ARCH_RESERVED_KERNEL_PAGES
3206extern unsigned long arch_reserved_kernel_pages(void);
3207#endif
3208
3209extern __printf(3, 4)
3210void warn_alloc(gfp_t gfp_mask, nodemask_t *nodemask, const char *fmt, ...);
3211
3212extern void setup_per_cpu_pageset(void);
3213
3214/* nommu.c */
3215extern atomic_long_t mmap_pages_allocated;
3216extern int nommu_shrink_inode_mappings(struct inode *, size_t, size_t);
3217
3218/* interval_tree.c */
3219void vma_interval_tree_insert(struct vm_area_struct *node,
3220 struct rb_root_cached *root);
3221void vma_interval_tree_insert_after(struct vm_area_struct *node,
3222 struct vm_area_struct *prev,
3223 struct rb_root_cached *root);
3224void vma_interval_tree_remove(struct vm_area_struct *node,
3225 struct rb_root_cached *root);
3226struct vm_area_struct *vma_interval_tree_iter_first(struct rb_root_cached *root,
3227 unsigned long start, unsigned long last);
3228struct vm_area_struct *vma_interval_tree_iter_next(struct vm_area_struct *node,
3229 unsigned long start, unsigned long last);
3230
3231#define vma_interval_tree_foreach(vma, root, start, last) \
3232 for (vma = vma_interval_tree_iter_first(root, start, last); \
3233 vma; vma = vma_interval_tree_iter_next(vma, start, last))
3234
3235void anon_vma_interval_tree_insert(struct anon_vma_chain *node,
3236 struct rb_root_cached *root);
3237void anon_vma_interval_tree_remove(struct anon_vma_chain *node,
3238 struct rb_root_cached *root);
3239struct anon_vma_chain *
3240anon_vma_interval_tree_iter_first(struct rb_root_cached *root,
3241 unsigned long start, unsigned long last);
3242struct anon_vma_chain *anon_vma_interval_tree_iter_next(
3243 struct anon_vma_chain *node, unsigned long start, unsigned long last);
3244#ifdef CONFIG_DEBUG_VM_RB
3245void anon_vma_interval_tree_verify(struct anon_vma_chain *node);
3246#endif
3247
3248#define anon_vma_interval_tree_foreach(avc, root, start, last) \
3249 for (avc = anon_vma_interval_tree_iter_first(root, start, last); \
3250 avc; avc = anon_vma_interval_tree_iter_next(avc, start, last))
3251
3252/* mmap.c */
3253extern int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin);
3254extern int vma_expand(struct vma_iterator *vmi, struct vm_area_struct *vma,
3255 unsigned long start, unsigned long end, pgoff_t pgoff,
3256 struct vm_area_struct *next);
3257extern int vma_shrink(struct vma_iterator *vmi, struct vm_area_struct *vma,
3258 unsigned long start, unsigned long end, pgoff_t pgoff);
3259extern struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *);
3260extern int insert_vm_struct(struct mm_struct *, struct vm_area_struct *);
3261extern void unlink_file_vma(struct vm_area_struct *);
3262extern struct vm_area_struct *copy_vma(struct vm_area_struct **,
3263 unsigned long addr, unsigned long len, pgoff_t pgoff,
3264 bool *need_rmap_locks);
3265extern void exit_mmap(struct mm_struct *);
3266struct vm_area_struct *vma_modify(struct vma_iterator *vmi,
3267 struct vm_area_struct *prev,
3268 struct vm_area_struct *vma,
3269 unsigned long start, unsigned long end,
3270 unsigned long vm_flags,
3271 struct mempolicy *policy,
3272 struct vm_userfaultfd_ctx uffd_ctx,
3273 struct anon_vma_name *anon_name);
3274
3275/* We are about to modify the VMA's flags. */
3276static inline struct vm_area_struct
3277*vma_modify_flags(struct vma_iterator *vmi,
3278 struct vm_area_struct *prev,
3279 struct vm_area_struct *vma,
3280 unsigned long start, unsigned long end,
3281 unsigned long new_flags)
3282{
3283 return vma_modify(vmi, prev, vma, start, end, new_flags,
3284 vma_policy(vma), vma->vm_userfaultfd_ctx,
3285 anon_vma_name(vma));
3286}
3287
3288/* We are about to modify the VMA's flags and/or anon_name. */
3289static inline struct vm_area_struct
3290*vma_modify_flags_name(struct vma_iterator *vmi,
3291 struct vm_area_struct *prev,
3292 struct vm_area_struct *vma,
3293 unsigned long start,
3294 unsigned long end,
3295 unsigned long new_flags,
3296 struct anon_vma_name *new_name)
3297{
3298 return vma_modify(vmi, prev, vma, start, end, new_flags,
3299 vma_policy(vma), vma->vm_userfaultfd_ctx, new_name);
3300}
3301
3302/* We are about to modify the VMA's memory policy. */
3303static inline struct vm_area_struct
3304*vma_modify_policy(struct vma_iterator *vmi,
3305 struct vm_area_struct *prev,
3306 struct vm_area_struct *vma,
3307 unsigned long start, unsigned long end,
3308 struct mempolicy *new_pol)
3309{
3310 return vma_modify(vmi, prev, vma, start, end, vma->vm_flags,
3311 new_pol, vma->vm_userfaultfd_ctx, anon_vma_name(vma));
3312}
3313
3314/* We are about to modify the VMA's flags and/or uffd context. */
3315static inline struct vm_area_struct
3316*vma_modify_flags_uffd(struct vma_iterator *vmi,
3317 struct vm_area_struct *prev,
3318 struct vm_area_struct *vma,
3319 unsigned long start, unsigned long end,
3320 unsigned long new_flags,
3321 struct vm_userfaultfd_ctx new_ctx)
3322{
3323 return vma_modify(vmi, prev, vma, start, end, new_flags,
3324 vma_policy(vma), new_ctx, anon_vma_name(vma));
3325}
3326
3327static inline int check_data_rlimit(unsigned long rlim,
3328 unsigned long new,
3329 unsigned long start,
3330 unsigned long end_data,
3331 unsigned long start_data)
3332{
3333 if (rlim < RLIM_INFINITY) {
3334 if (((new - start) + (end_data - start_data)) > rlim)
3335 return -ENOSPC;
3336 }
3337
3338 return 0;
3339}
3340
3341extern int mm_take_all_locks(struct mm_struct *mm);
3342extern void mm_drop_all_locks(struct mm_struct *mm);
3343
3344extern int set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file);
3345extern int replace_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file);
3346extern struct file *get_mm_exe_file(struct mm_struct *mm);
3347extern struct file *get_task_exe_file(struct task_struct *task);
3348
3349extern bool may_expand_vm(struct mm_struct *, vm_flags_t, unsigned long npages);
3350extern void vm_stat_account(struct mm_struct *, vm_flags_t, long npages);
3351
3352extern bool vma_is_special_mapping(const struct vm_area_struct *vma,
3353 const struct vm_special_mapping *sm);
3354extern struct vm_area_struct *_install_special_mapping(struct mm_struct *mm,
3355 unsigned long addr, unsigned long len,
3356 unsigned long flags,
3357 const struct vm_special_mapping *spec);
3358/* This is an obsolete alternative to _install_special_mapping. */
3359extern int install_special_mapping(struct mm_struct *mm,
3360 unsigned long addr, unsigned long len,
3361 unsigned long flags, struct page **pages);
3362
3363unsigned long randomize_stack_top(unsigned long stack_top);
3364unsigned long randomize_page(unsigned long start, unsigned long range);
3365
3366extern unsigned long get_unmapped_area(struct file *, unsigned long, unsigned long, unsigned long, unsigned long);
3367
3368extern unsigned long mmap_region(struct file *file, unsigned long addr,
3369 unsigned long len, vm_flags_t vm_flags, unsigned long pgoff,
3370 struct list_head *uf);
3371extern unsigned long do_mmap(struct file *file, unsigned long addr,
3372 unsigned long len, unsigned long prot, unsigned long flags,
3373 vm_flags_t vm_flags, unsigned long pgoff, unsigned long *populate,
3374 struct list_head *uf);
3375extern int do_vmi_munmap(struct vma_iterator *vmi, struct mm_struct *mm,
3376 unsigned long start, size_t len, struct list_head *uf,
3377 bool unlock);
3378extern int do_munmap(struct mm_struct *, unsigned long, size_t,
3379 struct list_head *uf);
3380extern int do_madvise(struct mm_struct *mm, unsigned long start, size_t len_in, int behavior);
3381
3382#ifdef CONFIG_MMU
3383extern int do_vma_munmap(struct vma_iterator *vmi, struct vm_area_struct *vma,
3384 unsigned long start, unsigned long end,
3385 struct list_head *uf, bool unlock);
3386extern int __mm_populate(unsigned long addr, unsigned long len,
3387 int ignore_errors);
3388static inline void mm_populate(unsigned long addr, unsigned long len)
3389{
3390 /* Ignore errors */
3391 (void) __mm_populate(addr, len, 1);
3392}
3393#else
3394static inline void mm_populate(unsigned long addr, unsigned long len) {}
3395#endif
3396
3397/* This takes the mm semaphore itself */
3398extern int __must_check vm_brk_flags(unsigned long, unsigned long, unsigned long);
3399extern int vm_munmap(unsigned long, size_t);
3400extern unsigned long __must_check vm_mmap(struct file *, unsigned long,
3401 unsigned long, unsigned long,
3402 unsigned long, unsigned long);
3403
3404struct vm_unmapped_area_info {
3405#define VM_UNMAPPED_AREA_TOPDOWN 1
3406 unsigned long flags;
3407 unsigned long length;
3408 unsigned long low_limit;
3409 unsigned long high_limit;
3410 unsigned long align_mask;
3411 unsigned long align_offset;
3412};
3413
3414extern unsigned long vm_unmapped_area(struct vm_unmapped_area_info *info);
3415
3416/* truncate.c */
3417extern void truncate_inode_pages(struct address_space *, loff_t);
3418extern void truncate_inode_pages_range(struct address_space *,
3419 loff_t lstart, loff_t lend);
3420extern void truncate_inode_pages_final(struct address_space *);
3421
3422/* generic vm_area_ops exported for stackable file systems */
3423extern vm_fault_t filemap_fault(struct vm_fault *vmf);
3424extern vm_fault_t filemap_map_pages(struct vm_fault *vmf,
3425 pgoff_t start_pgoff, pgoff_t end_pgoff);
3426extern vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf);
3427
3428extern unsigned long stack_guard_gap;
3429/* Generic expand stack which grows the stack according to GROWS{UP,DOWN} */
3430int expand_stack_locked(struct vm_area_struct *vma, unsigned long address);
3431struct vm_area_struct *expand_stack(struct mm_struct * mm, unsigned long addr);
3432
3433/* CONFIG_STACK_GROWSUP still needs to grow downwards at some places */
3434int expand_downwards(struct vm_area_struct *vma, unsigned long address);
3435
3436/* Look up the first VMA which satisfies addr < vm_end, NULL if none. */
3437extern struct vm_area_struct * find_vma(struct mm_struct * mm, unsigned long addr);
3438extern struct vm_area_struct * find_vma_prev(struct mm_struct * mm, unsigned long addr,
3439 struct vm_area_struct **pprev);
3440
3441/*
3442 * Look up the first VMA which intersects the interval [start_addr, end_addr)
3443 * NULL if none. Assume start_addr < end_addr.
3444 */
3445struct vm_area_struct *find_vma_intersection(struct mm_struct *mm,
3446 unsigned long start_addr, unsigned long end_addr);
3447
3448/**
3449 * vma_lookup() - Find a VMA at a specific address
3450 * @mm: The process address space.
3451 * @addr: The user address.
3452 *
3453 * Return: The vm_area_struct at the given address, %NULL otherwise.
3454 */
3455static inline
3456struct vm_area_struct *vma_lookup(struct mm_struct *mm, unsigned long addr)
3457{
3458 return mtree_load(&mm->mm_mt, addr);
3459}
3460
3461static inline unsigned long stack_guard_start_gap(struct vm_area_struct *vma)
3462{
3463 if (vma->vm_flags & VM_GROWSDOWN)
3464 return stack_guard_gap;
3465
3466 /* See reasoning around the VM_SHADOW_STACK definition */
3467 if (vma->vm_flags & VM_SHADOW_STACK)
3468 return PAGE_SIZE;
3469
3470 return 0;
3471}
3472
3473static inline unsigned long vm_start_gap(struct vm_area_struct *vma)
3474{
3475 unsigned long gap = stack_guard_start_gap(vma);
3476 unsigned long vm_start = vma->vm_start;
3477
3478 vm_start -= gap;
3479 if (vm_start > vma->vm_start)
3480 vm_start = 0;
3481 return vm_start;
3482}
3483
3484static inline unsigned long vm_end_gap(struct vm_area_struct *vma)
3485{
3486 unsigned long vm_end = vma->vm_end;
3487
3488 if (vma->vm_flags & VM_GROWSUP) {
3489 vm_end += stack_guard_gap;
3490 if (vm_end < vma->vm_end)
3491 vm_end = -PAGE_SIZE;
3492 }
3493 return vm_end;
3494}
3495
3496static inline unsigned long vma_pages(struct vm_area_struct *vma)
3497{
3498 return (vma->vm_end - vma->vm_start) >> PAGE_SHIFT;
3499}
3500
3501/* Look up the first VMA which exactly match the interval vm_start ... vm_end */
3502static inline struct vm_area_struct *find_exact_vma(struct mm_struct *mm,
3503 unsigned long vm_start, unsigned long vm_end)
3504{
3505 struct vm_area_struct *vma = vma_lookup(mm, vm_start);
3506
3507 if (vma && (vma->vm_start != vm_start || vma->vm_end != vm_end))
3508 vma = NULL;
3509
3510 return vma;
3511}
3512
3513static inline bool range_in_vma(struct vm_area_struct *vma,
3514 unsigned long start, unsigned long end)
3515{
3516 return (vma && vma->vm_start <= start && end <= vma->vm_end);
3517}
3518
3519#ifdef CONFIG_MMU
3520pgprot_t vm_get_page_prot(unsigned long vm_flags);
3521void vma_set_page_prot(struct vm_area_struct *vma);
3522#else
3523static inline pgprot_t vm_get_page_prot(unsigned long vm_flags)
3524{
3525 return __pgprot(0);
3526}
3527static inline void vma_set_page_prot(struct vm_area_struct *vma)
3528{
3529 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
3530}
3531#endif
3532
3533void vma_set_file(struct vm_area_struct *vma, struct file *file);
3534
3535#ifdef CONFIG_NUMA_BALANCING
3536unsigned long change_prot_numa(struct vm_area_struct *vma,
3537 unsigned long start, unsigned long end);
3538#endif
3539
3540struct vm_area_struct *find_extend_vma_locked(struct mm_struct *,
3541 unsigned long addr);
3542int remap_pfn_range(struct vm_area_struct *, unsigned long addr,
3543 unsigned long pfn, unsigned long size, pgprot_t);
3544int remap_pfn_range_notrack(struct vm_area_struct *vma, unsigned long addr,
3545 unsigned long pfn, unsigned long size, pgprot_t prot);
3546int vm_insert_page(struct vm_area_struct *, unsigned long addr, struct page *);
3547int vm_insert_pages(struct vm_area_struct *vma, unsigned long addr,
3548 struct page **pages, unsigned long *num);
3549int vm_map_pages(struct vm_area_struct *vma, struct page **pages,
3550 unsigned long num);
3551int vm_map_pages_zero(struct vm_area_struct *vma, struct page **pages,
3552 unsigned long num);
3553vm_fault_t vmf_insert_pfn(struct vm_area_struct *vma, unsigned long addr,
3554 unsigned long pfn);
3555vm_fault_t vmf_insert_pfn_prot(struct vm_area_struct *vma, unsigned long addr,
3556 unsigned long pfn, pgprot_t pgprot);
3557vm_fault_t vmf_insert_mixed(struct vm_area_struct *vma, unsigned long addr,
3558 pfn_t pfn);
3559vm_fault_t vmf_insert_mixed_mkwrite(struct vm_area_struct *vma,
3560 unsigned long addr, pfn_t pfn);
3561int vm_iomap_memory(struct vm_area_struct *vma, phys_addr_t start, unsigned long len);
3562
3563static inline vm_fault_t vmf_insert_page(struct vm_area_struct *vma,
3564 unsigned long addr, struct page *page)
3565{
3566 int err = vm_insert_page(vma, addr, page);
3567
3568 if (err == -ENOMEM)
3569 return VM_FAULT_OOM;
3570 if (err < 0 && err != -EBUSY)
3571 return VM_FAULT_SIGBUS;
3572
3573 return VM_FAULT_NOPAGE;
3574}
3575
3576#ifndef io_remap_pfn_range
3577static inline int io_remap_pfn_range(struct vm_area_struct *vma,
3578 unsigned long addr, unsigned long pfn,
3579 unsigned long size, pgprot_t prot)
3580{
3581 return remap_pfn_range(vma, addr, pfn, size, pgprot_decrypted(prot));
3582}
3583#endif
3584
3585static inline vm_fault_t vmf_error(int err)
3586{
3587 if (err == -ENOMEM)
3588 return VM_FAULT_OOM;
3589 else if (err == -EHWPOISON)
3590 return VM_FAULT_HWPOISON;
3591 return VM_FAULT_SIGBUS;
3592}
3593
3594/*
3595 * Convert errno to return value for ->page_mkwrite() calls.
3596 *
3597 * This should eventually be merged with vmf_error() above, but will need a
3598 * careful audit of all vmf_error() callers.
3599 */
3600static inline vm_fault_t vmf_fs_error(int err)
3601{
3602 if (err == 0)
3603 return VM_FAULT_LOCKED;
3604 if (err == -EFAULT || err == -EAGAIN)
3605 return VM_FAULT_NOPAGE;
3606 if (err == -ENOMEM)
3607 return VM_FAULT_OOM;
3608 /* -ENOSPC, -EDQUOT, -EIO ... */
3609 return VM_FAULT_SIGBUS;
3610}
3611
3612struct page *follow_page(struct vm_area_struct *vma, unsigned long address,
3613 unsigned int foll_flags);
3614
3615static inline int vm_fault_to_errno(vm_fault_t vm_fault, int foll_flags)
3616{
3617 if (vm_fault & VM_FAULT_OOM)
3618 return -ENOMEM;
3619 if (vm_fault & (VM_FAULT_HWPOISON | VM_FAULT_HWPOISON_LARGE))
3620 return (foll_flags & FOLL_HWPOISON) ? -EHWPOISON : -EFAULT;
3621 if (vm_fault & (VM_FAULT_SIGBUS | VM_FAULT_SIGSEGV))
3622 return -EFAULT;
3623 return 0;
3624}
3625
3626/*
3627 * Indicates whether GUP can follow a PROT_NONE mapped page, or whether
3628 * a (NUMA hinting) fault is required.
3629 */
3630static inline bool gup_can_follow_protnone(struct vm_area_struct *vma,
3631 unsigned int flags)
3632{
3633 /*
3634 * If callers don't want to honor NUMA hinting faults, no need to
3635 * determine if we would actually have to trigger a NUMA hinting fault.
3636 */
3637 if (!(flags & FOLL_HONOR_NUMA_FAULT))
3638 return true;
3639
3640 /*
3641 * NUMA hinting faults don't apply in inaccessible (PROT_NONE) VMAs.
3642 *
3643 * Requiring a fault here even for inaccessible VMAs would mean that
3644 * FOLL_FORCE cannot make any progress, because handle_mm_fault()
3645 * refuses to process NUMA hinting faults in inaccessible VMAs.
3646 */
3647 return !vma_is_accessible(vma);
3648}
3649
3650typedef int (*pte_fn_t)(pte_t *pte, unsigned long addr, void *data);
3651extern int apply_to_page_range(struct mm_struct *mm, unsigned long address,
3652 unsigned long size, pte_fn_t fn, void *data);
3653extern int apply_to_existing_page_range(struct mm_struct *mm,
3654 unsigned long address, unsigned long size,
3655 pte_fn_t fn, void *data);
3656
3657#ifdef CONFIG_PAGE_POISONING
3658extern void __kernel_poison_pages(struct page *page, int numpages);
3659extern void __kernel_unpoison_pages(struct page *page, int numpages);
3660extern bool _page_poisoning_enabled_early;
3661DECLARE_STATIC_KEY_FALSE(_page_poisoning_enabled);
3662static inline bool page_poisoning_enabled(void)
3663{
3664 return _page_poisoning_enabled_early;
3665}
3666/*
3667 * For use in fast paths after init_mem_debugging() has run, or when a
3668 * false negative result is not harmful when called too early.
3669 */
3670static inline bool page_poisoning_enabled_static(void)
3671{
3672 return static_branch_unlikely(&_page_poisoning_enabled);
3673}
3674static inline void kernel_poison_pages(struct page *page, int numpages)
3675{
3676 if (page_poisoning_enabled_static())
3677 __kernel_poison_pages(page, numpages);
3678}
3679static inline void kernel_unpoison_pages(struct page *page, int numpages)
3680{
3681 if (page_poisoning_enabled_static())
3682 __kernel_unpoison_pages(page, numpages);
3683}
3684#else
3685static inline bool page_poisoning_enabled(void) { return false; }
3686static inline bool page_poisoning_enabled_static(void) { return false; }
3687static inline void __kernel_poison_pages(struct page *page, int nunmpages) { }
3688static inline void kernel_poison_pages(struct page *page, int numpages) { }
3689static inline void kernel_unpoison_pages(struct page *page, int numpages) { }
3690#endif
3691
3692DECLARE_STATIC_KEY_MAYBE(CONFIG_INIT_ON_ALLOC_DEFAULT_ON, init_on_alloc);
3693static inline bool want_init_on_alloc(gfp_t flags)
3694{
3695 if (static_branch_maybe(CONFIG_INIT_ON_ALLOC_DEFAULT_ON,
3696 &init_on_alloc))
3697 return true;
3698 return flags & __GFP_ZERO;
3699}
3700
3701DECLARE_STATIC_KEY_MAYBE(CONFIG_INIT_ON_FREE_DEFAULT_ON, init_on_free);
3702static inline bool want_init_on_free(void)
3703{
3704 return static_branch_maybe(CONFIG_INIT_ON_FREE_DEFAULT_ON,
3705 &init_on_free);
3706}
3707
3708extern bool _debug_pagealloc_enabled_early;
3709DECLARE_STATIC_KEY_FALSE(_debug_pagealloc_enabled);
3710
3711static inline bool debug_pagealloc_enabled(void)
3712{
3713 return IS_ENABLED(CONFIG_DEBUG_PAGEALLOC) &&
3714 _debug_pagealloc_enabled_early;
3715}
3716
3717/*
3718 * For use in fast paths after mem_debugging_and_hardening_init() has run,
3719 * or when a false negative result is not harmful when called too early.
3720 */
3721static inline bool debug_pagealloc_enabled_static(void)
3722{
3723 if (!IS_ENABLED(CONFIG_DEBUG_PAGEALLOC))
3724 return false;
3725
3726 return static_branch_unlikely(&_debug_pagealloc_enabled);
3727}
3728
3729/*
3730 * To support DEBUG_PAGEALLOC architecture must ensure that
3731 * __kernel_map_pages() never fails
3732 */
3733extern void __kernel_map_pages(struct page *page, int numpages, int enable);
3734#ifdef CONFIG_DEBUG_PAGEALLOC
3735static inline void debug_pagealloc_map_pages(struct page *page, int numpages)
3736{
3737 if (debug_pagealloc_enabled_static())
3738 __kernel_map_pages(page, numpages, 1);
3739}
3740
3741static inline void debug_pagealloc_unmap_pages(struct page *page, int numpages)
3742{
3743 if (debug_pagealloc_enabled_static())
3744 __kernel_map_pages(page, numpages, 0);
3745}
3746
3747extern unsigned int _debug_guardpage_minorder;
3748DECLARE_STATIC_KEY_FALSE(_debug_guardpage_enabled);
3749
3750static inline unsigned int debug_guardpage_minorder(void)
3751{
3752 return _debug_guardpage_minorder;
3753}
3754
3755static inline bool debug_guardpage_enabled(void)
3756{
3757 return static_branch_unlikely(&_debug_guardpage_enabled);
3758}
3759
3760static inline bool page_is_guard(struct page *page)
3761{
3762 if (!debug_guardpage_enabled())
3763 return false;
3764
3765 return PageGuard(page);
3766}
3767
3768bool __set_page_guard(struct zone *zone, struct page *page, unsigned int order,
3769 int migratetype);
3770static inline bool set_page_guard(struct zone *zone, struct page *page,
3771 unsigned int order, int migratetype)
3772{
3773 if (!debug_guardpage_enabled())
3774 return false;
3775 return __set_page_guard(zone, page, order, migratetype);
3776}
3777
3778void __clear_page_guard(struct zone *zone, struct page *page, unsigned int order,
3779 int migratetype);
3780static inline void clear_page_guard(struct zone *zone, struct page *page,
3781 unsigned int order, int migratetype)
3782{
3783 if (!debug_guardpage_enabled())
3784 return;
3785 __clear_page_guard(zone, page, order, migratetype);
3786}
3787
3788#else /* CONFIG_DEBUG_PAGEALLOC */
3789static inline void debug_pagealloc_map_pages(struct page *page, int numpages) {}
3790static inline void debug_pagealloc_unmap_pages(struct page *page, int numpages) {}
3791static inline unsigned int debug_guardpage_minorder(void) { return 0; }
3792static inline bool debug_guardpage_enabled(void) { return false; }
3793static inline bool page_is_guard(struct page *page) { return false; }
3794static inline bool set_page_guard(struct zone *zone, struct page *page,
3795 unsigned int order, int migratetype) { return false; }
3796static inline void clear_page_guard(struct zone *zone, struct page *page,
3797 unsigned int order, int migratetype) {}
3798#endif /* CONFIG_DEBUG_PAGEALLOC */
3799
3800#ifdef __HAVE_ARCH_GATE_AREA
3801extern struct vm_area_struct *get_gate_vma(struct mm_struct *mm);
3802extern int in_gate_area_no_mm(unsigned long addr);
3803extern int in_gate_area(struct mm_struct *mm, unsigned long addr);
3804#else
3805static inline struct vm_area_struct *get_gate_vma(struct mm_struct *mm)
3806{
3807 return NULL;
3808}
3809static inline int in_gate_area_no_mm(unsigned long addr) { return 0; }
3810static inline int in_gate_area(struct mm_struct *mm, unsigned long addr)
3811{
3812 return 0;
3813}
3814#endif /* __HAVE_ARCH_GATE_AREA */
3815
3816extern bool process_shares_mm(struct task_struct *p, struct mm_struct *mm);
3817
3818#ifdef CONFIG_SYSCTL
3819extern int sysctl_drop_caches;
3820int drop_caches_sysctl_handler(struct ctl_table *, int, void *, size_t *,
3821 loff_t *);
3822#endif
3823
3824void drop_slab(void);
3825
3826#ifndef CONFIG_MMU
3827#define randomize_va_space 0
3828#else
3829extern int randomize_va_space;
3830#endif
3831
3832const char * arch_vma_name(struct vm_area_struct *vma);
3833#ifdef CONFIG_MMU
3834void print_vma_addr(char *prefix, unsigned long rip);
3835#else
3836static inline void print_vma_addr(char *prefix, unsigned long rip)
3837{
3838}
3839#endif
3840
3841void *sparse_buffer_alloc(unsigned long size);
3842struct page * __populate_section_memmap(unsigned long pfn,
3843 unsigned long nr_pages, int nid, struct vmem_altmap *altmap,
3844 struct dev_pagemap *pgmap);
3845void pmd_init(void *addr);
3846void pud_init(void *addr);
3847pgd_t *vmemmap_pgd_populate(unsigned long addr, int node);
3848p4d_t *vmemmap_p4d_populate(pgd_t *pgd, unsigned long addr, int node);
3849pud_t *vmemmap_pud_populate(p4d_t *p4d, unsigned long addr, int node);
3850pmd_t *vmemmap_pmd_populate(pud_t *pud, unsigned long addr, int node);
3851pte_t *vmemmap_pte_populate(pmd_t *pmd, unsigned long addr, int node,
3852 struct vmem_altmap *altmap, struct page *reuse);
3853void *vmemmap_alloc_block(unsigned long size, int node);
3854struct vmem_altmap;
3855void *vmemmap_alloc_block_buf(unsigned long size, int node,
3856 struct vmem_altmap *altmap);
3857void vmemmap_verify(pte_t *, int, unsigned long, unsigned long);
3858void vmemmap_set_pmd(pmd_t *pmd, void *p, int node,
3859 unsigned long addr, unsigned long next);
3860int vmemmap_check_pmd(pmd_t *pmd, int node,
3861 unsigned long addr, unsigned long next);
3862int vmemmap_populate_basepages(unsigned long start, unsigned long end,
3863 int node, struct vmem_altmap *altmap);
3864int vmemmap_populate_hugepages(unsigned long start, unsigned long end,
3865 int node, struct vmem_altmap *altmap);
3866int vmemmap_populate(unsigned long start, unsigned long end, int node,
3867 struct vmem_altmap *altmap);
3868void vmemmap_populate_print_last(void);
3869#ifdef CONFIG_MEMORY_HOTPLUG
3870void vmemmap_free(unsigned long start, unsigned long end,
3871 struct vmem_altmap *altmap);
3872#endif
3873
3874#ifdef CONFIG_SPARSEMEM_VMEMMAP
3875static inline unsigned long vmem_altmap_offset(struct vmem_altmap *altmap)
3876{
3877 /* number of pfns from base where pfn_to_page() is valid */
3878 if (altmap)
3879 return altmap->reserve + altmap->free;
3880 return 0;
3881}
3882
3883static inline void vmem_altmap_free(struct vmem_altmap *altmap,
3884 unsigned long nr_pfns)
3885{
3886 altmap->alloc -= nr_pfns;
3887}
3888#else
3889static inline unsigned long vmem_altmap_offset(struct vmem_altmap *altmap)
3890{
3891 return 0;
3892}
3893
3894static inline void vmem_altmap_free(struct vmem_altmap *altmap,
3895 unsigned long nr_pfns)
3896{
3897}
3898#endif
3899
3900#define VMEMMAP_RESERVE_NR 2
3901#ifdef CONFIG_ARCH_WANT_OPTIMIZE_DAX_VMEMMAP
3902static inline bool __vmemmap_can_optimize(struct vmem_altmap *altmap,
3903 struct dev_pagemap *pgmap)
3904{
3905 unsigned long nr_pages;
3906 unsigned long nr_vmemmap_pages;
3907
3908 if (!pgmap || !is_power_of_2(sizeof(struct page)))
3909 return false;
3910
3911 nr_pages = pgmap_vmemmap_nr(pgmap);
3912 nr_vmemmap_pages = ((nr_pages * sizeof(struct page)) >> PAGE_SHIFT);
3913 /*
3914 * For vmemmap optimization with DAX we need minimum 2 vmemmap
3915 * pages. See layout diagram in Documentation/mm/vmemmap_dedup.rst
3916 */
3917 return !altmap && (nr_vmemmap_pages > VMEMMAP_RESERVE_NR);
3918}
3919/*
3920 * If we don't have an architecture override, use the generic rule
3921 */
3922#ifndef vmemmap_can_optimize
3923#define vmemmap_can_optimize __vmemmap_can_optimize
3924#endif
3925
3926#else
3927static inline bool vmemmap_can_optimize(struct vmem_altmap *altmap,
3928 struct dev_pagemap *pgmap)
3929{
3930 return false;
3931}
3932#endif
3933
3934void register_page_bootmem_memmap(unsigned long section_nr, struct page *map,
3935 unsigned long nr_pages);
3936
3937enum mf_flags {
3938 MF_COUNT_INCREASED = 1 << 0,
3939 MF_ACTION_REQUIRED = 1 << 1,
3940 MF_MUST_KILL = 1 << 2,
3941 MF_SOFT_OFFLINE = 1 << 3,
3942 MF_UNPOISON = 1 << 4,
3943 MF_SW_SIMULATED = 1 << 5,
3944 MF_NO_RETRY = 1 << 6,
3945 MF_MEM_PRE_REMOVE = 1 << 7,
3946};
3947int mf_dax_kill_procs(struct address_space *mapping, pgoff_t index,
3948 unsigned long count, int mf_flags);
3949extern int memory_failure(unsigned long pfn, int flags);
3950extern void memory_failure_queue_kick(int cpu);
3951extern int unpoison_memory(unsigned long pfn);
3952extern void shake_page(struct page *p);
3953extern atomic_long_t num_poisoned_pages __read_mostly;
3954extern int soft_offline_page(unsigned long pfn, int flags);
3955#ifdef CONFIG_MEMORY_FAILURE
3956/*
3957 * Sysfs entries for memory failure handling statistics.
3958 */
3959extern const struct attribute_group memory_failure_attr_group;
3960extern void memory_failure_queue(unsigned long pfn, int flags);
3961extern int __get_huge_page_for_hwpoison(unsigned long pfn, int flags,
3962 bool *migratable_cleared);
3963void num_poisoned_pages_inc(unsigned long pfn);
3964void num_poisoned_pages_sub(unsigned long pfn, long i);
3965struct task_struct *task_early_kill(struct task_struct *tsk, int force_early);
3966#else
3967static inline void memory_failure_queue(unsigned long pfn, int flags)
3968{
3969}
3970
3971static inline int __get_huge_page_for_hwpoison(unsigned long pfn, int flags,
3972 bool *migratable_cleared)
3973{
3974 return 0;
3975}
3976
3977static inline void num_poisoned_pages_inc(unsigned long pfn)
3978{
3979}
3980
3981static inline void num_poisoned_pages_sub(unsigned long pfn, long i)
3982{
3983}
3984#endif
3985
3986#if defined(CONFIG_MEMORY_FAILURE) && defined(CONFIG_KSM)
3987void add_to_kill_ksm(struct task_struct *tsk, struct page *p,
3988 struct vm_area_struct *vma, struct list_head *to_kill,
3989 unsigned long ksm_addr);
3990#endif
3991
3992#if defined(CONFIG_MEMORY_FAILURE) && defined(CONFIG_MEMORY_HOTPLUG)
3993extern void memblk_nr_poison_inc(unsigned long pfn);
3994extern void memblk_nr_poison_sub(unsigned long pfn, long i);
3995#else
3996static inline void memblk_nr_poison_inc(unsigned long pfn)
3997{
3998}
3999
4000static inline void memblk_nr_poison_sub(unsigned long pfn, long i)
4001{
4002}
4003#endif
4004
4005#ifndef arch_memory_failure
4006static inline int arch_memory_failure(unsigned long pfn, int flags)
4007{
4008 return -ENXIO;
4009}
4010#endif
4011
4012#ifndef arch_is_platform_page
4013static inline bool arch_is_platform_page(u64 paddr)
4014{
4015 return false;
4016}
4017#endif
4018
4019/*
4020 * Error handlers for various types of pages.
4021 */
4022enum mf_result {
4023 MF_IGNORED, /* Error: cannot be handled */
4024 MF_FAILED, /* Error: handling failed */
4025 MF_DELAYED, /* Will be handled later */
4026 MF_RECOVERED, /* Successfully recovered */
4027};
4028
4029enum mf_action_page_type {
4030 MF_MSG_KERNEL,
4031 MF_MSG_KERNEL_HIGH_ORDER,
4032 MF_MSG_SLAB,
4033 MF_MSG_DIFFERENT_COMPOUND,
4034 MF_MSG_HUGE,
4035 MF_MSG_FREE_HUGE,
4036 MF_MSG_UNMAP_FAILED,
4037 MF_MSG_DIRTY_SWAPCACHE,
4038 MF_MSG_CLEAN_SWAPCACHE,
4039 MF_MSG_DIRTY_MLOCKED_LRU,
4040 MF_MSG_CLEAN_MLOCKED_LRU,
4041 MF_MSG_DIRTY_UNEVICTABLE_LRU,
4042 MF_MSG_CLEAN_UNEVICTABLE_LRU,
4043 MF_MSG_DIRTY_LRU,
4044 MF_MSG_CLEAN_LRU,
4045 MF_MSG_TRUNCATED_LRU,
4046 MF_MSG_BUDDY,
4047 MF_MSG_DAX,
4048 MF_MSG_UNSPLIT_THP,
4049 MF_MSG_UNKNOWN,
4050};
4051
4052#if defined(CONFIG_TRANSPARENT_HUGEPAGE) || defined(CONFIG_HUGETLBFS)
4053extern void clear_huge_page(struct page *page,
4054 unsigned long addr_hint,
4055 unsigned int pages_per_huge_page);
4056int copy_user_large_folio(struct folio *dst, struct folio *src,
4057 unsigned long addr_hint,
4058 struct vm_area_struct *vma);
4059long copy_folio_from_user(struct folio *dst_folio,
4060 const void __user *usr_src,
4061 bool allow_pagefault);
4062
4063/**
4064 * vma_is_special_huge - Are transhuge page-table entries considered special?
4065 * @vma: Pointer to the struct vm_area_struct to consider
4066 *
4067 * Whether transhuge page-table entries are considered "special" following
4068 * the definition in vm_normal_page().
4069 *
4070 * Return: true if transhuge page-table entries should be considered special,
4071 * false otherwise.
4072 */
4073static inline bool vma_is_special_huge(const struct vm_area_struct *vma)
4074{
4075 return vma_is_dax(vma) || (vma->vm_file &&
4076 (vma->vm_flags & (VM_PFNMAP | VM_MIXEDMAP)));
4077}
4078
4079#endif /* CONFIG_TRANSPARENT_HUGEPAGE || CONFIG_HUGETLBFS */
4080
4081#if MAX_NUMNODES > 1
4082void __init setup_nr_node_ids(void);
4083#else
4084static inline void setup_nr_node_ids(void) {}
4085#endif
4086
4087extern int memcmp_pages(struct page *page1, struct page *page2);
4088
4089static inline int pages_identical(struct page *page1, struct page *page2)
4090{
4091 return !memcmp_pages(page1, page2);
4092}
4093
4094#ifdef CONFIG_MAPPING_DIRTY_HELPERS
4095unsigned long clean_record_shared_mapping_range(struct address_space *mapping,
4096 pgoff_t first_index, pgoff_t nr,
4097 pgoff_t bitmap_pgoff,
4098 unsigned long *bitmap,
4099 pgoff_t *start,
4100 pgoff_t *end);
4101
4102unsigned long wp_shared_mapping_range(struct address_space *mapping,
4103 pgoff_t first_index, pgoff_t nr);
4104#endif
4105
4106extern int sysctl_nr_trim_pages;
4107
4108#ifdef CONFIG_PRINTK
4109void mem_dump_obj(void *object);
4110#else
4111static inline void mem_dump_obj(void *object) {}
4112#endif
4113
4114/**
4115 * seal_check_write - Check for F_SEAL_WRITE or F_SEAL_FUTURE_WRITE flags and
4116 * handle them.
4117 * @seals: the seals to check
4118 * @vma: the vma to operate on
4119 *
4120 * Check whether F_SEAL_WRITE or F_SEAL_FUTURE_WRITE are set; if so, do proper
4121 * check/handling on the vma flags. Return 0 if check pass, or <0 for errors.
4122 */
4123static inline int seal_check_write(int seals, struct vm_area_struct *vma)
4124{
4125 if (seals & (F_SEAL_WRITE | F_SEAL_FUTURE_WRITE)) {
4126 /*
4127 * New PROT_WRITE and MAP_SHARED mmaps are not allowed when
4128 * write seals are active.
4129 */
4130 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_WRITE))
4131 return -EPERM;
4132
4133 /*
4134 * Since an F_SEAL_[FUTURE_]WRITE sealed memfd can be mapped as
4135 * MAP_SHARED and read-only, take care to not allow mprotect to
4136 * revert protections on such mappings. Do this only for shared
4137 * mappings. For private mappings, don't need to mask
4138 * VM_MAYWRITE as we still want them to be COW-writable.
4139 */
4140 if (vma->vm_flags & VM_SHARED)
4141 vm_flags_clear(vma, VM_MAYWRITE);
4142 }
4143
4144 return 0;
4145}
4146
4147#ifdef CONFIG_ANON_VMA_NAME
4148int madvise_set_anon_name(struct mm_struct *mm, unsigned long start,
4149 unsigned long len_in,
4150 struct anon_vma_name *anon_name);
4151#else
4152static inline int
4153madvise_set_anon_name(struct mm_struct *mm, unsigned long start,
4154 unsigned long len_in, struct anon_vma_name *anon_name) {
4155 return 0;
4156}
4157#endif
4158
4159#ifdef CONFIG_UNACCEPTED_MEMORY
4160
4161bool range_contains_unaccepted_memory(phys_addr_t start, phys_addr_t end);
4162void accept_memory(phys_addr_t start, phys_addr_t end);
4163
4164#else
4165
4166static inline bool range_contains_unaccepted_memory(phys_addr_t start,
4167 phys_addr_t end)
4168{
4169 return false;
4170}
4171
4172static inline void accept_memory(phys_addr_t start, phys_addr_t end)
4173{
4174}
4175
4176#endif
4177
4178static inline bool pfn_is_unaccepted_memory(unsigned long pfn)
4179{
4180 phys_addr_t paddr = pfn << PAGE_SHIFT;
4181
4182 return range_contains_unaccepted_memory(paddr, paddr + PAGE_SIZE);
4183}
4184
4185#endif /* _LINUX_MM_H */