1 /* SPDX-License-Identifier: GPL-2.0 */
5 #include <linux/errno.h>
6 #include <linux/mmdebug.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>
36 struct anon_vma_chain;
41 extern int sysctl_page_lock_unfairness;
43 void mm_core_init(void);
44 void init_mm_internals(void);
46 #ifndef CONFIG_NUMA /* Don't use mapnrs, do it properly */
47 extern unsigned long max_mapnr;
49 static inline void set_max_mapnr(unsigned long limit)
54 static inline void set_max_mapnr(unsigned long limit) { }
57 extern atomic_long_t _totalram_pages;
58 static inline unsigned long totalram_pages(void)
60 return (unsigned long)atomic_long_read(&_totalram_pages);
63 static inline void totalram_pages_inc(void)
65 atomic_long_inc(&_totalram_pages);
68 static inline void totalram_pages_dec(void)
70 atomic_long_dec(&_totalram_pages);
73 static inline void totalram_pages_add(long count)
75 atomic_long_add(count, &_totalram_pages);
78 extern void * high_memory;
79 extern int page_cluster;
80 extern const int page_cluster_max;
83 extern int sysctl_legacy_va_layout;
85 #define sysctl_legacy_va_layout 0
88 #ifdef CONFIG_HAVE_ARCH_MMAP_RND_BITS
89 extern const int mmap_rnd_bits_min;
90 extern int mmap_rnd_bits_max __ro_after_init;
91 extern int mmap_rnd_bits __read_mostly;
93 #ifdef CONFIG_HAVE_ARCH_MMAP_RND_COMPAT_BITS
94 extern const int mmap_rnd_compat_bits_min;
95 extern const int mmap_rnd_compat_bits_max;
96 extern int mmap_rnd_compat_bits __read_mostly;
100 #include <asm/processor.h>
103 #define __pa_symbol(x) __pa(RELOC_HIDE((unsigned long)(x), 0))
107 #define page_to_virt(x) __va(PFN_PHYS(page_to_pfn(x)))
111 #define lm_alias(x) __va(__pa_symbol(x))
115 * To prevent common memory management code establishing
116 * a zero page mapping on a read fault.
117 * This macro should be defined within <asm/pgtable.h>.
118 * s390 does this to prevent multiplexing of hardware bits
119 * related to the physical page in case of virtualization.
121 #ifndef mm_forbids_zeropage
122 #define mm_forbids_zeropage(X) (0)
126 * On some architectures it is expensive to call memset() for small sizes.
127 * If an architecture decides to implement their own version of
128 * mm_zero_struct_page they should wrap the defines below in a #ifndef and
129 * define their own version of this macro in <asm/pgtable.h>
131 #if BITS_PER_LONG == 64
132 /* This function must be updated when the size of struct page grows above 96
133 * or reduces below 56. The idea that compiler optimizes out switch()
134 * statement, and only leaves move/store instructions. Also the compiler can
135 * combine write statements if they are both assignments and can be reordered,
136 * this can result in several of the writes here being dropped.
138 #define mm_zero_struct_page(pp) __mm_zero_struct_page(pp)
139 static inline void __mm_zero_struct_page(struct page *page)
141 unsigned long *_pp = (void *)page;
143 /* Check that struct page is either 56, 64, 72, 80, 88 or 96 bytes */
144 BUILD_BUG_ON(sizeof(struct page) & 7);
145 BUILD_BUG_ON(sizeof(struct page) < 56);
146 BUILD_BUG_ON(sizeof(struct page) > 96);
148 switch (sizeof(struct page)) {
175 #define mm_zero_struct_page(pp) ((void)memset((pp), 0, sizeof(struct page)))
179 * Default maximum number of active map areas, this limits the number of vmas
180 * per mm struct. Users can overwrite this number by sysctl but there is a
183 * When a program's coredump is generated as ELF format, a section is created
184 * per a vma. In ELF, the number of sections is represented in unsigned short.
185 * This means the number of sections should be smaller than 65535 at coredump.
186 * Because the kernel adds some informative sections to a image of program at
187 * generating coredump, we need some margin. The number of extra sections is
188 * 1-3 now and depends on arch. We use "5" as safe margin, here.
190 * ELF extended numbering allows more than 65535 sections, so 16-bit bound is
191 * not a hard limit any more. Although some userspace tools can be surprised by
194 #define MAPCOUNT_ELF_CORE_MARGIN (5)
195 #define DEFAULT_MAX_MAP_COUNT (USHRT_MAX - MAPCOUNT_ELF_CORE_MARGIN)
197 extern int sysctl_max_map_count;
199 extern unsigned long sysctl_user_reserve_kbytes;
200 extern unsigned long sysctl_admin_reserve_kbytes;
202 extern int sysctl_overcommit_memory;
203 extern int sysctl_overcommit_ratio;
204 extern unsigned long sysctl_overcommit_kbytes;
206 int overcommit_ratio_handler(struct ctl_table *, int, void *, size_t *,
208 int overcommit_kbytes_handler(struct ctl_table *, int, void *, size_t *,
210 int overcommit_policy_handler(struct ctl_table *, int, void *, size_t *,
213 #if defined(CONFIG_SPARSEMEM) && !defined(CONFIG_SPARSEMEM_VMEMMAP)
214 #define nth_page(page,n) pfn_to_page(page_to_pfn((page)) + (n))
215 #define folio_page_idx(folio, p) (page_to_pfn(p) - folio_pfn(folio))
217 #define nth_page(page,n) ((page) + (n))
218 #define folio_page_idx(folio, p) ((p) - &(folio)->page)
221 /* to align the pointer to the (next) page boundary */
222 #define PAGE_ALIGN(addr) ALIGN(addr, PAGE_SIZE)
224 /* to align the pointer to the (prev) page boundary */
225 #define PAGE_ALIGN_DOWN(addr) ALIGN_DOWN(addr, PAGE_SIZE)
227 /* test whether an address (unsigned long or pointer) is aligned to PAGE_SIZE */
228 #define PAGE_ALIGNED(addr) IS_ALIGNED((unsigned long)(addr), PAGE_SIZE)
230 static inline struct folio *lru_to_folio(struct list_head *head)
232 return list_entry((head)->prev, struct folio, lru);
235 void setup_initial_init_mm(void *start_code, void *end_code,
236 void *end_data, void *brk);
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
247 struct vm_area_struct *vm_area_alloc(struct mm_struct *);
248 struct vm_area_struct *vm_area_dup(struct vm_area_struct *);
249 void vm_area_free(struct vm_area_struct *);
250 /* Use only if VMA has no other users */
251 void __vm_area_free(struct vm_area_struct *vma);
254 extern struct rb_root nommu_region_tree;
255 extern struct rw_semaphore nommu_region_sem;
257 extern unsigned int kobjsize(const void *objp);
261 * vm_flags in vm_area_struct, see mm_types.h.
262 * When changing, update also include/trace/events/mmflags.h
264 #define VM_NONE 0x00000000
266 #define VM_READ 0x00000001 /* currently active flags */
267 #define VM_WRITE 0x00000002
268 #define VM_EXEC 0x00000004
269 #define VM_SHARED 0x00000008
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
277 #define VM_GROWSDOWN 0x00000100 /* general info on the segment */
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 */
287 #define VM_LOCKED 0x00002000
288 #define VM_IO 0x00004000 /* Memory mapped I/O or similar */
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 */
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 */
305 #ifdef CONFIG_MEM_SOFT_DIRTY
306 # define VM_SOFTDIRTY 0x08000000 /* Not soft dirty clean area */
308 # define VM_SOFTDIRTY 0
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 */
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 */
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
338 # define VM_PKEY_BIT4 VM_HIGH_ARCH_4
340 # define VM_PKEY_BIT4 0
342 #endif /* CONFIG_ARCH_HAS_PKEYS */
344 #ifdef CONFIG_X86_USER_SHADOW_STACK
346 * VM_SHADOW_STACK should not be set with VM_SHARED because of lack of
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.
354 # define VM_SHADOW_STACK VM_HIGH_ARCH_5
356 # define VM_SHADOW_STACK VM_NONE
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) */
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 */
379 # define VM_MTE VM_NONE
380 # define VM_MTE_ALLOWED VM_NONE
384 # define VM_GROWSUP VM_NONE
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 */
395 * This flag is used to connect VFIO to arch specific KVM code. It
396 * indicates that the memory under this VMA is safe for use with any
397 * non-cachable memory type inside KVM. Some VFIO devices, on some
398 * platforms, are thought to be unsafe and can cause machine crashes
399 * if KVM does not lock down the memory type.
402 #define VM_ALLOW_ANY_UNCACHED_BIT 39
403 #define VM_ALLOW_ANY_UNCACHED BIT(VM_ALLOW_ANY_UNCACHED_BIT)
405 #define VM_ALLOW_ANY_UNCACHED VM_NONE
408 /* Bits set in the VMA until the stack is in its final location */
409 #define VM_STACK_INCOMPLETE_SETUP (VM_RAND_READ | VM_SEQ_READ | VM_STACK_EARLY)
411 #define TASK_EXEC ((current->personality & READ_IMPLIES_EXEC) ? VM_EXEC : 0)
413 /* Common data flag combinations */
414 #define VM_DATA_FLAGS_TSK_EXEC (VM_READ | VM_WRITE | TASK_EXEC | \
415 VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC)
416 #define VM_DATA_FLAGS_NON_EXEC (VM_READ | VM_WRITE | VM_MAYREAD | \
417 VM_MAYWRITE | VM_MAYEXEC)
418 #define VM_DATA_FLAGS_EXEC (VM_READ | VM_WRITE | VM_EXEC | \
419 VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC)
421 #ifndef VM_DATA_DEFAULT_FLAGS /* arch can override this */
422 #define VM_DATA_DEFAULT_FLAGS VM_DATA_FLAGS_EXEC
425 #ifndef VM_STACK_DEFAULT_FLAGS /* arch can override this */
426 #define VM_STACK_DEFAULT_FLAGS VM_DATA_DEFAULT_FLAGS
429 #define VM_STARTGAP_FLAGS (VM_GROWSDOWN | VM_SHADOW_STACK)
431 #ifdef CONFIG_STACK_GROWSUP
432 #define VM_STACK VM_GROWSUP
433 #define VM_STACK_EARLY VM_GROWSDOWN
435 #define VM_STACK VM_GROWSDOWN
436 #define VM_STACK_EARLY 0
439 #define VM_STACK_FLAGS (VM_STACK | VM_STACK_DEFAULT_FLAGS | VM_ACCOUNT)
441 /* VMA basic access permission flags */
442 #define VM_ACCESS_FLAGS (VM_READ | VM_WRITE | VM_EXEC)
446 * Special vmas that are non-mergable, non-mlock()able.
448 #define VM_SPECIAL (VM_IO | VM_DONTEXPAND | VM_PFNMAP | VM_MIXEDMAP)
450 /* This mask prevents VMA from being scanned with khugepaged */
451 #define VM_NO_KHUGEPAGED (VM_SPECIAL | VM_HUGETLB)
453 /* This mask defines which mm->def_flags a process can inherit its parent */
454 #define VM_INIT_DEF_MASK VM_NOHUGEPAGE
456 /* This mask represents all the VMA flag bits used by mlock */
457 #define VM_LOCKED_MASK (VM_LOCKED | VM_LOCKONFAULT)
459 /* Arch-specific flags to clear when updating VM flags on protection change */
460 #ifndef VM_ARCH_CLEAR
461 # define VM_ARCH_CLEAR VM_NONE
463 #define VM_FLAGS_CLEAR (ARCH_VM_PKEY_FLAGS | VM_ARCH_CLEAR)
466 * mapping from the currently active vm_flags protection bits (the
467 * low four bits) to a page protection mask..
471 * The default fault flags that should be used by most of the
472 * arch-specific page fault handlers.
474 #define FAULT_FLAG_DEFAULT (FAULT_FLAG_ALLOW_RETRY | \
475 FAULT_FLAG_KILLABLE | \
476 FAULT_FLAG_INTERRUPTIBLE)
479 * fault_flag_allow_retry_first - check ALLOW_RETRY the first time
480 * @flags: Fault flags.
482 * This is mostly used for places where we want to try to avoid taking
483 * the mmap_lock for too long a time when waiting for another condition
484 * to change, in which case we can try to be polite to release the
485 * mmap_lock in the first round to avoid potential starvation of other
486 * processes that would also want the mmap_lock.
488 * Return: true if the page fault allows retry and this is the first
489 * attempt of the fault handling; false otherwise.
491 static inline bool fault_flag_allow_retry_first(enum fault_flag flags)
493 return (flags & FAULT_FLAG_ALLOW_RETRY) &&
494 (!(flags & FAULT_FLAG_TRIED));
497 #define FAULT_FLAG_TRACE \
498 { FAULT_FLAG_WRITE, "WRITE" }, \
499 { FAULT_FLAG_MKWRITE, "MKWRITE" }, \
500 { FAULT_FLAG_ALLOW_RETRY, "ALLOW_RETRY" }, \
501 { FAULT_FLAG_RETRY_NOWAIT, "RETRY_NOWAIT" }, \
502 { FAULT_FLAG_KILLABLE, "KILLABLE" }, \
503 { FAULT_FLAG_TRIED, "TRIED" }, \
504 { FAULT_FLAG_USER, "USER" }, \
505 { FAULT_FLAG_REMOTE, "REMOTE" }, \
506 { FAULT_FLAG_INSTRUCTION, "INSTRUCTION" }, \
507 { FAULT_FLAG_INTERRUPTIBLE, "INTERRUPTIBLE" }, \
508 { FAULT_FLAG_VMA_LOCK, "VMA_LOCK" }
511 * vm_fault is filled by the pagefault handler and passed to the vma's
512 * ->fault function. The vma's ->fault is responsible for returning a bitmask
513 * of VM_FAULT_xxx flags that give details about how the fault was handled.
515 * MM layer fills up gfp_mask for page allocations but fault handler might
516 * alter it if its implementation requires a different allocation context.
518 * pgoff should be used in favour of virtual_address, if possible.
522 struct vm_area_struct *vma; /* Target VMA */
523 gfp_t gfp_mask; /* gfp mask to be used for allocations */
524 pgoff_t pgoff; /* Logical page offset based on vma */
525 unsigned long address; /* Faulting virtual address - masked */
526 unsigned long real_address; /* Faulting virtual address - unmasked */
528 enum fault_flag flags; /* FAULT_FLAG_xxx flags
529 * XXX: should really be 'const' */
530 pmd_t *pmd; /* Pointer to pmd entry matching
532 pud_t *pud; /* Pointer to pud entry matching
536 pte_t orig_pte; /* Value of PTE at the time of fault */
537 pmd_t orig_pmd; /* Value of PMD at the time of fault,
538 * used by PMD fault only.
542 struct page *cow_page; /* Page handler may use for COW fault */
543 struct page *page; /* ->fault handlers should return a
544 * page here, unless VM_FAULT_NOPAGE
545 * is set (which is also implied by
548 /* These three entries are valid only while holding ptl lock */
549 pte_t *pte; /* Pointer to pte entry matching
550 * the 'address'. NULL if the page
551 * table hasn't been allocated.
553 spinlock_t *ptl; /* Page table lock.
554 * Protects pte page table if 'pte'
555 * is not NULL, otherwise pmd.
557 pgtable_t prealloc_pte; /* Pre-allocated pte page table.
558 * vm_ops->map_pages() sets up a page
559 * table from atomic context.
560 * do_fault_around() pre-allocates
561 * page table to avoid allocation from
567 * These are the virtual MM functions - opening of an area, closing and
568 * unmapping it (needed to keep files on disk up-to-date etc), pointer
569 * to the functions called when a no-page or a wp-page exception occurs.
571 struct vm_operations_struct {
572 void (*open)(struct vm_area_struct * area);
574 * @close: Called when the VMA is being removed from the MM.
575 * Context: User context. May sleep. Caller holds mmap_lock.
577 void (*close)(struct vm_area_struct * area);
578 /* Called any time before splitting to check if it's allowed */
579 int (*may_split)(struct vm_area_struct *area, unsigned long addr);
580 int (*mremap)(struct vm_area_struct *area);
582 * Called by mprotect() to make driver-specific permission
583 * checks before mprotect() is finalised. The VMA must not
584 * be modified. Returns 0 if mprotect() can proceed.
586 int (*mprotect)(struct vm_area_struct *vma, unsigned long start,
587 unsigned long end, unsigned long newflags);
588 vm_fault_t (*fault)(struct vm_fault *vmf);
589 vm_fault_t (*huge_fault)(struct vm_fault *vmf, unsigned int order);
590 vm_fault_t (*map_pages)(struct vm_fault *vmf,
591 pgoff_t start_pgoff, pgoff_t end_pgoff);
592 unsigned long (*pagesize)(struct vm_area_struct * area);
594 /* notification that a previously read-only page is about to become
595 * writable, if an error is returned it will cause a SIGBUS */
596 vm_fault_t (*page_mkwrite)(struct vm_fault *vmf);
598 /* same as page_mkwrite when using VM_PFNMAP|VM_MIXEDMAP */
599 vm_fault_t (*pfn_mkwrite)(struct vm_fault *vmf);
601 /* called by access_process_vm when get_user_pages() fails, typically
602 * for use by special VMAs. See also generic_access_phys() for a generic
603 * implementation useful for any iomem mapping.
605 int (*access)(struct vm_area_struct *vma, unsigned long addr,
606 void *buf, int len, int write);
608 /* Called by the /proc/PID/maps code to ask the vma whether it
609 * has a special name. Returning non-NULL will also cause this
610 * vma to be dumped unconditionally. */
611 const char *(*name)(struct vm_area_struct *vma);
615 * set_policy() op must add a reference to any non-NULL @new mempolicy
616 * to hold the policy upon return. Caller should pass NULL @new to
617 * remove a policy and fall back to surrounding context--i.e. do not
618 * install a MPOL_DEFAULT policy, nor the task or system default
621 int (*set_policy)(struct vm_area_struct *vma, struct mempolicy *new);
624 * get_policy() op must add reference [mpol_get()] to any policy at
625 * (vma,addr) marked as MPOL_SHARED. The shared policy infrastructure
626 * in mm/mempolicy.c will do this automatically.
627 * get_policy() must NOT add a ref if the policy at (vma,addr) is not
628 * marked as MPOL_SHARED. vma policies are protected by the mmap_lock.
629 * If no [shared/vma] mempolicy exists at the addr, get_policy() op
630 * must return NULL--i.e., do not "fallback" to task or system default
633 struct mempolicy *(*get_policy)(struct vm_area_struct *vma,
634 unsigned long addr, pgoff_t *ilx);
637 * Called by vm_normal_page() for special PTEs to find the
638 * page for @addr. This is useful if the default behavior
639 * (using pte_page()) would not find the correct page.
641 struct page *(*find_special_page)(struct vm_area_struct *vma,
645 #ifdef CONFIG_NUMA_BALANCING
646 static inline void vma_numab_state_init(struct vm_area_struct *vma)
648 vma->numab_state = NULL;
650 static inline void vma_numab_state_free(struct vm_area_struct *vma)
652 kfree(vma->numab_state);
655 static inline void vma_numab_state_init(struct vm_area_struct *vma) {}
656 static inline void vma_numab_state_free(struct vm_area_struct *vma) {}
657 #endif /* CONFIG_NUMA_BALANCING */
659 #ifdef CONFIG_PER_VMA_LOCK
661 * Try to read-lock a vma. The function is allowed to occasionally yield false
662 * locked result to avoid performance overhead, in which case we fall back to
663 * using mmap_lock. The function should never yield false unlocked result.
665 static inline bool vma_start_read(struct vm_area_struct *vma)
668 * Check before locking. A race might cause false locked result.
669 * We can use READ_ONCE() for the mm_lock_seq here, and don't need
670 * ACQUIRE semantics, because this is just a lockless check whose result
671 * we don't rely on for anything - the mm_lock_seq read against which we
672 * need ordering is below.
674 if (READ_ONCE(vma->vm_lock_seq) == READ_ONCE(vma->vm_mm->mm_lock_seq))
677 if (unlikely(down_read_trylock(&vma->vm_lock->lock) == 0))
681 * Overflow might produce false locked result.
682 * False unlocked result is impossible because we modify and check
683 * vma->vm_lock_seq under vma->vm_lock protection and mm->mm_lock_seq
684 * modification invalidates all existing locks.
686 * We must use ACQUIRE semantics for the mm_lock_seq so that if we are
687 * racing with vma_end_write_all(), we only start reading from the VMA
688 * after it has been unlocked.
689 * This pairs with RELEASE semantics in vma_end_write_all().
691 if (unlikely(vma->vm_lock_seq == smp_load_acquire(&vma->vm_mm->mm_lock_seq))) {
692 up_read(&vma->vm_lock->lock);
698 static inline void vma_end_read(struct vm_area_struct *vma)
700 rcu_read_lock(); /* keeps vma alive till the end of up_read */
701 up_read(&vma->vm_lock->lock);
705 /* WARNING! Can only be used if mmap_lock is expected to be write-locked */
706 static bool __is_vma_write_locked(struct vm_area_struct *vma, int *mm_lock_seq)
708 mmap_assert_write_locked(vma->vm_mm);
711 * current task is holding mmap_write_lock, both vma->vm_lock_seq and
712 * mm->mm_lock_seq can't be concurrently modified.
714 *mm_lock_seq = vma->vm_mm->mm_lock_seq;
715 return (vma->vm_lock_seq == *mm_lock_seq);
719 * Begin writing to a VMA.
720 * Exclude concurrent readers under the per-VMA lock until the currently
721 * write-locked mmap_lock is dropped or downgraded.
723 static inline void vma_start_write(struct vm_area_struct *vma)
727 if (__is_vma_write_locked(vma, &mm_lock_seq))
730 down_write(&vma->vm_lock->lock);
732 * We should use WRITE_ONCE() here because we can have concurrent reads
733 * from the early lockless pessimistic check in vma_start_read().
734 * We don't really care about the correctness of that early check, but
735 * we should use WRITE_ONCE() for cleanliness and to keep KCSAN happy.
737 WRITE_ONCE(vma->vm_lock_seq, mm_lock_seq);
738 up_write(&vma->vm_lock->lock);
741 static inline void vma_assert_write_locked(struct vm_area_struct *vma)
745 VM_BUG_ON_VMA(!__is_vma_write_locked(vma, &mm_lock_seq), vma);
748 static inline void vma_assert_locked(struct vm_area_struct *vma)
750 if (!rwsem_is_locked(&vma->vm_lock->lock))
751 vma_assert_write_locked(vma);
754 static inline void vma_mark_detached(struct vm_area_struct *vma, bool detached)
756 /* When detaching vma should be write-locked */
758 vma_assert_write_locked(vma);
759 vma->detached = detached;
762 static inline void release_fault_lock(struct vm_fault *vmf)
764 if (vmf->flags & FAULT_FLAG_VMA_LOCK)
765 vma_end_read(vmf->vma);
767 mmap_read_unlock(vmf->vma->vm_mm);
770 static inline void assert_fault_locked(struct vm_fault *vmf)
772 if (vmf->flags & FAULT_FLAG_VMA_LOCK)
773 vma_assert_locked(vmf->vma);
775 mmap_assert_locked(vmf->vma->vm_mm);
778 struct vm_area_struct *lock_vma_under_rcu(struct mm_struct *mm,
779 unsigned long address);
781 #else /* CONFIG_PER_VMA_LOCK */
783 static inline bool vma_start_read(struct vm_area_struct *vma)
785 static inline void vma_end_read(struct vm_area_struct *vma) {}
786 static inline void vma_start_write(struct vm_area_struct *vma) {}
787 static inline void vma_assert_write_locked(struct vm_area_struct *vma)
788 { mmap_assert_write_locked(vma->vm_mm); }
789 static inline void vma_mark_detached(struct vm_area_struct *vma,
792 static inline struct vm_area_struct *lock_vma_under_rcu(struct mm_struct *mm,
793 unsigned long address)
798 static inline void vma_assert_locked(struct vm_area_struct *vma)
800 mmap_assert_locked(vma->vm_mm);
803 static inline void release_fault_lock(struct vm_fault *vmf)
805 mmap_read_unlock(vmf->vma->vm_mm);
808 static inline void assert_fault_locked(struct vm_fault *vmf)
810 mmap_assert_locked(vmf->vma->vm_mm);
813 #endif /* CONFIG_PER_VMA_LOCK */
815 extern const struct vm_operations_struct vma_dummy_vm_ops;
818 * WARNING: vma_init does not initialize vma->vm_lock.
819 * Use vm_area_alloc()/vm_area_free() if vma needs locking.
821 static inline void vma_init(struct vm_area_struct *vma, struct mm_struct *mm)
823 memset(vma, 0, sizeof(*vma));
825 vma->vm_ops = &vma_dummy_vm_ops;
826 INIT_LIST_HEAD(&vma->anon_vma_chain);
827 vma_mark_detached(vma, false);
828 vma_numab_state_init(vma);
831 /* Use when VMA is not part of the VMA tree and needs no locking */
832 static inline void vm_flags_init(struct vm_area_struct *vma,
835 ACCESS_PRIVATE(vma, __vm_flags) = flags;
839 * Use when VMA is part of the VMA tree and modifications need coordination
840 * Note: vm_flags_reset and vm_flags_reset_once do not lock the vma and
841 * it should be locked explicitly beforehand.
843 static inline void vm_flags_reset(struct vm_area_struct *vma,
846 vma_assert_write_locked(vma);
847 vm_flags_init(vma, flags);
850 static inline void vm_flags_reset_once(struct vm_area_struct *vma,
853 vma_assert_write_locked(vma);
854 WRITE_ONCE(ACCESS_PRIVATE(vma, __vm_flags), flags);
857 static inline void vm_flags_set(struct vm_area_struct *vma,
860 vma_start_write(vma);
861 ACCESS_PRIVATE(vma, __vm_flags) |= flags;
864 static inline void vm_flags_clear(struct vm_area_struct *vma,
867 vma_start_write(vma);
868 ACCESS_PRIVATE(vma, __vm_flags) &= ~flags;
872 * Use only if VMA is not part of the VMA tree or has no other users and
873 * therefore needs no locking.
875 static inline void __vm_flags_mod(struct vm_area_struct *vma,
876 vm_flags_t set, vm_flags_t clear)
878 vm_flags_init(vma, (vma->vm_flags | set) & ~clear);
882 * Use only when the order of set/clear operations is unimportant, otherwise
883 * use vm_flags_{set|clear} explicitly.
885 static inline void vm_flags_mod(struct vm_area_struct *vma,
886 vm_flags_t set, vm_flags_t clear)
888 vma_start_write(vma);
889 __vm_flags_mod(vma, set, clear);
892 static inline void vma_set_anonymous(struct vm_area_struct *vma)
897 static inline bool vma_is_anonymous(struct vm_area_struct *vma)
903 * Indicate if the VMA is a heap for the given task; for
904 * /proc/PID/maps that is the heap of the main task.
906 static inline bool vma_is_initial_heap(const struct vm_area_struct *vma)
908 return vma->vm_start < vma->vm_mm->brk &&
909 vma->vm_end > vma->vm_mm->start_brk;
913 * Indicate if the VMA is a stack for the given task; for
914 * /proc/PID/maps that is the stack of the main task.
916 static inline bool vma_is_initial_stack(const struct vm_area_struct *vma)
919 * We make no effort to guess what a given thread considers to be
920 * its "stack". It's not even well-defined for programs written
923 return vma->vm_start <= vma->vm_mm->start_stack &&
924 vma->vm_end >= vma->vm_mm->start_stack;
927 static inline bool vma_is_temporary_stack(struct vm_area_struct *vma)
929 int maybe_stack = vma->vm_flags & (VM_GROWSDOWN | VM_GROWSUP);
934 if ((vma->vm_flags & VM_STACK_INCOMPLETE_SETUP) ==
935 VM_STACK_INCOMPLETE_SETUP)
941 static inline bool vma_is_foreign(struct vm_area_struct *vma)
946 if (current->mm != vma->vm_mm)
952 static inline bool vma_is_accessible(struct vm_area_struct *vma)
954 return vma->vm_flags & VM_ACCESS_FLAGS;
957 static inline bool is_shared_maywrite(vm_flags_t vm_flags)
959 return (vm_flags & (VM_SHARED | VM_MAYWRITE)) ==
960 (VM_SHARED | VM_MAYWRITE);
963 static inline bool vma_is_shared_maywrite(struct vm_area_struct *vma)
965 return is_shared_maywrite(vma->vm_flags);
969 struct vm_area_struct *vma_find(struct vma_iterator *vmi, unsigned long max)
971 return mas_find(&vmi->mas, max - 1);
974 static inline struct vm_area_struct *vma_next(struct vma_iterator *vmi)
977 * Uses mas_find() to get the first VMA when the iterator starts.
978 * Calling mas_next() could skip the first entry.
980 return mas_find(&vmi->mas, ULONG_MAX);
984 struct vm_area_struct *vma_iter_next_range(struct vma_iterator *vmi)
986 return mas_next_range(&vmi->mas, ULONG_MAX);
990 static inline struct vm_area_struct *vma_prev(struct vma_iterator *vmi)
992 return mas_prev(&vmi->mas, 0);
996 struct vm_area_struct *vma_iter_prev_range(struct vma_iterator *vmi)
998 return mas_prev_range(&vmi->mas, 0);
1001 static inline unsigned long vma_iter_addr(struct vma_iterator *vmi)
1003 return vmi->mas.index;
1006 static inline unsigned long vma_iter_end(struct vma_iterator *vmi)
1008 return vmi->mas.last + 1;
1010 static inline int vma_iter_bulk_alloc(struct vma_iterator *vmi,
1011 unsigned long count)
1013 return mas_expected_entries(&vmi->mas, count);
1016 static inline int vma_iter_clear_gfp(struct vma_iterator *vmi,
1017 unsigned long start, unsigned long end, gfp_t gfp)
1019 __mas_set_range(&vmi->mas, start, end - 1);
1020 mas_store_gfp(&vmi->mas, NULL, gfp);
1021 if (unlikely(mas_is_err(&vmi->mas)))
1027 /* Free any unused preallocations */
1028 static inline void vma_iter_free(struct vma_iterator *vmi)
1030 mas_destroy(&vmi->mas);
1033 static inline int vma_iter_bulk_store(struct vma_iterator *vmi,
1034 struct vm_area_struct *vma)
1036 vmi->mas.index = vma->vm_start;
1037 vmi->mas.last = vma->vm_end - 1;
1038 mas_store(&vmi->mas, vma);
1039 if (unlikely(mas_is_err(&vmi->mas)))
1045 static inline void vma_iter_invalidate(struct vma_iterator *vmi)
1047 mas_pause(&vmi->mas);
1050 static inline void vma_iter_set(struct vma_iterator *vmi, unsigned long addr)
1052 mas_set(&vmi->mas, addr);
1055 #define for_each_vma(__vmi, __vma) \
1056 while (((__vma) = vma_next(&(__vmi))) != NULL)
1058 /* The MM code likes to work with exclusive end addresses */
1059 #define for_each_vma_range(__vmi, __vma, __end) \
1060 while (((__vma) = vma_find(&(__vmi), (__end))) != NULL)
1064 * The vma_is_shmem is not inline because it is used only by slow
1065 * paths in userfault.
1067 bool vma_is_shmem(struct vm_area_struct *vma);
1068 bool vma_is_anon_shmem(struct vm_area_struct *vma);
1070 static inline bool vma_is_shmem(struct vm_area_struct *vma) { return false; }
1071 static inline bool vma_is_anon_shmem(struct vm_area_struct *vma) { return false; }
1074 int vma_is_stack_for_current(struct vm_area_struct *vma);
1076 /* flush_tlb_range() takes a vma, not a mm, and can care about flags */
1077 #define TLB_FLUSH_VMA(mm,flags) { .vm_mm = (mm), .vm_flags = (flags) }
1083 * compound_order() can be called without holding a reference, which means
1084 * that niceties like page_folio() don't work. These callers should be
1085 * prepared to handle wild return values. For example, PG_head may be
1086 * set before the order is initialised, or this may be a tail page.
1087 * See compaction.c for some good examples.
1089 static inline unsigned int compound_order(struct page *page)
1091 struct folio *folio = (struct folio *)page;
1093 if (!test_bit(PG_head, &folio->flags))
1095 return folio->_flags_1 & 0xff;
1099 * folio_order - The allocation order of a folio.
1100 * @folio: The folio.
1102 * A folio is composed of 2^order pages. See get_order() for the definition
1105 * Return: The order of the folio.
1107 static inline unsigned int folio_order(struct folio *folio)
1109 if (!folio_test_large(folio))
1111 return folio->_flags_1 & 0xff;
1114 #include <linux/huge_mm.h>
1117 * Methods to modify the page usage count.
1119 * What counts for a page usage:
1120 * - cache mapping (page->mapping)
1121 * - private data (page->private)
1122 * - page mapped in a task's page tables, each mapping
1123 * is counted separately
1125 * Also, many kernel routines increase the page count before a critical
1126 * routine so they can be sure the page doesn't go away from under them.
1130 * Drop a ref, return true if the refcount fell to zero (the page has no users)
1132 static inline int put_page_testzero(struct page *page)
1134 VM_BUG_ON_PAGE(page_ref_count(page) == 0, page);
1135 return page_ref_dec_and_test(page);
1138 static inline int folio_put_testzero(struct folio *folio)
1140 return put_page_testzero(&folio->page);
1144 * Try to grab a ref unless the page has a refcount of zero, return false if
1146 * This can be called when MMU is off so it must not access
1147 * any of the virtual mappings.
1149 static inline bool get_page_unless_zero(struct page *page)
1151 return page_ref_add_unless(page, 1, 0);
1154 static inline struct folio *folio_get_nontail_page(struct page *page)
1156 if (unlikely(!get_page_unless_zero(page)))
1158 return (struct folio *)page;
1161 extern int page_is_ram(unsigned long pfn);
1169 int region_intersects(resource_size_t offset, size_t size, unsigned long flags,
1170 unsigned long desc);
1172 /* Support for virtually mapped pages */
1173 struct page *vmalloc_to_page(const void *addr);
1174 unsigned long vmalloc_to_pfn(const void *addr);
1177 * Determine if an address is within the vmalloc range
1179 * On nommu, vmalloc/vfree wrap through kmalloc/kfree directly, so there
1180 * is no special casing required.
1183 extern bool is_vmalloc_addr(const void *x);
1184 extern int is_vmalloc_or_module_addr(const void *x);
1186 static inline bool is_vmalloc_addr(const void *x)
1190 static inline int is_vmalloc_or_module_addr(const void *x)
1197 * How many times the entire folio is mapped as a single unit (eg by a
1198 * PMD or PUD entry). This is probably not what you want, except for
1199 * debugging purposes - it does not include PTE-mapped sub-pages; look
1200 * at folio_mapcount() or page_mapcount() instead.
1202 static inline int folio_entire_mapcount(struct folio *folio)
1204 VM_BUG_ON_FOLIO(!folio_test_large(folio), folio);
1205 return atomic_read(&folio->_entire_mapcount) + 1;
1209 * The atomic page->_mapcount, starts from -1: so that transitions
1210 * both from it and to it can be tracked, using atomic_inc_and_test
1211 * and atomic_add_negative(-1).
1213 static inline void page_mapcount_reset(struct page *page)
1215 atomic_set(&(page)->_mapcount, -1);
1219 * page_mapcount() - Number of times this precise page is mapped.
1222 * The number of times this page is mapped. If this page is part of
1223 * a large folio, it includes the number of times this page is mapped
1224 * as part of that folio.
1226 * Will report 0 for pages which cannot be mapped into userspace, eg
1227 * slab, page tables and similar.
1229 static inline int page_mapcount(struct page *page)
1231 int mapcount = atomic_read(&page->_mapcount) + 1;
1233 /* Handle page_has_type() pages */
1236 if (unlikely(PageCompound(page)))
1237 mapcount += folio_entire_mapcount(page_folio(page));
1242 int folio_total_mapcount(struct folio *folio);
1245 * folio_mapcount() - Calculate the number of mappings of this folio.
1246 * @folio: The folio.
1248 * A large folio tracks both how many times the entire folio is mapped,
1249 * and how many times each individual page in the folio is mapped.
1250 * This function calculates the total number of times the folio is
1253 * Return: The number of times this folio is mapped.
1255 static inline int folio_mapcount(struct folio *folio)
1257 if (likely(!folio_test_large(folio)))
1258 return atomic_read(&folio->_mapcount) + 1;
1259 return folio_total_mapcount(folio);
1262 static inline bool folio_large_is_mapped(struct folio *folio)
1265 * Reading _entire_mapcount below could be omitted if hugetlb
1266 * participated in incrementing nr_pages_mapped when compound mapped.
1268 return atomic_read(&folio->_nr_pages_mapped) > 0 ||
1269 atomic_read(&folio->_entire_mapcount) >= 0;
1273 * folio_mapped - Is this folio mapped into userspace?
1274 * @folio: The folio.
1276 * Return: True if any page in this folio is referenced by user page tables.
1278 static inline bool folio_mapped(struct folio *folio)
1280 if (likely(!folio_test_large(folio)))
1281 return atomic_read(&folio->_mapcount) >= 0;
1282 return folio_large_is_mapped(folio);
1286 * Return true if this page is mapped into pagetables.
1287 * For compound page it returns true if any sub-page of compound page is mapped,
1288 * even if this particular sub-page is not itself mapped by any PTE or PMD.
1290 static inline bool page_mapped(struct page *page)
1292 if (likely(!PageCompound(page)))
1293 return atomic_read(&page->_mapcount) >= 0;
1294 return folio_large_is_mapped(page_folio(page));
1297 static inline struct page *virt_to_head_page(const void *x)
1299 struct page *page = virt_to_page(x);
1301 return compound_head(page);
1304 static inline struct folio *virt_to_folio(const void *x)
1306 struct page *page = virt_to_page(x);
1308 return page_folio(page);
1311 void __folio_put(struct folio *folio);
1313 void put_pages_list(struct list_head *pages);
1315 void split_page(struct page *page, unsigned int order);
1316 void folio_copy(struct folio *dst, struct folio *src);
1318 unsigned long nr_free_buffer_pages(void);
1320 void destroy_large_folio(struct folio *folio);
1322 /* Returns the number of bytes in this potentially compound page. */
1323 static inline unsigned long page_size(struct page *page)
1325 return PAGE_SIZE << compound_order(page);
1328 /* Returns the number of bits needed for the number of bytes in a page */
1329 static inline unsigned int page_shift(struct page *page)
1331 return PAGE_SHIFT + compound_order(page);
1335 * thp_order - Order of a transparent huge page.
1336 * @page: Head page of a transparent huge page.
1338 static inline unsigned int thp_order(struct page *page)
1340 VM_BUG_ON_PGFLAGS(PageTail(page), page);
1341 return compound_order(page);
1345 * thp_size - Size of a transparent huge page.
1346 * @page: Head page of a transparent huge page.
1348 * Return: Number of bytes in this page.
1350 static inline unsigned long thp_size(struct page *page)
1352 return PAGE_SIZE << thp_order(page);
1357 * Do pte_mkwrite, but only if the vma says VM_WRITE. We do this when
1358 * servicing faults for write access. In the normal case, do always want
1359 * pte_mkwrite. But get_user_pages can cause write faults for mappings
1360 * that do not have writing enabled, when used by access_process_vm.
1362 static inline pte_t maybe_mkwrite(pte_t pte, struct vm_area_struct *vma)
1364 if (likely(vma->vm_flags & VM_WRITE))
1365 pte = pte_mkwrite(pte, vma);
1369 vm_fault_t do_set_pmd(struct vm_fault *vmf, struct page *page);
1370 void set_pte_range(struct vm_fault *vmf, struct folio *folio,
1371 struct page *page, unsigned int nr, unsigned long addr);
1373 vm_fault_t finish_fault(struct vm_fault *vmf);
1377 * Multiple processes may "see" the same page. E.g. for untouched
1378 * mappings of /dev/null, all processes see the same page full of
1379 * zeroes, and text pages of executables and shared libraries have
1380 * only one copy in memory, at most, normally.
1382 * For the non-reserved pages, page_count(page) denotes a reference count.
1383 * page_count() == 0 means the page is free. page->lru is then used for
1384 * freelist management in the buddy allocator.
1385 * page_count() > 0 means the page has been allocated.
1387 * Pages are allocated by the slab allocator in order to provide memory
1388 * to kmalloc and kmem_cache_alloc. In this case, the management of the
1389 * page, and the fields in 'struct page' are the responsibility of mm/slab.c
1390 * unless a particular usage is carefully commented. (the responsibility of
1391 * freeing the kmalloc memory is the caller's, of course).
1393 * A page may be used by anyone else who does a __get_free_page().
1394 * In this case, page_count still tracks the references, and should only
1395 * be used through the normal accessor functions. The top bits of page->flags
1396 * and page->virtual store page management information, but all other fields
1397 * are unused and could be used privately, carefully. The management of this
1398 * page is the responsibility of the one who allocated it, and those who have
1399 * subsequently been given references to it.
1401 * The other pages (we may call them "pagecache pages") are completely
1402 * managed by the Linux memory manager: I/O, buffers, swapping etc.
1403 * The following discussion applies only to them.
1405 * A pagecache page contains an opaque `private' member, which belongs to the
1406 * page's address_space. Usually, this is the address of a circular list of
1407 * the page's disk buffers. PG_private must be set to tell the VM to call
1408 * into the filesystem to release these pages.
1410 * A page may belong to an inode's memory mapping. In this case, page->mapping
1411 * is the pointer to the inode, and page->index is the file offset of the page,
1412 * in units of PAGE_SIZE.
1414 * If pagecache pages are not associated with an inode, they are said to be
1415 * anonymous pages. These may become associated with the swapcache, and in that
1416 * case PG_swapcache is set, and page->private is an offset into the swapcache.
1418 * In either case (swapcache or inode backed), the pagecache itself holds one
1419 * reference to the page. Setting PG_private should also increment the
1420 * refcount. The each user mapping also has a reference to the page.
1422 * The pagecache pages are stored in a per-mapping radix tree, which is
1423 * rooted at mapping->i_pages, and indexed by offset.
1424 * Where 2.4 and early 2.6 kernels kept dirty/clean pages in per-address_space
1425 * lists, we instead now tag pages as dirty/writeback in the radix tree.
1427 * All pagecache pages may be subject to I/O:
1428 * - inode pages may need to be read from disk,
1429 * - inode pages which have been modified and are MAP_SHARED may need
1430 * to be written back to the inode on disk,
1431 * - anonymous pages (including MAP_PRIVATE file mappings) which have been
1432 * modified may need to be swapped out to swap space and (later) to be read
1436 #if defined(CONFIG_ZONE_DEVICE) && defined(CONFIG_FS_DAX)
1437 DECLARE_STATIC_KEY_FALSE(devmap_managed_key);
1439 bool __put_devmap_managed_page_refs(struct page *page, int refs);
1440 static inline bool put_devmap_managed_page_refs(struct page *page, int refs)
1442 if (!static_branch_unlikely(&devmap_managed_key))
1444 if (!is_zone_device_page(page))
1446 return __put_devmap_managed_page_refs(page, refs);
1448 #else /* CONFIG_ZONE_DEVICE && CONFIG_FS_DAX */
1449 static inline bool put_devmap_managed_page_refs(struct page *page, int refs)
1453 #endif /* CONFIG_ZONE_DEVICE && CONFIG_FS_DAX */
1455 static inline bool put_devmap_managed_page(struct page *page)
1457 return put_devmap_managed_page_refs(page, 1);
1460 /* 127: arbitrary random number, small enough to assemble well */
1461 #define folio_ref_zero_or_close_to_overflow(folio) \
1462 ((unsigned int) folio_ref_count(folio) + 127u <= 127u)
1465 * folio_get - Increment the reference count on a folio.
1466 * @folio: The folio.
1468 * Context: May be called in any context, as long as you know that
1469 * you have a refcount on the folio. If you do not already have one,
1470 * folio_try_get() may be the right interface for you to use.
1472 static inline void folio_get(struct folio *folio)
1474 VM_BUG_ON_FOLIO(folio_ref_zero_or_close_to_overflow(folio), folio);
1475 folio_ref_inc(folio);
1478 static inline void get_page(struct page *page)
1480 folio_get(page_folio(page));
1483 static inline __must_check bool try_get_page(struct page *page)
1485 page = compound_head(page);
1486 if (WARN_ON_ONCE(page_ref_count(page) <= 0))
1493 * folio_put - Decrement the reference count on a folio.
1494 * @folio: The folio.
1496 * If the folio's reference count reaches zero, the memory will be
1497 * released back to the page allocator and may be used by another
1498 * allocation immediately. Do not access the memory or the struct folio
1499 * after calling folio_put() unless you can be sure that it wasn't the
1502 * Context: May be called in process or interrupt context, but not in NMI
1503 * context. May be called while holding a spinlock.
1505 static inline void folio_put(struct folio *folio)
1507 if (folio_put_testzero(folio))
1512 * folio_put_refs - Reduce the reference count on a folio.
1513 * @folio: The folio.
1514 * @refs: The amount to subtract from the folio's reference count.
1516 * If the folio's reference count reaches zero, the memory will be
1517 * released back to the page allocator and may be used by another
1518 * allocation immediately. Do not access the memory or the struct folio
1519 * after calling folio_put_refs() unless you can be sure that these weren't
1520 * the last references.
1522 * Context: May be called in process or interrupt context, but not in NMI
1523 * context. May be called while holding a spinlock.
1525 static inline void folio_put_refs(struct folio *folio, int refs)
1527 if (folio_ref_sub_and_test(folio, refs))
1531 void folios_put_refs(struct folio_batch *folios, unsigned int *refs);
1534 * union release_pages_arg - an array of pages or folios
1536 * release_pages() releases a simple array of multiple pages, and
1537 * accepts various different forms of said page array: either
1538 * a regular old boring array of pages, an array of folios, or
1539 * an array of encoded page pointers.
1541 * The transparent union syntax for this kind of "any of these
1542 * argument types" is all kinds of ugly, so look away.
1545 struct page **pages;
1546 struct folio **folios;
1547 struct encoded_page **encoded_pages;
1548 } release_pages_arg __attribute__ ((__transparent_union__));
1550 void release_pages(release_pages_arg, int nr);
1553 * folios_put - Decrement the reference count on an array of folios.
1554 * @folios: The folios.
1556 * Like folio_put(), but for a batch of folios. This is more efficient
1557 * than writing the loop yourself as it will optimise the locks which need
1558 * to be taken if the folios are freed. The folios batch is returned
1559 * empty and ready to be reused for another batch; there is no need to
1562 * Context: May be called in process or interrupt context, but not in NMI
1563 * context. May be called while holding a spinlock.
1565 static inline void folios_put(struct folio_batch *folios)
1567 folios_put_refs(folios, NULL);
1570 static inline void put_page(struct page *page)
1572 struct folio *folio = page_folio(page);
1575 * For some devmap managed pages we need to catch refcount transition
1578 if (put_devmap_managed_page(&folio->page))
1584 * GUP_PIN_COUNTING_BIAS, and the associated functions that use it, overload
1585 * the page's refcount so that two separate items are tracked: the original page
1586 * reference count, and also a new count of how many pin_user_pages() calls were
1587 * made against the page. ("gup-pinned" is another term for the latter).
1589 * With this scheme, pin_user_pages() becomes special: such pages are marked as
1590 * distinct from normal pages. As such, the unpin_user_page() call (and its
1591 * variants) must be used in order to release gup-pinned pages.
1595 * By making GUP_PIN_COUNTING_BIAS a power of two, debugging of page reference
1596 * counts with respect to pin_user_pages() and unpin_user_page() becomes
1597 * simpler, due to the fact that adding an even power of two to the page
1598 * refcount has the effect of using only the upper N bits, for the code that
1599 * counts up using the bias value. This means that the lower bits are left for
1600 * the exclusive use of the original code that increments and decrements by one
1601 * (or at least, by much smaller values than the bias value).
1603 * Of course, once the lower bits overflow into the upper bits (and this is
1604 * OK, because subtraction recovers the original values), then visual inspection
1605 * no longer suffices to directly view the separate counts. However, for normal
1606 * applications that don't have huge page reference counts, this won't be an
1609 * Locking: the lockless algorithm described in folio_try_get_rcu()
1610 * provides safe operation for get_user_pages(), page_mkclean() and
1611 * other calls that race to set up page table entries.
1613 #define GUP_PIN_COUNTING_BIAS (1U << 10)
1615 void unpin_user_page(struct page *page);
1616 void unpin_user_pages_dirty_lock(struct page **pages, unsigned long npages,
1618 void unpin_user_page_range_dirty_lock(struct page *page, unsigned long npages,
1620 void unpin_user_pages(struct page **pages, unsigned long npages);
1622 static inline bool is_cow_mapping(vm_flags_t flags)
1624 return (flags & (VM_SHARED | VM_MAYWRITE)) == VM_MAYWRITE;
1628 static inline bool is_nommu_shared_mapping(vm_flags_t flags)
1631 * NOMMU shared mappings are ordinary MAP_SHARED mappings and selected
1632 * R/O MAP_PRIVATE file mappings that are an effective R/O overlay of
1633 * a file mapping. R/O MAP_PRIVATE mappings might still modify
1634 * underlying memory if ptrace is active, so this is only possible if
1635 * ptrace does not apply. Note that there is no mprotect() to upgrade
1636 * write permissions later.
1638 return flags & (VM_MAYSHARE | VM_MAYOVERLAY);
1642 #if defined(CONFIG_SPARSEMEM) && !defined(CONFIG_SPARSEMEM_VMEMMAP)
1643 #define SECTION_IN_PAGE_FLAGS
1647 * The identification function is mainly used by the buddy allocator for
1648 * determining if two pages could be buddies. We are not really identifying
1649 * the zone since we could be using the section number id if we do not have
1650 * node id available in page flags.
1651 * We only guarantee that it will return the same value for two combinable
1654 static inline int page_zone_id(struct page *page)
1656 return (page->flags >> ZONEID_PGSHIFT) & ZONEID_MASK;
1659 #ifdef NODE_NOT_IN_PAGE_FLAGS
1660 int page_to_nid(const struct page *page);
1662 static inline int page_to_nid(const struct page *page)
1664 return (PF_POISONED_CHECK(page)->flags >> NODES_PGSHIFT) & NODES_MASK;
1668 static inline int folio_nid(const struct folio *folio)
1670 return page_to_nid(&folio->page);
1673 #ifdef CONFIG_NUMA_BALANCING
1674 /* page access time bits needs to hold at least 4 seconds */
1675 #define PAGE_ACCESS_TIME_MIN_BITS 12
1676 #if LAST_CPUPID_SHIFT < PAGE_ACCESS_TIME_MIN_BITS
1677 #define PAGE_ACCESS_TIME_BUCKETS \
1678 (PAGE_ACCESS_TIME_MIN_BITS - LAST_CPUPID_SHIFT)
1680 #define PAGE_ACCESS_TIME_BUCKETS 0
1683 #define PAGE_ACCESS_TIME_MASK \
1684 (LAST_CPUPID_MASK << PAGE_ACCESS_TIME_BUCKETS)
1686 static inline int cpu_pid_to_cpupid(int cpu, int pid)
1688 return ((cpu & LAST__CPU_MASK) << LAST__PID_SHIFT) | (pid & LAST__PID_MASK);
1691 static inline int cpupid_to_pid(int cpupid)
1693 return cpupid & LAST__PID_MASK;
1696 static inline int cpupid_to_cpu(int cpupid)
1698 return (cpupid >> LAST__PID_SHIFT) & LAST__CPU_MASK;
1701 static inline int cpupid_to_nid(int cpupid)
1703 return cpu_to_node(cpupid_to_cpu(cpupid));
1706 static inline bool cpupid_pid_unset(int cpupid)
1708 return cpupid_to_pid(cpupid) == (-1 & LAST__PID_MASK);
1711 static inline bool cpupid_cpu_unset(int cpupid)
1713 return cpupid_to_cpu(cpupid) == (-1 & LAST__CPU_MASK);
1716 static inline bool __cpupid_match_pid(pid_t task_pid, int cpupid)
1718 return (task_pid & LAST__PID_MASK) == cpupid_to_pid(cpupid);
1721 #define cpupid_match_pid(task, cpupid) __cpupid_match_pid(task->pid, cpupid)
1722 #ifdef LAST_CPUPID_NOT_IN_PAGE_FLAGS
1723 static inline int folio_xchg_last_cpupid(struct folio *folio, int cpupid)
1725 return xchg(&folio->_last_cpupid, cpupid & LAST_CPUPID_MASK);
1728 static inline int folio_last_cpupid(struct folio *folio)
1730 return folio->_last_cpupid;
1732 static inline void page_cpupid_reset_last(struct page *page)
1734 page->_last_cpupid = -1 & LAST_CPUPID_MASK;
1737 static inline int folio_last_cpupid(struct folio *folio)
1739 return (folio->flags >> LAST_CPUPID_PGSHIFT) & LAST_CPUPID_MASK;
1742 int folio_xchg_last_cpupid(struct folio *folio, int cpupid);
1744 static inline void page_cpupid_reset_last(struct page *page)
1746 page->flags |= LAST_CPUPID_MASK << LAST_CPUPID_PGSHIFT;
1748 #endif /* LAST_CPUPID_NOT_IN_PAGE_FLAGS */
1750 static inline int folio_xchg_access_time(struct folio *folio, int time)
1754 last_time = folio_xchg_last_cpupid(folio,
1755 time >> PAGE_ACCESS_TIME_BUCKETS);
1756 return last_time << PAGE_ACCESS_TIME_BUCKETS;
1759 static inline void vma_set_access_pid_bit(struct vm_area_struct *vma)
1761 unsigned int pid_bit;
1763 pid_bit = hash_32(current->pid, ilog2(BITS_PER_LONG));
1764 if (vma->numab_state && !test_bit(pid_bit, &vma->numab_state->pids_active[1])) {
1765 __set_bit(pid_bit, &vma->numab_state->pids_active[1]);
1768 #else /* !CONFIG_NUMA_BALANCING */
1769 static inline int folio_xchg_last_cpupid(struct folio *folio, int cpupid)
1771 return folio_nid(folio); /* XXX */
1774 static inline int folio_xchg_access_time(struct folio *folio, int time)
1779 static inline int folio_last_cpupid(struct folio *folio)
1781 return folio_nid(folio); /* XXX */
1784 static inline int cpupid_to_nid(int cpupid)
1789 static inline int cpupid_to_pid(int cpupid)
1794 static inline int cpupid_to_cpu(int cpupid)
1799 static inline int cpu_pid_to_cpupid(int nid, int pid)
1804 static inline bool cpupid_pid_unset(int cpupid)
1809 static inline void page_cpupid_reset_last(struct page *page)
1813 static inline bool cpupid_match_pid(struct task_struct *task, int cpupid)
1818 static inline void vma_set_access_pid_bit(struct vm_area_struct *vma)
1821 #endif /* CONFIG_NUMA_BALANCING */
1823 #if defined(CONFIG_KASAN_SW_TAGS) || defined(CONFIG_KASAN_HW_TAGS)
1826 * KASAN per-page tags are stored xor'ed with 0xff. This allows to avoid
1827 * setting tags for all pages to native kernel tag value 0xff, as the default
1828 * value 0x00 maps to 0xff.
1831 static inline u8 page_kasan_tag(const struct page *page)
1833 u8 tag = KASAN_TAG_KERNEL;
1835 if (kasan_enabled()) {
1836 tag = (page->flags >> KASAN_TAG_PGSHIFT) & KASAN_TAG_MASK;
1843 static inline void page_kasan_tag_set(struct page *page, u8 tag)
1845 unsigned long old_flags, flags;
1847 if (!kasan_enabled())
1851 old_flags = READ_ONCE(page->flags);
1854 flags &= ~(KASAN_TAG_MASK << KASAN_TAG_PGSHIFT);
1855 flags |= (tag & KASAN_TAG_MASK) << KASAN_TAG_PGSHIFT;
1856 } while (unlikely(!try_cmpxchg(&page->flags, &old_flags, flags)));
1859 static inline void page_kasan_tag_reset(struct page *page)
1861 if (kasan_enabled())
1862 page_kasan_tag_set(page, KASAN_TAG_KERNEL);
1865 #else /* CONFIG_KASAN_SW_TAGS || CONFIG_KASAN_HW_TAGS */
1867 static inline u8 page_kasan_tag(const struct page *page)
1872 static inline void page_kasan_tag_set(struct page *page, u8 tag) { }
1873 static inline void page_kasan_tag_reset(struct page *page) { }
1875 #endif /* CONFIG_KASAN_SW_TAGS || CONFIG_KASAN_HW_TAGS */
1877 static inline struct zone *page_zone(const struct page *page)
1879 return &NODE_DATA(page_to_nid(page))->node_zones[page_zonenum(page)];
1882 static inline pg_data_t *page_pgdat(const struct page *page)
1884 return NODE_DATA(page_to_nid(page));
1887 static inline struct zone *folio_zone(const struct folio *folio)
1889 return page_zone(&folio->page);
1892 static inline pg_data_t *folio_pgdat(const struct folio *folio)
1894 return page_pgdat(&folio->page);
1897 #ifdef SECTION_IN_PAGE_FLAGS
1898 static inline void set_page_section(struct page *page, unsigned long section)
1900 page->flags &= ~(SECTIONS_MASK << SECTIONS_PGSHIFT);
1901 page->flags |= (section & SECTIONS_MASK) << SECTIONS_PGSHIFT;
1904 static inline unsigned long page_to_section(const struct page *page)
1906 return (page->flags >> SECTIONS_PGSHIFT) & SECTIONS_MASK;
1911 * folio_pfn - Return the Page Frame Number of a folio.
1912 * @folio: The folio.
1914 * A folio may contain multiple pages. The pages have consecutive
1915 * Page Frame Numbers.
1917 * Return: The Page Frame Number of the first page in the folio.
1919 static inline unsigned long folio_pfn(struct folio *folio)
1921 return page_to_pfn(&folio->page);
1924 static inline struct folio *pfn_folio(unsigned long pfn)
1926 return page_folio(pfn_to_page(pfn));
1930 * folio_maybe_dma_pinned - Report if a folio may be pinned for DMA.
1931 * @folio: The folio.
1933 * This function checks if a folio has been pinned via a call to
1934 * a function in the pin_user_pages() family.
1936 * For small folios, the return value is partially fuzzy: false is not fuzzy,
1937 * because it means "definitely not pinned for DMA", but true means "probably
1938 * pinned for DMA, but possibly a false positive due to having at least
1939 * GUP_PIN_COUNTING_BIAS worth of normal folio references".
1941 * False positives are OK, because: a) it's unlikely for a folio to
1942 * get that many refcounts, and b) all the callers of this routine are
1943 * expected to be able to deal gracefully with a false positive.
1945 * For large folios, the result will be exactly correct. That's because
1946 * we have more tracking data available: the _pincount field is used
1947 * instead of the GUP_PIN_COUNTING_BIAS scheme.
1949 * For more information, please see Documentation/core-api/pin_user_pages.rst.
1951 * Return: True, if it is likely that the page has been "dma-pinned".
1952 * False, if the page is definitely not dma-pinned.
1954 static inline bool folio_maybe_dma_pinned(struct folio *folio)
1956 if (folio_test_large(folio))
1957 return atomic_read(&folio->_pincount) > 0;
1960 * folio_ref_count() is signed. If that refcount overflows, then
1961 * folio_ref_count() returns a negative value, and callers will avoid
1962 * further incrementing the refcount.
1964 * Here, for that overflow case, use the sign bit to count a little
1965 * bit higher via unsigned math, and thus still get an accurate result.
1967 return ((unsigned int)folio_ref_count(folio)) >=
1968 GUP_PIN_COUNTING_BIAS;
1971 static inline bool page_maybe_dma_pinned(struct page *page)
1973 return folio_maybe_dma_pinned(page_folio(page));
1977 * This should most likely only be called during fork() to see whether we
1978 * should break the cow immediately for an anon page on the src mm.
1980 * The caller has to hold the PT lock and the vma->vm_mm->->write_protect_seq.
1982 static inline bool folio_needs_cow_for_dma(struct vm_area_struct *vma,
1983 struct folio *folio)
1985 VM_BUG_ON(!(raw_read_seqcount(&vma->vm_mm->write_protect_seq) & 1));
1987 if (!test_bit(MMF_HAS_PINNED, &vma->vm_mm->flags))
1990 return folio_maybe_dma_pinned(folio);
1994 * is_zero_page - Query if a page is a zero page
1995 * @page: The page to query
1997 * This returns true if @page is one of the permanent zero pages.
1999 static inline bool is_zero_page(const struct page *page)
2001 return is_zero_pfn(page_to_pfn(page));
2005 * is_zero_folio - Query if a folio is a zero page
2006 * @folio: The folio to query
2008 * This returns true if @folio is one of the permanent zero pages.
2010 static inline bool is_zero_folio(const struct folio *folio)
2012 return is_zero_page(&folio->page);
2015 /* MIGRATE_CMA and ZONE_MOVABLE do not allow pin folios */
2016 #ifdef CONFIG_MIGRATION
2017 static inline bool folio_is_longterm_pinnable(struct folio *folio)
2020 int mt = folio_migratetype(folio);
2022 if (mt == MIGRATE_CMA || mt == MIGRATE_ISOLATE)
2025 /* The zero page can be "pinned" but gets special handling. */
2026 if (is_zero_folio(folio))
2029 /* Coherent device memory must always allow eviction. */
2030 if (folio_is_device_coherent(folio))
2033 /* Otherwise, non-movable zone folios can be pinned. */
2034 return !folio_is_zone_movable(folio);
2038 static inline bool folio_is_longterm_pinnable(struct folio *folio)
2044 static inline void set_page_zone(struct page *page, enum zone_type zone)
2046 page->flags &= ~(ZONES_MASK << ZONES_PGSHIFT);
2047 page->flags |= (zone & ZONES_MASK) << ZONES_PGSHIFT;
2050 static inline void set_page_node(struct page *page, unsigned long node)
2052 page->flags &= ~(NODES_MASK << NODES_PGSHIFT);
2053 page->flags |= (node & NODES_MASK) << NODES_PGSHIFT;
2056 static inline void set_page_links(struct page *page, enum zone_type zone,
2057 unsigned long node, unsigned long pfn)
2059 set_page_zone(page, zone);
2060 set_page_node(page, node);
2061 #ifdef SECTION_IN_PAGE_FLAGS
2062 set_page_section(page, pfn_to_section_nr(pfn));
2067 * folio_nr_pages - The number of pages in the folio.
2068 * @folio: The folio.
2070 * Return: A positive power of two.
2072 static inline long folio_nr_pages(struct folio *folio)
2074 if (!folio_test_large(folio))
2077 return folio->_folio_nr_pages;
2079 return 1L << (folio->_flags_1 & 0xff);
2083 /* Only hugetlbfs can allocate folios larger than MAX_ORDER */
2084 #ifdef CONFIG_ARCH_HAS_GIGANTIC_PAGE
2085 #define MAX_FOLIO_NR_PAGES (1UL << PUD_ORDER)
2087 #define MAX_FOLIO_NR_PAGES MAX_ORDER_NR_PAGES
2091 * compound_nr() returns the number of pages in this potentially compound
2092 * page. compound_nr() can be called on a tail page, and is defined to
2093 * return 1 in that case.
2095 static inline unsigned long compound_nr(struct page *page)
2097 struct folio *folio = (struct folio *)page;
2099 if (!test_bit(PG_head, &folio->flags))
2102 return folio->_folio_nr_pages;
2104 return 1L << (folio->_flags_1 & 0xff);
2109 * thp_nr_pages - The number of regular pages in this huge page.
2110 * @page: The head page of a huge page.
2112 static inline int thp_nr_pages(struct page *page)
2114 return folio_nr_pages((struct folio *)page);
2118 * folio_next - Move to the next physical folio.
2119 * @folio: The folio we're currently operating on.
2121 * If you have physically contiguous memory which may span more than
2122 * one folio (eg a &struct bio_vec), use this function to move from one
2123 * folio to the next. Do not use it if the memory is only virtually
2124 * contiguous as the folios are almost certainly not adjacent to each
2125 * other. This is the folio equivalent to writing ``page++``.
2127 * Context: We assume that the folios are refcounted and/or locked at a
2128 * higher level and do not adjust the reference counts.
2129 * Return: The next struct folio.
2131 static inline struct folio *folio_next(struct folio *folio)
2133 return (struct folio *)folio_page(folio, folio_nr_pages(folio));
2137 * folio_shift - The size of the memory described by this folio.
2138 * @folio: The folio.
2140 * A folio represents a number of bytes which is a power-of-two in size.
2141 * This function tells you which power-of-two the folio is. See also
2142 * folio_size() and folio_order().
2144 * Context: The caller should have a reference on the folio to prevent
2145 * it from being split. It is not necessary for the folio to be locked.
2146 * Return: The base-2 logarithm of the size of this folio.
2148 static inline unsigned int folio_shift(struct folio *folio)
2150 return PAGE_SHIFT + folio_order(folio);
2154 * folio_size - The number of bytes in a folio.
2155 * @folio: The folio.
2157 * Context: The caller should have a reference on the folio to prevent
2158 * it from being split. It is not necessary for the folio to be locked.
2159 * Return: The number of bytes in this folio.
2161 static inline size_t folio_size(struct folio *folio)
2163 return PAGE_SIZE << folio_order(folio);
2167 * folio_estimated_sharers - Estimate the number of sharers of a folio.
2168 * @folio: The folio.
2170 * folio_estimated_sharers() aims to serve as a function to efficiently
2171 * estimate the number of processes sharing a folio. This is done by
2172 * looking at the precise mapcount of the first subpage in the folio, and
2173 * assuming the other subpages are the same. This may not be true for large
2174 * folios. If you want exact mapcounts for exact calculations, look at
2175 * page_mapcount() or folio_total_mapcount().
2177 * Return: The estimated number of processes sharing a folio.
2179 static inline int folio_estimated_sharers(struct folio *folio)
2181 return page_mapcount(folio_page(folio, 0));
2184 #ifndef HAVE_ARCH_MAKE_PAGE_ACCESSIBLE
2185 static inline int arch_make_page_accessible(struct page *page)
2191 #ifndef HAVE_ARCH_MAKE_FOLIO_ACCESSIBLE
2192 static inline int arch_make_folio_accessible(struct folio *folio)
2195 long i, nr = folio_nr_pages(folio);
2197 for (i = 0; i < nr; i++) {
2198 ret = arch_make_page_accessible(folio_page(folio, i));
2208 * Some inline functions in vmstat.h depend on page_zone()
2210 #include <linux/vmstat.h>
2212 #if defined(CONFIG_HIGHMEM) && !defined(WANT_PAGE_VIRTUAL)
2213 #define HASHED_PAGE_VIRTUAL
2216 #if defined(WANT_PAGE_VIRTUAL)
2217 static inline void *page_address(const struct page *page)
2219 return page->virtual;
2221 static inline void set_page_address(struct page *page, void *address)
2223 page->virtual = address;
2225 #define page_address_init() do { } while(0)
2228 #if defined(HASHED_PAGE_VIRTUAL)
2229 void *page_address(const struct page *page);
2230 void set_page_address(struct page *page, void *virtual);
2231 void page_address_init(void);
2234 static __always_inline void *lowmem_page_address(const struct page *page)
2236 return page_to_virt(page);
2239 #if !defined(HASHED_PAGE_VIRTUAL) && !defined(WANT_PAGE_VIRTUAL)
2240 #define page_address(page) lowmem_page_address(page)
2241 #define set_page_address(page, address) do { } while(0)
2242 #define page_address_init() do { } while(0)
2245 static inline void *folio_address(const struct folio *folio)
2247 return page_address(&folio->page);
2250 extern pgoff_t __page_file_index(struct page *page);
2253 * Return the pagecache index of the passed page. Regular pagecache pages
2254 * use ->index whereas swapcache pages use swp_offset(->private)
2256 static inline pgoff_t page_index(struct page *page)
2258 if (unlikely(PageSwapCache(page)))
2259 return __page_file_index(page);
2264 * Return true only if the page has been allocated with
2265 * ALLOC_NO_WATERMARKS and the low watermark was not
2266 * met implying that the system is under some pressure.
2268 static inline bool page_is_pfmemalloc(const struct page *page)
2271 * lru.next has bit 1 set if the page is allocated from the
2272 * pfmemalloc reserves. Callers may simply overwrite it if
2273 * they do not need to preserve that information.
2275 return (uintptr_t)page->lru.next & BIT(1);
2279 * Return true only if the folio has been allocated with
2280 * ALLOC_NO_WATERMARKS and the low watermark was not
2281 * met implying that the system is under some pressure.
2283 static inline bool folio_is_pfmemalloc(const struct folio *folio)
2286 * lru.next has bit 1 set if the page is allocated from the
2287 * pfmemalloc reserves. Callers may simply overwrite it if
2288 * they do not need to preserve that information.
2290 return (uintptr_t)folio->lru.next & BIT(1);
2294 * Only to be called by the page allocator on a freshly allocated
2297 static inline void set_page_pfmemalloc(struct page *page)
2299 page->lru.next = (void *)BIT(1);
2302 static inline void clear_page_pfmemalloc(struct page *page)
2304 page->lru.next = NULL;
2308 * Can be called by the pagefault handler when it gets a VM_FAULT_OOM.
2310 extern void pagefault_out_of_memory(void);
2312 #define offset_in_page(p) ((unsigned long)(p) & ~PAGE_MASK)
2313 #define offset_in_thp(page, p) ((unsigned long)(p) & (thp_size(page) - 1))
2314 #define offset_in_folio(folio, p) ((unsigned long)(p) & (folio_size(folio) - 1))
2317 * Parameter block passed down to zap_pte_range in exceptional cases.
2319 struct zap_details {
2320 struct folio *single_folio; /* Locked folio to be unmapped */
2321 bool even_cows; /* Zap COWed private pages too? */
2322 zap_flags_t zap_flags; /* Extra flags for zapping */
2326 * Whether to drop the pte markers, for example, the uffd-wp information for
2327 * file-backed memory. This should only be specified when we will completely
2328 * drop the page in the mm, either by truncation or unmapping of the vma. By
2329 * default, the flag is not set.
2331 #define ZAP_FLAG_DROP_MARKER ((__force zap_flags_t) BIT(0))
2332 /* Set in unmap_vmas() to indicate a final unmap call. Only used by hugetlb */
2333 #define ZAP_FLAG_UNMAP ((__force zap_flags_t) BIT(1))
2335 #ifdef CONFIG_SCHED_MM_CID
2336 void sched_mm_cid_before_execve(struct task_struct *t);
2337 void sched_mm_cid_after_execve(struct task_struct *t);
2338 void sched_mm_cid_fork(struct task_struct *t);
2339 void sched_mm_cid_exit_signals(struct task_struct *t);
2340 static inline int task_mm_cid(struct task_struct *t)
2345 static inline void sched_mm_cid_before_execve(struct task_struct *t) { }
2346 static inline void sched_mm_cid_after_execve(struct task_struct *t) { }
2347 static inline void sched_mm_cid_fork(struct task_struct *t) { }
2348 static inline void sched_mm_cid_exit_signals(struct task_struct *t) { }
2349 static inline int task_mm_cid(struct task_struct *t)
2352 * Use the processor id as a fall-back when the mm cid feature is
2353 * disabled. This provides functional per-cpu data structure accesses
2354 * in user-space, althrough it won't provide the memory usage benefits.
2356 return raw_smp_processor_id();
2361 extern bool can_do_mlock(void);
2363 static inline bool can_do_mlock(void) { return false; }
2365 extern int user_shm_lock(size_t, struct ucounts *);
2366 extern void user_shm_unlock(size_t, struct ucounts *);
2368 struct folio *vm_normal_folio(struct vm_area_struct *vma, unsigned long addr,
2370 struct page *vm_normal_page(struct vm_area_struct *vma, unsigned long addr,
2372 struct folio *vm_normal_folio_pmd(struct vm_area_struct *vma,
2373 unsigned long addr, pmd_t pmd);
2374 struct page *vm_normal_page_pmd(struct vm_area_struct *vma, unsigned long addr,
2377 void zap_vma_ptes(struct vm_area_struct *vma, unsigned long address,
2378 unsigned long size);
2379 void zap_page_range_single(struct vm_area_struct *vma, unsigned long address,
2380 unsigned long size, struct zap_details *details);
2381 static inline void zap_vma_pages(struct vm_area_struct *vma)
2383 zap_page_range_single(vma, vma->vm_start,
2384 vma->vm_end - vma->vm_start, NULL);
2386 void unmap_vmas(struct mmu_gather *tlb, struct ma_state *mas,
2387 struct vm_area_struct *start_vma, unsigned long start,
2388 unsigned long end, unsigned long tree_end, bool mm_wr_locked);
2390 struct mmu_notifier_range;
2392 void free_pgd_range(struct mmu_gather *tlb, unsigned long addr,
2393 unsigned long end, unsigned long floor, unsigned long ceiling);
2395 copy_page_range(struct vm_area_struct *dst_vma, struct vm_area_struct *src_vma);
2396 int follow_pte(struct mm_struct *mm, unsigned long address,
2397 pte_t **ptepp, spinlock_t **ptlp);
2398 int follow_pfn(struct vm_area_struct *vma, unsigned long address,
2399 unsigned long *pfn);
2400 int follow_phys(struct vm_area_struct *vma, unsigned long address,
2401 unsigned int flags, unsigned long *prot, resource_size_t *phys);
2402 int generic_access_phys(struct vm_area_struct *vma, unsigned long addr,
2403 void *buf, int len, int write);
2405 extern void truncate_pagecache(struct inode *inode, loff_t new);
2406 extern void truncate_setsize(struct inode *inode, loff_t newsize);
2407 void pagecache_isize_extended(struct inode *inode, loff_t from, loff_t to);
2408 void truncate_pagecache_range(struct inode *inode, loff_t offset, loff_t end);
2409 int generic_error_remove_folio(struct address_space *mapping,
2410 struct folio *folio);
2412 struct vm_area_struct *lock_mm_and_find_vma(struct mm_struct *mm,
2413 unsigned long address, struct pt_regs *regs);
2416 extern vm_fault_t handle_mm_fault(struct vm_area_struct *vma,
2417 unsigned long address, unsigned int flags,
2418 struct pt_regs *regs);
2419 extern int fixup_user_fault(struct mm_struct *mm,
2420 unsigned long address, unsigned int fault_flags,
2422 void unmap_mapping_pages(struct address_space *mapping,
2423 pgoff_t start, pgoff_t nr, bool even_cows);
2424 void unmap_mapping_range(struct address_space *mapping,
2425 loff_t const holebegin, loff_t const holelen, int even_cows);
2427 static inline vm_fault_t handle_mm_fault(struct vm_area_struct *vma,
2428 unsigned long address, unsigned int flags,
2429 struct pt_regs *regs)
2431 /* should never happen if there's no MMU */
2433 return VM_FAULT_SIGBUS;
2435 static inline int fixup_user_fault(struct mm_struct *mm, unsigned long address,
2436 unsigned int fault_flags, bool *unlocked)
2438 /* should never happen if there's no MMU */
2442 static inline void unmap_mapping_pages(struct address_space *mapping,
2443 pgoff_t start, pgoff_t nr, bool even_cows) { }
2444 static inline void unmap_mapping_range(struct address_space *mapping,
2445 loff_t const holebegin, loff_t const holelen, int even_cows) { }
2448 static inline void unmap_shared_mapping_range(struct address_space *mapping,
2449 loff_t const holebegin, loff_t const holelen)
2451 unmap_mapping_range(mapping, holebegin, holelen, 0);
2454 static inline struct vm_area_struct *vma_lookup(struct mm_struct *mm,
2455 unsigned long addr);
2457 extern int access_process_vm(struct task_struct *tsk, unsigned long addr,
2458 void *buf, int len, unsigned int gup_flags);
2459 extern int access_remote_vm(struct mm_struct *mm, unsigned long addr,
2460 void *buf, int len, unsigned int gup_flags);
2462 long get_user_pages_remote(struct mm_struct *mm,
2463 unsigned long start, unsigned long nr_pages,
2464 unsigned int gup_flags, struct page **pages,
2466 long pin_user_pages_remote(struct mm_struct *mm,
2467 unsigned long start, unsigned long nr_pages,
2468 unsigned int gup_flags, struct page **pages,
2472 * Retrieves a single page alongside its VMA. Does not support FOLL_NOWAIT.
2474 static inline struct page *get_user_page_vma_remote(struct mm_struct *mm,
2477 struct vm_area_struct **vmap)
2480 struct vm_area_struct *vma;
2483 if (WARN_ON_ONCE(unlikely(gup_flags & FOLL_NOWAIT)))
2484 return ERR_PTR(-EINVAL);
2486 got = get_user_pages_remote(mm, addr, 1, gup_flags, &page, NULL);
2489 return ERR_PTR(got);
2491 vma = vma_lookup(mm, addr);
2492 if (WARN_ON_ONCE(!vma)) {
2494 return ERR_PTR(-EINVAL);
2501 long get_user_pages(unsigned long start, unsigned long nr_pages,
2502 unsigned int gup_flags, struct page **pages);
2503 long pin_user_pages(unsigned long start, unsigned long nr_pages,
2504 unsigned int gup_flags, struct page **pages);
2505 long get_user_pages_unlocked(unsigned long start, unsigned long nr_pages,
2506 struct page **pages, unsigned int gup_flags);
2507 long pin_user_pages_unlocked(unsigned long start, unsigned long nr_pages,
2508 struct page **pages, unsigned int gup_flags);
2510 int get_user_pages_fast(unsigned long start, int nr_pages,
2511 unsigned int gup_flags, struct page **pages);
2512 int pin_user_pages_fast(unsigned long start, int nr_pages,
2513 unsigned int gup_flags, struct page **pages);
2514 void folio_add_pin(struct folio *folio);
2516 int account_locked_vm(struct mm_struct *mm, unsigned long pages, bool inc);
2517 int __account_locked_vm(struct mm_struct *mm, unsigned long pages, bool inc,
2518 struct task_struct *task, bool bypass_rlim);
2521 struct page *get_dump_page(unsigned long addr);
2523 bool folio_mark_dirty(struct folio *folio);
2524 bool set_page_dirty(struct page *page);
2525 int set_page_dirty_lock(struct page *page);
2527 int get_cmdline(struct task_struct *task, char *buffer, int buflen);
2529 extern unsigned long move_page_tables(struct vm_area_struct *vma,
2530 unsigned long old_addr, struct vm_area_struct *new_vma,
2531 unsigned long new_addr, unsigned long len,
2532 bool need_rmap_locks, bool for_stack);
2535 * Flags used by change_protection(). For now we make it a bitmap so
2536 * that we can pass in multiple flags just like parameters. However
2537 * for now all the callers are only use one of the flags at the same
2541 * Whether we should manually check if we can map individual PTEs writable,
2542 * because something (e.g., COW, uffd-wp) blocks that from happening for all
2543 * PTEs automatically in a writable mapping.
2545 #define MM_CP_TRY_CHANGE_WRITABLE (1UL << 0)
2546 /* Whether this protection change is for NUMA hints */
2547 #define MM_CP_PROT_NUMA (1UL << 1)
2548 /* Whether this change is for write protecting */
2549 #define MM_CP_UFFD_WP (1UL << 2) /* do wp */
2550 #define MM_CP_UFFD_WP_RESOLVE (1UL << 3) /* Resolve wp */
2551 #define MM_CP_UFFD_WP_ALL (MM_CP_UFFD_WP | \
2552 MM_CP_UFFD_WP_RESOLVE)
2554 bool vma_needs_dirty_tracking(struct vm_area_struct *vma);
2555 int vma_wants_writenotify(struct vm_area_struct *vma, pgprot_t vm_page_prot);
2556 static inline bool vma_wants_manual_pte_write_upgrade(struct vm_area_struct *vma)
2559 * We want to check manually if we can change individual PTEs writable
2560 * if we can't do that automatically for all PTEs in a mapping. For
2561 * private mappings, that's always the case when we have write
2562 * permissions as we properly have to handle COW.
2564 if (vma->vm_flags & VM_SHARED)
2565 return vma_wants_writenotify(vma, vma->vm_page_prot);
2566 return !!(vma->vm_flags & VM_WRITE);
2569 bool can_change_pte_writable(struct vm_area_struct *vma, unsigned long addr,
2571 extern long change_protection(struct mmu_gather *tlb,
2572 struct vm_area_struct *vma, unsigned long start,
2573 unsigned long end, unsigned long cp_flags);
2574 extern int mprotect_fixup(struct vma_iterator *vmi, struct mmu_gather *tlb,
2575 struct vm_area_struct *vma, struct vm_area_struct **pprev,
2576 unsigned long start, unsigned long end, unsigned long newflags);
2579 * doesn't attempt to fault and will return short.
2581 int get_user_pages_fast_only(unsigned long start, int nr_pages,
2582 unsigned int gup_flags, struct page **pages);
2584 static inline bool get_user_page_fast_only(unsigned long addr,
2585 unsigned int gup_flags, struct page **pagep)
2587 return get_user_pages_fast_only(addr, 1, gup_flags, pagep) == 1;
2590 * per-process(per-mm_struct) statistics.
2592 static inline unsigned long get_mm_counter(struct mm_struct *mm, int member)
2594 return percpu_counter_read_positive(&mm->rss_stat[member]);
2597 void mm_trace_rss_stat(struct mm_struct *mm, int member);
2599 static inline void add_mm_counter(struct mm_struct *mm, int member, long value)
2601 percpu_counter_add(&mm->rss_stat[member], value);
2603 mm_trace_rss_stat(mm, member);
2606 static inline void inc_mm_counter(struct mm_struct *mm, int member)
2608 percpu_counter_inc(&mm->rss_stat[member]);
2610 mm_trace_rss_stat(mm, member);
2613 static inline void dec_mm_counter(struct mm_struct *mm, int member)
2615 percpu_counter_dec(&mm->rss_stat[member]);
2617 mm_trace_rss_stat(mm, member);
2620 /* Optimized variant when folio is already known not to be anon */
2621 static inline int mm_counter_file(struct folio *folio)
2623 if (folio_test_swapbacked(folio))
2624 return MM_SHMEMPAGES;
2625 return MM_FILEPAGES;
2628 static inline int mm_counter(struct folio *folio)
2630 if (folio_test_anon(folio))
2631 return MM_ANONPAGES;
2632 return mm_counter_file(folio);
2635 static inline unsigned long get_mm_rss(struct mm_struct *mm)
2637 return get_mm_counter(mm, MM_FILEPAGES) +
2638 get_mm_counter(mm, MM_ANONPAGES) +
2639 get_mm_counter(mm, MM_SHMEMPAGES);
2642 static inline unsigned long get_mm_hiwater_rss(struct mm_struct *mm)
2644 return max(mm->hiwater_rss, get_mm_rss(mm));
2647 static inline unsigned long get_mm_hiwater_vm(struct mm_struct *mm)
2649 return max(mm->hiwater_vm, mm->total_vm);
2652 static inline void update_hiwater_rss(struct mm_struct *mm)
2654 unsigned long _rss = get_mm_rss(mm);
2656 if ((mm)->hiwater_rss < _rss)
2657 (mm)->hiwater_rss = _rss;
2660 static inline void update_hiwater_vm(struct mm_struct *mm)
2662 if (mm->hiwater_vm < mm->total_vm)
2663 mm->hiwater_vm = mm->total_vm;
2666 static inline void reset_mm_hiwater_rss(struct mm_struct *mm)
2668 mm->hiwater_rss = get_mm_rss(mm);
2671 static inline void setmax_mm_hiwater_rss(unsigned long *maxrss,
2672 struct mm_struct *mm)
2674 unsigned long hiwater_rss = get_mm_hiwater_rss(mm);
2676 if (*maxrss < hiwater_rss)
2677 *maxrss = hiwater_rss;
2680 #ifndef CONFIG_ARCH_HAS_PTE_SPECIAL
2681 static inline int pte_special(pte_t pte)
2686 static inline pte_t pte_mkspecial(pte_t pte)
2692 #ifndef CONFIG_ARCH_HAS_PTE_DEVMAP
2693 static inline int pte_devmap(pte_t pte)
2699 extern pte_t *__get_locked_pte(struct mm_struct *mm, unsigned long addr,
2701 static inline pte_t *get_locked_pte(struct mm_struct *mm, unsigned long addr,
2705 __cond_lock(*ptl, ptep = __get_locked_pte(mm, addr, ptl));
2709 #ifdef __PAGETABLE_P4D_FOLDED
2710 static inline int __p4d_alloc(struct mm_struct *mm, pgd_t *pgd,
2711 unsigned long address)
2716 int __p4d_alloc(struct mm_struct *mm, pgd_t *pgd, unsigned long address);
2719 #if defined(__PAGETABLE_PUD_FOLDED) || !defined(CONFIG_MMU)
2720 static inline int __pud_alloc(struct mm_struct *mm, p4d_t *p4d,
2721 unsigned long address)
2725 static inline void mm_inc_nr_puds(struct mm_struct *mm) {}
2726 static inline void mm_dec_nr_puds(struct mm_struct *mm) {}
2729 int __pud_alloc(struct mm_struct *mm, p4d_t *p4d, unsigned long address);
2731 static inline void mm_inc_nr_puds(struct mm_struct *mm)
2733 if (mm_pud_folded(mm))
2735 atomic_long_add(PTRS_PER_PUD * sizeof(pud_t), &mm->pgtables_bytes);
2738 static inline void mm_dec_nr_puds(struct mm_struct *mm)
2740 if (mm_pud_folded(mm))
2742 atomic_long_sub(PTRS_PER_PUD * sizeof(pud_t), &mm->pgtables_bytes);
2746 #if defined(__PAGETABLE_PMD_FOLDED) || !defined(CONFIG_MMU)
2747 static inline int __pmd_alloc(struct mm_struct *mm, pud_t *pud,
2748 unsigned long address)
2753 static inline void mm_inc_nr_pmds(struct mm_struct *mm) {}
2754 static inline void mm_dec_nr_pmds(struct mm_struct *mm) {}
2757 int __pmd_alloc(struct mm_struct *mm, pud_t *pud, unsigned long address);
2759 static inline void mm_inc_nr_pmds(struct mm_struct *mm)
2761 if (mm_pmd_folded(mm))
2763 atomic_long_add(PTRS_PER_PMD * sizeof(pmd_t), &mm->pgtables_bytes);
2766 static inline void mm_dec_nr_pmds(struct mm_struct *mm)
2768 if (mm_pmd_folded(mm))
2770 atomic_long_sub(PTRS_PER_PMD * sizeof(pmd_t), &mm->pgtables_bytes);
2775 static inline void mm_pgtables_bytes_init(struct mm_struct *mm)
2777 atomic_long_set(&mm->pgtables_bytes, 0);
2780 static inline unsigned long mm_pgtables_bytes(const struct mm_struct *mm)
2782 return atomic_long_read(&mm->pgtables_bytes);
2785 static inline void mm_inc_nr_ptes(struct mm_struct *mm)
2787 atomic_long_add(PTRS_PER_PTE * sizeof(pte_t), &mm->pgtables_bytes);
2790 static inline void mm_dec_nr_ptes(struct mm_struct *mm)
2792 atomic_long_sub(PTRS_PER_PTE * sizeof(pte_t), &mm->pgtables_bytes);
2796 static inline void mm_pgtables_bytes_init(struct mm_struct *mm) {}
2797 static inline unsigned long mm_pgtables_bytes(const struct mm_struct *mm)
2802 static inline void mm_inc_nr_ptes(struct mm_struct *mm) {}
2803 static inline void mm_dec_nr_ptes(struct mm_struct *mm) {}
2806 int __pte_alloc(struct mm_struct *mm, pmd_t *pmd);
2807 int __pte_alloc_kernel(pmd_t *pmd);
2809 #if defined(CONFIG_MMU)
2811 static inline p4d_t *p4d_alloc(struct mm_struct *mm, pgd_t *pgd,
2812 unsigned long address)
2814 return (unlikely(pgd_none(*pgd)) && __p4d_alloc(mm, pgd, address)) ?
2815 NULL : p4d_offset(pgd, address);
2818 static inline pud_t *pud_alloc(struct mm_struct *mm, p4d_t *p4d,
2819 unsigned long address)
2821 return (unlikely(p4d_none(*p4d)) && __pud_alloc(mm, p4d, address)) ?
2822 NULL : pud_offset(p4d, address);
2825 static inline pmd_t *pmd_alloc(struct mm_struct *mm, pud_t *pud, unsigned long address)
2827 return (unlikely(pud_none(*pud)) && __pmd_alloc(mm, pud, address))?
2828 NULL: pmd_offset(pud, address);
2830 #endif /* CONFIG_MMU */
2832 static inline struct ptdesc *virt_to_ptdesc(const void *x)
2834 return page_ptdesc(virt_to_page(x));
2837 static inline void *ptdesc_to_virt(const struct ptdesc *pt)
2839 return page_to_virt(ptdesc_page(pt));
2842 static inline void *ptdesc_address(const struct ptdesc *pt)
2844 return folio_address(ptdesc_folio(pt));
2847 static inline bool pagetable_is_reserved(struct ptdesc *pt)
2849 return folio_test_reserved(ptdesc_folio(pt));
2853 * pagetable_alloc - Allocate pagetables
2855 * @order: desired pagetable order
2857 * pagetable_alloc allocates memory for page tables as well as a page table
2858 * descriptor to describe that memory.
2860 * Return: The ptdesc describing the allocated page tables.
2862 static inline struct ptdesc *pagetable_alloc(gfp_t gfp, unsigned int order)
2864 struct page *page = alloc_pages(gfp | __GFP_COMP, order);
2866 return page_ptdesc(page);
2870 * pagetable_free - Free pagetables
2871 * @pt: The page table descriptor
2873 * pagetable_free frees the memory of all page tables described by a page
2874 * table descriptor and the memory for the descriptor itself.
2876 static inline void pagetable_free(struct ptdesc *pt)
2878 struct page *page = ptdesc_page(pt);
2880 __free_pages(page, compound_order(page));
2883 #if USE_SPLIT_PTE_PTLOCKS
2884 #if ALLOC_SPLIT_PTLOCKS
2885 void __init ptlock_cache_init(void);
2886 bool ptlock_alloc(struct ptdesc *ptdesc);
2887 void ptlock_free(struct ptdesc *ptdesc);
2889 static inline spinlock_t *ptlock_ptr(struct ptdesc *ptdesc)
2893 #else /* ALLOC_SPLIT_PTLOCKS */
2894 static inline void ptlock_cache_init(void)
2898 static inline bool ptlock_alloc(struct ptdesc *ptdesc)
2903 static inline void ptlock_free(struct ptdesc *ptdesc)
2907 static inline spinlock_t *ptlock_ptr(struct ptdesc *ptdesc)
2909 return &ptdesc->ptl;
2911 #endif /* ALLOC_SPLIT_PTLOCKS */
2913 static inline spinlock_t *pte_lockptr(struct mm_struct *mm, pmd_t *pmd)
2915 return ptlock_ptr(page_ptdesc(pmd_page(*pmd)));
2918 static inline bool ptlock_init(struct ptdesc *ptdesc)
2921 * prep_new_page() initialize page->private (and therefore page->ptl)
2922 * with 0. Make sure nobody took it in use in between.
2924 * It can happen if arch try to use slab for page table allocation:
2925 * slab code uses page->slab_cache, which share storage with page->ptl.
2927 VM_BUG_ON_PAGE(*(unsigned long *)&ptdesc->ptl, ptdesc_page(ptdesc));
2928 if (!ptlock_alloc(ptdesc))
2930 spin_lock_init(ptlock_ptr(ptdesc));
2934 #else /* !USE_SPLIT_PTE_PTLOCKS */
2936 * We use mm->page_table_lock to guard all pagetable pages of the mm.
2938 static inline spinlock_t *pte_lockptr(struct mm_struct *mm, pmd_t *pmd)
2940 return &mm->page_table_lock;
2942 static inline void ptlock_cache_init(void) {}
2943 static inline bool ptlock_init(struct ptdesc *ptdesc) { return true; }
2944 static inline void ptlock_free(struct ptdesc *ptdesc) {}
2945 #endif /* USE_SPLIT_PTE_PTLOCKS */
2947 static inline bool pagetable_pte_ctor(struct ptdesc *ptdesc)
2949 struct folio *folio = ptdesc_folio(ptdesc);
2951 if (!ptlock_init(ptdesc))
2953 __folio_set_pgtable(folio);
2954 lruvec_stat_add_folio(folio, NR_PAGETABLE);
2958 static inline void pagetable_pte_dtor(struct ptdesc *ptdesc)
2960 struct folio *folio = ptdesc_folio(ptdesc);
2962 ptlock_free(ptdesc);
2963 __folio_clear_pgtable(folio);
2964 lruvec_stat_sub_folio(folio, NR_PAGETABLE);
2967 pte_t *__pte_offset_map(pmd_t *pmd, unsigned long addr, pmd_t *pmdvalp);
2968 static inline pte_t *pte_offset_map(pmd_t *pmd, unsigned long addr)
2970 return __pte_offset_map(pmd, addr, NULL);
2973 pte_t *__pte_offset_map_lock(struct mm_struct *mm, pmd_t *pmd,
2974 unsigned long addr, spinlock_t **ptlp);
2975 static inline pte_t *pte_offset_map_lock(struct mm_struct *mm, pmd_t *pmd,
2976 unsigned long addr, spinlock_t **ptlp)
2980 __cond_lock(*ptlp, pte = __pte_offset_map_lock(mm, pmd, addr, ptlp));
2984 pte_t *pte_offset_map_nolock(struct mm_struct *mm, pmd_t *pmd,
2985 unsigned long addr, spinlock_t **ptlp);
2987 #define pte_unmap_unlock(pte, ptl) do { \
2992 #define pte_alloc(mm, pmd) (unlikely(pmd_none(*(pmd))) && __pte_alloc(mm, pmd))
2994 #define pte_alloc_map(mm, pmd, address) \
2995 (pte_alloc(mm, pmd) ? NULL : pte_offset_map(pmd, address))
2997 #define pte_alloc_map_lock(mm, pmd, address, ptlp) \
2998 (pte_alloc(mm, pmd) ? \
2999 NULL : pte_offset_map_lock(mm, pmd, address, ptlp))
3001 #define pte_alloc_kernel(pmd, address) \
3002 ((unlikely(pmd_none(*(pmd))) && __pte_alloc_kernel(pmd))? \
3003 NULL: pte_offset_kernel(pmd, address))
3005 #if USE_SPLIT_PMD_PTLOCKS
3007 static inline struct page *pmd_pgtable_page(pmd_t *pmd)
3009 unsigned long mask = ~(PTRS_PER_PMD * sizeof(pmd_t) - 1);
3010 return virt_to_page((void *)((unsigned long) pmd & mask));
3013 static inline struct ptdesc *pmd_ptdesc(pmd_t *pmd)
3015 return page_ptdesc(pmd_pgtable_page(pmd));
3018 static inline spinlock_t *pmd_lockptr(struct mm_struct *mm, pmd_t *pmd)
3020 return ptlock_ptr(pmd_ptdesc(pmd));
3023 static inline bool pmd_ptlock_init(struct ptdesc *ptdesc)
3025 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
3026 ptdesc->pmd_huge_pte = NULL;
3028 return ptlock_init(ptdesc);
3031 static inline void pmd_ptlock_free(struct ptdesc *ptdesc)
3033 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
3034 VM_BUG_ON_PAGE(ptdesc->pmd_huge_pte, ptdesc_page(ptdesc));
3036 ptlock_free(ptdesc);
3039 #define pmd_huge_pte(mm, pmd) (pmd_ptdesc(pmd)->pmd_huge_pte)
3043 static inline spinlock_t *pmd_lockptr(struct mm_struct *mm, pmd_t *pmd)
3045 return &mm->page_table_lock;
3048 static inline bool pmd_ptlock_init(struct ptdesc *ptdesc) { return true; }
3049 static inline void pmd_ptlock_free(struct ptdesc *ptdesc) {}
3051 #define pmd_huge_pte(mm, pmd) ((mm)->pmd_huge_pte)
3055 static inline spinlock_t *pmd_lock(struct mm_struct *mm, pmd_t *pmd)
3057 spinlock_t *ptl = pmd_lockptr(mm, pmd);
3062 static inline bool pagetable_pmd_ctor(struct ptdesc *ptdesc)
3064 struct folio *folio = ptdesc_folio(ptdesc);
3066 if (!pmd_ptlock_init(ptdesc))
3068 __folio_set_pgtable(folio);
3069 lruvec_stat_add_folio(folio, NR_PAGETABLE);
3073 static inline void pagetable_pmd_dtor(struct ptdesc *ptdesc)
3075 struct folio *folio = ptdesc_folio(ptdesc);
3077 pmd_ptlock_free(ptdesc);
3078 __folio_clear_pgtable(folio);
3079 lruvec_stat_sub_folio(folio, NR_PAGETABLE);
3083 * No scalability reason to split PUD locks yet, but follow the same pattern
3084 * as the PMD locks to make it easier if we decide to. The VM should not be
3085 * considered ready to switch to split PUD locks yet; there may be places
3086 * which need to be converted from page_table_lock.
3088 static inline spinlock_t *pud_lockptr(struct mm_struct *mm, pud_t *pud)
3090 return &mm->page_table_lock;
3093 static inline spinlock_t *pud_lock(struct mm_struct *mm, pud_t *pud)
3095 spinlock_t *ptl = pud_lockptr(mm, pud);
3101 static inline void pagetable_pud_ctor(struct ptdesc *ptdesc)
3103 struct folio *folio = ptdesc_folio(ptdesc);
3105 __folio_set_pgtable(folio);
3106 lruvec_stat_add_folio(folio, NR_PAGETABLE);
3109 static inline void pagetable_pud_dtor(struct ptdesc *ptdesc)
3111 struct folio *folio = ptdesc_folio(ptdesc);
3113 __folio_clear_pgtable(folio);
3114 lruvec_stat_sub_folio(folio, NR_PAGETABLE);
3117 extern void __init pagecache_init(void);
3118 extern void free_initmem(void);
3121 * Free reserved pages within range [PAGE_ALIGN(start), end & PAGE_MASK)
3122 * into the buddy system. The freed pages will be poisoned with pattern
3123 * "poison" if it's within range [0, UCHAR_MAX].
3124 * Return pages freed into the buddy system.
3126 extern unsigned long free_reserved_area(void *start, void *end,
3127 int poison, const char *s);
3129 extern void adjust_managed_page_count(struct page *page, long count);
3131 extern void reserve_bootmem_region(phys_addr_t start,
3132 phys_addr_t end, int nid);
3134 /* Free the reserved page into the buddy system, so it gets managed. */
3135 static inline void free_reserved_page(struct page *page)
3137 ClearPageReserved(page);
3138 init_page_count(page);
3140 adjust_managed_page_count(page, 1);
3142 #define free_highmem_page(page) free_reserved_page(page)
3144 static inline void mark_page_reserved(struct page *page)
3146 SetPageReserved(page);
3147 adjust_managed_page_count(page, -1);
3150 static inline void free_reserved_ptdesc(struct ptdesc *pt)
3152 free_reserved_page(ptdesc_page(pt));
3156 * Default method to free all the __init memory into the buddy system.
3157 * The freed pages will be poisoned with pattern "poison" if it's within
3158 * range [0, UCHAR_MAX].
3159 * Return pages freed into the buddy system.
3161 static inline unsigned long free_initmem_default(int poison)
3163 extern char __init_begin[], __init_end[];
3165 return free_reserved_area(&__init_begin, &__init_end,
3166 poison, "unused kernel image (initmem)");
3169 static inline unsigned long get_num_physpages(void)
3172 unsigned long phys_pages = 0;
3174 for_each_online_node(nid)
3175 phys_pages += node_present_pages(nid);
3181 * Using memblock node mappings, an architecture may initialise its
3182 * zones, allocate the backing mem_map and account for memory holes in an
3183 * architecture independent manner.
3185 * An architecture is expected to register range of page frames backed by
3186 * physical memory with memblock_add[_node]() before calling
3187 * free_area_init() passing in the PFN each zone ends at. At a basic
3188 * usage, an architecture is expected to do something like
3190 * unsigned long max_zone_pfns[MAX_NR_ZONES] = {max_dma, max_normal_pfn,
3192 * for_each_valid_physical_page_range()
3193 * memblock_add_node(base, size, nid, MEMBLOCK_NONE)
3194 * free_area_init(max_zone_pfns);
3196 void free_area_init(unsigned long *max_zone_pfn);
3197 unsigned long node_map_pfn_alignment(void);
3198 unsigned long __absent_pages_in_range(int nid, unsigned long start_pfn,
3199 unsigned long end_pfn);
3200 extern unsigned long absent_pages_in_range(unsigned long start_pfn,
3201 unsigned long end_pfn);
3202 extern void get_pfn_range_for_nid(unsigned int nid,
3203 unsigned long *start_pfn, unsigned long *end_pfn);
3206 static inline int early_pfn_to_nid(unsigned long pfn)
3211 /* please see mm/page_alloc.c */
3212 extern int __meminit early_pfn_to_nid(unsigned long pfn);
3215 extern void set_dma_reserve(unsigned long new_dma_reserve);
3216 extern void mem_init(void);
3217 extern void __init mmap_init(void);
3219 extern void __show_mem(unsigned int flags, nodemask_t *nodemask, int max_zone_idx);
3220 static inline void show_mem(void)
3222 __show_mem(0, NULL, MAX_NR_ZONES - 1);
3224 extern long si_mem_available(void);
3225 extern void si_meminfo(struct sysinfo * val);
3226 extern void si_meminfo_node(struct sysinfo *val, int nid);
3227 #ifdef __HAVE_ARCH_RESERVED_KERNEL_PAGES
3228 extern unsigned long arch_reserved_kernel_pages(void);
3231 extern __printf(3, 4)
3232 void warn_alloc(gfp_t gfp_mask, nodemask_t *nodemask, const char *fmt, ...);
3234 extern void setup_per_cpu_pageset(void);
3237 extern atomic_long_t mmap_pages_allocated;
3238 extern int nommu_shrink_inode_mappings(struct inode *, size_t, size_t);
3240 /* interval_tree.c */
3241 void vma_interval_tree_insert(struct vm_area_struct *node,
3242 struct rb_root_cached *root);
3243 void vma_interval_tree_insert_after(struct vm_area_struct *node,
3244 struct vm_area_struct *prev,
3245 struct rb_root_cached *root);
3246 void vma_interval_tree_remove(struct vm_area_struct *node,
3247 struct rb_root_cached *root);
3248 struct vm_area_struct *vma_interval_tree_iter_first(struct rb_root_cached *root,
3249 unsigned long start, unsigned long last);
3250 struct vm_area_struct *vma_interval_tree_iter_next(struct vm_area_struct *node,
3251 unsigned long start, unsigned long last);
3253 #define vma_interval_tree_foreach(vma, root, start, last) \
3254 for (vma = vma_interval_tree_iter_first(root, start, last); \
3255 vma; vma = vma_interval_tree_iter_next(vma, start, last))
3257 void anon_vma_interval_tree_insert(struct anon_vma_chain *node,
3258 struct rb_root_cached *root);
3259 void anon_vma_interval_tree_remove(struct anon_vma_chain *node,
3260 struct rb_root_cached *root);
3261 struct anon_vma_chain *
3262 anon_vma_interval_tree_iter_first(struct rb_root_cached *root,
3263 unsigned long start, unsigned long last);
3264 struct anon_vma_chain *anon_vma_interval_tree_iter_next(
3265 struct anon_vma_chain *node, unsigned long start, unsigned long last);
3266 #ifdef CONFIG_DEBUG_VM_RB
3267 void anon_vma_interval_tree_verify(struct anon_vma_chain *node);
3270 #define anon_vma_interval_tree_foreach(avc, root, start, last) \
3271 for (avc = anon_vma_interval_tree_iter_first(root, start, last); \
3272 avc; avc = anon_vma_interval_tree_iter_next(avc, start, last))
3275 extern int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin);
3276 extern int vma_expand(struct vma_iterator *vmi, struct vm_area_struct *vma,
3277 unsigned long start, unsigned long end, pgoff_t pgoff,
3278 struct vm_area_struct *next);
3279 extern int vma_shrink(struct vma_iterator *vmi, struct vm_area_struct *vma,
3280 unsigned long start, unsigned long end, pgoff_t pgoff);
3281 extern struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *);
3282 extern int insert_vm_struct(struct mm_struct *, struct vm_area_struct *);
3283 extern void unlink_file_vma(struct vm_area_struct *);
3284 extern struct vm_area_struct *copy_vma(struct vm_area_struct **,
3285 unsigned long addr, unsigned long len, pgoff_t pgoff,
3286 bool *need_rmap_locks);
3287 extern void exit_mmap(struct mm_struct *);
3288 struct vm_area_struct *vma_modify(struct vma_iterator *vmi,
3289 struct vm_area_struct *prev,
3290 struct vm_area_struct *vma,
3291 unsigned long start, unsigned long end,
3292 unsigned long vm_flags,
3293 struct mempolicy *policy,
3294 struct vm_userfaultfd_ctx uffd_ctx,
3295 struct anon_vma_name *anon_name);
3297 /* We are about to modify the VMA's flags. */
3298 static inline struct vm_area_struct
3299 *vma_modify_flags(struct vma_iterator *vmi,
3300 struct vm_area_struct *prev,
3301 struct vm_area_struct *vma,
3302 unsigned long start, unsigned long end,
3303 unsigned long new_flags)
3305 return vma_modify(vmi, prev, vma, start, end, new_flags,
3306 vma_policy(vma), vma->vm_userfaultfd_ctx,
3307 anon_vma_name(vma));
3310 /* We are about to modify the VMA's flags and/or anon_name. */
3311 static inline struct vm_area_struct
3312 *vma_modify_flags_name(struct vma_iterator *vmi,
3313 struct vm_area_struct *prev,
3314 struct vm_area_struct *vma,
3315 unsigned long start,
3317 unsigned long new_flags,
3318 struct anon_vma_name *new_name)
3320 return vma_modify(vmi, prev, vma, start, end, new_flags,
3321 vma_policy(vma), vma->vm_userfaultfd_ctx, new_name);
3324 /* We are about to modify the VMA's memory policy. */
3325 static inline struct vm_area_struct
3326 *vma_modify_policy(struct vma_iterator *vmi,
3327 struct vm_area_struct *prev,
3328 struct vm_area_struct *vma,
3329 unsigned long start, unsigned long end,
3330 struct mempolicy *new_pol)
3332 return vma_modify(vmi, prev, vma, start, end, vma->vm_flags,
3333 new_pol, vma->vm_userfaultfd_ctx, anon_vma_name(vma));
3336 /* We are about to modify the VMA's flags and/or uffd context. */
3337 static inline struct vm_area_struct
3338 *vma_modify_flags_uffd(struct vma_iterator *vmi,
3339 struct vm_area_struct *prev,
3340 struct vm_area_struct *vma,
3341 unsigned long start, unsigned long end,
3342 unsigned long new_flags,
3343 struct vm_userfaultfd_ctx new_ctx)
3345 return vma_modify(vmi, prev, vma, start, end, new_flags,
3346 vma_policy(vma), new_ctx, anon_vma_name(vma));
3349 static inline int check_data_rlimit(unsigned long rlim,
3351 unsigned long start,
3352 unsigned long end_data,
3353 unsigned long start_data)
3355 if (rlim < RLIM_INFINITY) {
3356 if (((new - start) + (end_data - start_data)) > rlim)
3363 extern int mm_take_all_locks(struct mm_struct *mm);
3364 extern void mm_drop_all_locks(struct mm_struct *mm);
3366 extern int set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file);
3367 extern int replace_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file);
3368 extern struct file *get_mm_exe_file(struct mm_struct *mm);
3369 extern struct file *get_task_exe_file(struct task_struct *task);
3371 extern bool may_expand_vm(struct mm_struct *, vm_flags_t, unsigned long npages);
3372 extern void vm_stat_account(struct mm_struct *, vm_flags_t, long npages);
3374 extern bool vma_is_special_mapping(const struct vm_area_struct *vma,
3375 const struct vm_special_mapping *sm);
3376 extern struct vm_area_struct *_install_special_mapping(struct mm_struct *mm,
3377 unsigned long addr, unsigned long len,
3378 unsigned long flags,
3379 const struct vm_special_mapping *spec);
3380 /* This is an obsolete alternative to _install_special_mapping. */
3381 extern int install_special_mapping(struct mm_struct *mm,
3382 unsigned long addr, unsigned long len,
3383 unsigned long flags, struct page **pages);
3385 unsigned long randomize_stack_top(unsigned long stack_top);
3386 unsigned long randomize_page(unsigned long start, unsigned long range);
3388 extern unsigned long get_unmapped_area(struct file *, unsigned long, unsigned long, unsigned long, unsigned long);
3390 extern unsigned long mmap_region(struct file *file, unsigned long addr,
3391 unsigned long len, vm_flags_t vm_flags, unsigned long pgoff,
3392 struct list_head *uf);
3393 extern unsigned long do_mmap(struct file *file, unsigned long addr,
3394 unsigned long len, unsigned long prot, unsigned long flags,
3395 vm_flags_t vm_flags, unsigned long pgoff, unsigned long *populate,
3396 struct list_head *uf);
3397 extern int do_vmi_munmap(struct vma_iterator *vmi, struct mm_struct *mm,
3398 unsigned long start, size_t len, struct list_head *uf,
3400 extern int do_munmap(struct mm_struct *, unsigned long, size_t,
3401 struct list_head *uf);
3402 extern int do_madvise(struct mm_struct *mm, unsigned long start, size_t len_in, int behavior);
3405 extern int do_vma_munmap(struct vma_iterator *vmi, struct vm_area_struct *vma,
3406 unsigned long start, unsigned long end,
3407 struct list_head *uf, bool unlock);
3408 extern int __mm_populate(unsigned long addr, unsigned long len,
3410 static inline void mm_populate(unsigned long addr, unsigned long len)
3413 (void) __mm_populate(addr, len, 1);
3416 static inline void mm_populate(unsigned long addr, unsigned long len) {}
3419 /* This takes the mm semaphore itself */
3420 extern int __must_check vm_brk_flags(unsigned long, unsigned long, unsigned long);
3421 extern int vm_munmap(unsigned long, size_t);
3422 extern unsigned long __must_check vm_mmap(struct file *, unsigned long,
3423 unsigned long, unsigned long,
3424 unsigned long, unsigned long);
3426 struct vm_unmapped_area_info {
3427 #define VM_UNMAPPED_AREA_TOPDOWN 1
3428 unsigned long flags;
3429 unsigned long length;
3430 unsigned long low_limit;
3431 unsigned long high_limit;
3432 unsigned long align_mask;
3433 unsigned long align_offset;
3436 extern unsigned long vm_unmapped_area(struct vm_unmapped_area_info *info);
3439 extern void truncate_inode_pages(struct address_space *, loff_t);
3440 extern void truncate_inode_pages_range(struct address_space *,
3441 loff_t lstart, loff_t lend);
3442 extern void truncate_inode_pages_final(struct address_space *);
3444 /* generic vm_area_ops exported for stackable file systems */
3445 extern vm_fault_t filemap_fault(struct vm_fault *vmf);
3446 extern vm_fault_t filemap_map_pages(struct vm_fault *vmf,
3447 pgoff_t start_pgoff, pgoff_t end_pgoff);
3448 extern vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf);
3450 extern unsigned long stack_guard_gap;
3451 /* Generic expand stack which grows the stack according to GROWS{UP,DOWN} */
3452 int expand_stack_locked(struct vm_area_struct *vma, unsigned long address);
3453 struct vm_area_struct *expand_stack(struct mm_struct * mm, unsigned long addr);
3455 /* CONFIG_STACK_GROWSUP still needs to grow downwards at some places */
3456 int expand_downwards(struct vm_area_struct *vma, unsigned long address);
3458 /* Look up the first VMA which satisfies addr < vm_end, NULL if none. */
3459 extern struct vm_area_struct * find_vma(struct mm_struct * mm, unsigned long addr);
3460 extern struct vm_area_struct * find_vma_prev(struct mm_struct * mm, unsigned long addr,
3461 struct vm_area_struct **pprev);
3464 * Look up the first VMA which intersects the interval [start_addr, end_addr)
3465 * NULL if none. Assume start_addr < end_addr.
3467 struct vm_area_struct *find_vma_intersection(struct mm_struct *mm,
3468 unsigned long start_addr, unsigned long end_addr);
3471 * vma_lookup() - Find a VMA at a specific address
3472 * @mm: The process address space.
3473 * @addr: The user address.
3475 * Return: The vm_area_struct at the given address, %NULL otherwise.
3478 struct vm_area_struct *vma_lookup(struct mm_struct *mm, unsigned long addr)
3480 return mtree_load(&mm->mm_mt, addr);
3483 static inline unsigned long stack_guard_start_gap(struct vm_area_struct *vma)
3485 if (vma->vm_flags & VM_GROWSDOWN)
3486 return stack_guard_gap;
3488 /* See reasoning around the VM_SHADOW_STACK definition */
3489 if (vma->vm_flags & VM_SHADOW_STACK)
3495 static inline unsigned long vm_start_gap(struct vm_area_struct *vma)
3497 unsigned long gap = stack_guard_start_gap(vma);
3498 unsigned long vm_start = vma->vm_start;
3501 if (vm_start > vma->vm_start)
3506 static inline unsigned long vm_end_gap(struct vm_area_struct *vma)
3508 unsigned long vm_end = vma->vm_end;
3510 if (vma->vm_flags & VM_GROWSUP) {
3511 vm_end += stack_guard_gap;
3512 if (vm_end < vma->vm_end)
3513 vm_end = -PAGE_SIZE;
3518 static inline unsigned long vma_pages(struct vm_area_struct *vma)
3520 return (vma->vm_end - vma->vm_start) >> PAGE_SHIFT;
3523 /* Look up the first VMA which exactly match the interval vm_start ... vm_end */
3524 static inline struct vm_area_struct *find_exact_vma(struct mm_struct *mm,
3525 unsigned long vm_start, unsigned long vm_end)
3527 struct vm_area_struct *vma = vma_lookup(mm, vm_start);
3529 if (vma && (vma->vm_start != vm_start || vma->vm_end != vm_end))
3535 static inline bool range_in_vma(struct vm_area_struct *vma,
3536 unsigned long start, unsigned long end)
3538 return (vma && vma->vm_start <= start && end <= vma->vm_end);
3542 pgprot_t vm_get_page_prot(unsigned long vm_flags);
3543 void vma_set_page_prot(struct vm_area_struct *vma);
3545 static inline pgprot_t vm_get_page_prot(unsigned long vm_flags)
3549 static inline void vma_set_page_prot(struct vm_area_struct *vma)
3551 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
3555 void vma_set_file(struct vm_area_struct *vma, struct file *file);
3557 #ifdef CONFIG_NUMA_BALANCING
3558 unsigned long change_prot_numa(struct vm_area_struct *vma,
3559 unsigned long start, unsigned long end);
3562 struct vm_area_struct *find_extend_vma_locked(struct mm_struct *,
3563 unsigned long addr);
3564 int remap_pfn_range(struct vm_area_struct *, unsigned long addr,
3565 unsigned long pfn, unsigned long size, pgprot_t);
3566 int remap_pfn_range_notrack(struct vm_area_struct *vma, unsigned long addr,
3567 unsigned long pfn, unsigned long size, pgprot_t prot);
3568 int vm_insert_page(struct vm_area_struct *, unsigned long addr, struct page *);
3569 int vm_insert_pages(struct vm_area_struct *vma, unsigned long addr,
3570 struct page **pages, unsigned long *num);
3571 int vm_map_pages(struct vm_area_struct *vma, struct page **pages,
3573 int vm_map_pages_zero(struct vm_area_struct *vma, struct page **pages,
3575 vm_fault_t vmf_insert_pfn(struct vm_area_struct *vma, unsigned long addr,
3577 vm_fault_t vmf_insert_pfn_prot(struct vm_area_struct *vma, unsigned long addr,
3578 unsigned long pfn, pgprot_t pgprot);
3579 vm_fault_t vmf_insert_mixed(struct vm_area_struct *vma, unsigned long addr,
3581 vm_fault_t vmf_insert_mixed_mkwrite(struct vm_area_struct *vma,
3582 unsigned long addr, pfn_t pfn);
3583 int vm_iomap_memory(struct vm_area_struct *vma, phys_addr_t start, unsigned long len);
3585 static inline vm_fault_t vmf_insert_page(struct vm_area_struct *vma,
3586 unsigned long addr, struct page *page)
3588 int err = vm_insert_page(vma, addr, page);
3591 return VM_FAULT_OOM;
3592 if (err < 0 && err != -EBUSY)
3593 return VM_FAULT_SIGBUS;
3595 return VM_FAULT_NOPAGE;
3598 #ifndef io_remap_pfn_range
3599 static inline int io_remap_pfn_range(struct vm_area_struct *vma,
3600 unsigned long addr, unsigned long pfn,
3601 unsigned long size, pgprot_t prot)
3603 return remap_pfn_range(vma, addr, pfn, size, pgprot_decrypted(prot));
3607 static inline vm_fault_t vmf_error(int err)
3610 return VM_FAULT_OOM;
3611 else if (err == -EHWPOISON)
3612 return VM_FAULT_HWPOISON;
3613 return VM_FAULT_SIGBUS;
3617 * Convert errno to return value for ->page_mkwrite() calls.
3619 * This should eventually be merged with vmf_error() above, but will need a
3620 * careful audit of all vmf_error() callers.
3622 static inline vm_fault_t vmf_fs_error(int err)
3625 return VM_FAULT_LOCKED;
3626 if (err == -EFAULT || err == -EAGAIN)
3627 return VM_FAULT_NOPAGE;
3629 return VM_FAULT_OOM;
3630 /* -ENOSPC, -EDQUOT, -EIO ... */
3631 return VM_FAULT_SIGBUS;
3634 struct page *follow_page(struct vm_area_struct *vma, unsigned long address,
3635 unsigned int foll_flags);
3637 static inline int vm_fault_to_errno(vm_fault_t vm_fault, int foll_flags)
3639 if (vm_fault & VM_FAULT_OOM)
3641 if (vm_fault & (VM_FAULT_HWPOISON | VM_FAULT_HWPOISON_LARGE))
3642 return (foll_flags & FOLL_HWPOISON) ? -EHWPOISON : -EFAULT;
3643 if (vm_fault & (VM_FAULT_SIGBUS | VM_FAULT_SIGSEGV))
3649 * Indicates whether GUP can follow a PROT_NONE mapped page, or whether
3650 * a (NUMA hinting) fault is required.
3652 static inline bool gup_can_follow_protnone(struct vm_area_struct *vma,
3656 * If callers don't want to honor NUMA hinting faults, no need to
3657 * determine if we would actually have to trigger a NUMA hinting fault.
3659 if (!(flags & FOLL_HONOR_NUMA_FAULT))
3663 * NUMA hinting faults don't apply in inaccessible (PROT_NONE) VMAs.
3665 * Requiring a fault here even for inaccessible VMAs would mean that
3666 * FOLL_FORCE cannot make any progress, because handle_mm_fault()
3667 * refuses to process NUMA hinting faults in inaccessible VMAs.
3669 return !vma_is_accessible(vma);
3672 typedef int (*pte_fn_t)(pte_t *pte, unsigned long addr, void *data);
3673 extern int apply_to_page_range(struct mm_struct *mm, unsigned long address,
3674 unsigned long size, pte_fn_t fn, void *data);
3675 extern int apply_to_existing_page_range(struct mm_struct *mm,
3676 unsigned long address, unsigned long size,
3677 pte_fn_t fn, void *data);
3679 #ifdef CONFIG_PAGE_POISONING
3680 extern void __kernel_poison_pages(struct page *page, int numpages);
3681 extern void __kernel_unpoison_pages(struct page *page, int numpages);
3682 extern bool _page_poisoning_enabled_early;
3683 DECLARE_STATIC_KEY_FALSE(_page_poisoning_enabled);
3684 static inline bool page_poisoning_enabled(void)
3686 return _page_poisoning_enabled_early;
3689 * For use in fast paths after init_mem_debugging() has run, or when a
3690 * false negative result is not harmful when called too early.
3692 static inline bool page_poisoning_enabled_static(void)
3694 return static_branch_unlikely(&_page_poisoning_enabled);
3696 static inline void kernel_poison_pages(struct page *page, int numpages)
3698 if (page_poisoning_enabled_static())
3699 __kernel_poison_pages(page, numpages);
3701 static inline void kernel_unpoison_pages(struct page *page, int numpages)
3703 if (page_poisoning_enabled_static())
3704 __kernel_unpoison_pages(page, numpages);
3707 static inline bool page_poisoning_enabled(void) { return false; }
3708 static inline bool page_poisoning_enabled_static(void) { return false; }
3709 static inline void __kernel_poison_pages(struct page *page, int nunmpages) { }
3710 static inline void kernel_poison_pages(struct page *page, int numpages) { }
3711 static inline void kernel_unpoison_pages(struct page *page, int numpages) { }
3714 DECLARE_STATIC_KEY_MAYBE(CONFIG_INIT_ON_ALLOC_DEFAULT_ON, init_on_alloc);
3715 static inline bool want_init_on_alloc(gfp_t flags)
3717 if (static_branch_maybe(CONFIG_INIT_ON_ALLOC_DEFAULT_ON,
3720 return flags & __GFP_ZERO;
3723 DECLARE_STATIC_KEY_MAYBE(CONFIG_INIT_ON_FREE_DEFAULT_ON, init_on_free);
3724 static inline bool want_init_on_free(void)
3726 return static_branch_maybe(CONFIG_INIT_ON_FREE_DEFAULT_ON,
3730 extern bool _debug_pagealloc_enabled_early;
3731 DECLARE_STATIC_KEY_FALSE(_debug_pagealloc_enabled);
3733 static inline bool debug_pagealloc_enabled(void)
3735 return IS_ENABLED(CONFIG_DEBUG_PAGEALLOC) &&
3736 _debug_pagealloc_enabled_early;
3740 * For use in fast paths after mem_debugging_and_hardening_init() has run,
3741 * or when a false negative result is not harmful when called too early.
3743 static inline bool debug_pagealloc_enabled_static(void)
3745 if (!IS_ENABLED(CONFIG_DEBUG_PAGEALLOC))
3748 return static_branch_unlikely(&_debug_pagealloc_enabled);
3752 * To support DEBUG_PAGEALLOC architecture must ensure that
3753 * __kernel_map_pages() never fails
3755 extern void __kernel_map_pages(struct page *page, int numpages, int enable);
3756 #ifdef CONFIG_DEBUG_PAGEALLOC
3757 static inline void debug_pagealloc_map_pages(struct page *page, int numpages)
3759 if (debug_pagealloc_enabled_static())
3760 __kernel_map_pages(page, numpages, 1);
3763 static inline void debug_pagealloc_unmap_pages(struct page *page, int numpages)
3765 if (debug_pagealloc_enabled_static())
3766 __kernel_map_pages(page, numpages, 0);
3769 extern unsigned int _debug_guardpage_minorder;
3770 DECLARE_STATIC_KEY_FALSE(_debug_guardpage_enabled);
3772 static inline unsigned int debug_guardpage_minorder(void)
3774 return _debug_guardpage_minorder;
3777 static inline bool debug_guardpage_enabled(void)
3779 return static_branch_unlikely(&_debug_guardpage_enabled);
3782 static inline bool page_is_guard(struct page *page)
3784 if (!debug_guardpage_enabled())
3787 return PageGuard(page);
3790 bool __set_page_guard(struct zone *zone, struct page *page, unsigned int order,
3792 static inline bool set_page_guard(struct zone *zone, struct page *page,
3793 unsigned int order, int migratetype)
3795 if (!debug_guardpage_enabled())
3797 return __set_page_guard(zone, page, order, migratetype);
3800 void __clear_page_guard(struct zone *zone, struct page *page, unsigned int order,
3802 static inline void clear_page_guard(struct zone *zone, struct page *page,
3803 unsigned int order, int migratetype)
3805 if (!debug_guardpage_enabled())
3807 __clear_page_guard(zone, page, order, migratetype);
3810 #else /* CONFIG_DEBUG_PAGEALLOC */
3811 static inline void debug_pagealloc_map_pages(struct page *page, int numpages) {}
3812 static inline void debug_pagealloc_unmap_pages(struct page *page, int numpages) {}
3813 static inline unsigned int debug_guardpage_minorder(void) { return 0; }
3814 static inline bool debug_guardpage_enabled(void) { return false; }
3815 static inline bool page_is_guard(struct page *page) { return false; }
3816 static inline bool set_page_guard(struct zone *zone, struct page *page,
3817 unsigned int order, int migratetype) { return false; }
3818 static inline void clear_page_guard(struct zone *zone, struct page *page,
3819 unsigned int order, int migratetype) {}
3820 #endif /* CONFIG_DEBUG_PAGEALLOC */
3822 #ifdef __HAVE_ARCH_GATE_AREA
3823 extern struct vm_area_struct *get_gate_vma(struct mm_struct *mm);
3824 extern int in_gate_area_no_mm(unsigned long addr);
3825 extern int in_gate_area(struct mm_struct *mm, unsigned long addr);
3827 static inline struct vm_area_struct *get_gate_vma(struct mm_struct *mm)
3831 static inline int in_gate_area_no_mm(unsigned long addr) { return 0; }
3832 static inline int in_gate_area(struct mm_struct *mm, unsigned long addr)
3836 #endif /* __HAVE_ARCH_GATE_AREA */
3838 extern bool process_shares_mm(struct task_struct *p, struct mm_struct *mm);
3840 #ifdef CONFIG_SYSCTL
3841 extern int sysctl_drop_caches;
3842 int drop_caches_sysctl_handler(struct ctl_table *, int, void *, size_t *,
3846 void drop_slab(void);
3849 #define randomize_va_space 0
3851 extern int randomize_va_space;
3854 const char * arch_vma_name(struct vm_area_struct *vma);
3856 void print_vma_addr(char *prefix, unsigned long rip);
3858 static inline void print_vma_addr(char *prefix, unsigned long rip)
3863 void *sparse_buffer_alloc(unsigned long size);
3864 struct page * __populate_section_memmap(unsigned long pfn,
3865 unsigned long nr_pages, int nid, struct vmem_altmap *altmap,
3866 struct dev_pagemap *pgmap);
3867 void pmd_init(void *addr);
3868 void pud_init(void *addr);
3869 pgd_t *vmemmap_pgd_populate(unsigned long addr, int node);
3870 p4d_t *vmemmap_p4d_populate(pgd_t *pgd, unsigned long addr, int node);
3871 pud_t *vmemmap_pud_populate(p4d_t *p4d, unsigned long addr, int node);
3872 pmd_t *vmemmap_pmd_populate(pud_t *pud, unsigned long addr, int node);
3873 pte_t *vmemmap_pte_populate(pmd_t *pmd, unsigned long addr, int node,
3874 struct vmem_altmap *altmap, struct page *reuse);
3875 void *vmemmap_alloc_block(unsigned long size, int node);
3877 void *vmemmap_alloc_block_buf(unsigned long size, int node,
3878 struct vmem_altmap *altmap);
3879 void vmemmap_verify(pte_t *, int, unsigned long, unsigned long);
3880 void vmemmap_set_pmd(pmd_t *pmd, void *p, int node,
3881 unsigned long addr, unsigned long next);
3882 int vmemmap_check_pmd(pmd_t *pmd, int node,
3883 unsigned long addr, unsigned long next);
3884 int vmemmap_populate_basepages(unsigned long start, unsigned long end,
3885 int node, struct vmem_altmap *altmap);
3886 int vmemmap_populate_hugepages(unsigned long start, unsigned long end,
3887 int node, struct vmem_altmap *altmap);
3888 int vmemmap_populate(unsigned long start, unsigned long end, int node,
3889 struct vmem_altmap *altmap);
3890 void vmemmap_populate_print_last(void);
3891 #ifdef CONFIG_MEMORY_HOTPLUG
3892 void vmemmap_free(unsigned long start, unsigned long end,
3893 struct vmem_altmap *altmap);
3896 #ifdef CONFIG_SPARSEMEM_VMEMMAP
3897 static inline unsigned long vmem_altmap_offset(struct vmem_altmap *altmap)
3899 /* number of pfns from base where pfn_to_page() is valid */
3901 return altmap->reserve + altmap->free;
3905 static inline void vmem_altmap_free(struct vmem_altmap *altmap,
3906 unsigned long nr_pfns)
3908 altmap->alloc -= nr_pfns;
3911 static inline unsigned long vmem_altmap_offset(struct vmem_altmap *altmap)
3916 static inline void vmem_altmap_free(struct vmem_altmap *altmap,
3917 unsigned long nr_pfns)
3922 #define VMEMMAP_RESERVE_NR 2
3923 #ifdef CONFIG_ARCH_WANT_OPTIMIZE_DAX_VMEMMAP
3924 static inline bool __vmemmap_can_optimize(struct vmem_altmap *altmap,
3925 struct dev_pagemap *pgmap)
3927 unsigned long nr_pages;
3928 unsigned long nr_vmemmap_pages;
3930 if (!pgmap || !is_power_of_2(sizeof(struct page)))
3933 nr_pages = pgmap_vmemmap_nr(pgmap);
3934 nr_vmemmap_pages = ((nr_pages * sizeof(struct page)) >> PAGE_SHIFT);
3936 * For vmemmap optimization with DAX we need minimum 2 vmemmap
3937 * pages. See layout diagram in Documentation/mm/vmemmap_dedup.rst
3939 return !altmap && (nr_vmemmap_pages > VMEMMAP_RESERVE_NR);
3942 * If we don't have an architecture override, use the generic rule
3944 #ifndef vmemmap_can_optimize
3945 #define vmemmap_can_optimize __vmemmap_can_optimize
3949 static inline bool vmemmap_can_optimize(struct vmem_altmap *altmap,
3950 struct dev_pagemap *pgmap)
3956 void register_page_bootmem_memmap(unsigned long section_nr, struct page *map,
3957 unsigned long nr_pages);
3960 MF_COUNT_INCREASED = 1 << 0,
3961 MF_ACTION_REQUIRED = 1 << 1,
3962 MF_MUST_KILL = 1 << 2,
3963 MF_SOFT_OFFLINE = 1 << 3,
3964 MF_UNPOISON = 1 << 4,
3965 MF_SW_SIMULATED = 1 << 5,
3966 MF_NO_RETRY = 1 << 6,
3967 MF_MEM_PRE_REMOVE = 1 << 7,
3969 int mf_dax_kill_procs(struct address_space *mapping, pgoff_t index,
3970 unsigned long count, int mf_flags);
3971 extern int memory_failure(unsigned long pfn, int flags);
3972 extern void memory_failure_queue_kick(int cpu);
3973 extern int unpoison_memory(unsigned long pfn);
3974 extern void shake_page(struct page *p);
3975 extern atomic_long_t num_poisoned_pages __read_mostly;
3976 extern int soft_offline_page(unsigned long pfn, int flags);
3977 #ifdef CONFIG_MEMORY_FAILURE
3979 * Sysfs entries for memory failure handling statistics.
3981 extern const struct attribute_group memory_failure_attr_group;
3982 extern void memory_failure_queue(unsigned long pfn, int flags);
3983 extern int __get_huge_page_for_hwpoison(unsigned long pfn, int flags,
3984 bool *migratable_cleared);
3985 void num_poisoned_pages_inc(unsigned long pfn);
3986 void num_poisoned_pages_sub(unsigned long pfn, long i);
3987 struct task_struct *task_early_kill(struct task_struct *tsk, int force_early);
3989 static inline void memory_failure_queue(unsigned long pfn, int flags)
3993 static inline int __get_huge_page_for_hwpoison(unsigned long pfn, int flags,
3994 bool *migratable_cleared)
3999 static inline void num_poisoned_pages_inc(unsigned long pfn)
4003 static inline void num_poisoned_pages_sub(unsigned long pfn, long i)
4008 #if defined(CONFIG_MEMORY_FAILURE) && defined(CONFIG_KSM)
4009 void add_to_kill_ksm(struct task_struct *tsk, struct page *p,
4010 struct vm_area_struct *vma, struct list_head *to_kill,
4011 unsigned long ksm_addr);
4014 #if defined(CONFIG_MEMORY_FAILURE) && defined(CONFIG_MEMORY_HOTPLUG)
4015 extern void memblk_nr_poison_inc(unsigned long pfn);
4016 extern void memblk_nr_poison_sub(unsigned long pfn, long i);
4018 static inline void memblk_nr_poison_inc(unsigned long pfn)
4022 static inline void memblk_nr_poison_sub(unsigned long pfn, long i)
4027 #ifndef arch_memory_failure
4028 static inline int arch_memory_failure(unsigned long pfn, int flags)
4034 #ifndef arch_is_platform_page
4035 static inline bool arch_is_platform_page(u64 paddr)
4042 * Error handlers for various types of pages.
4045 MF_IGNORED, /* Error: cannot be handled */
4046 MF_FAILED, /* Error: handling failed */
4047 MF_DELAYED, /* Will be handled later */
4048 MF_RECOVERED, /* Successfully recovered */
4051 enum mf_action_page_type {
4053 MF_MSG_KERNEL_HIGH_ORDER,
4055 MF_MSG_DIFFERENT_COMPOUND,
4058 MF_MSG_UNMAP_FAILED,
4059 MF_MSG_DIRTY_SWAPCACHE,
4060 MF_MSG_CLEAN_SWAPCACHE,
4061 MF_MSG_DIRTY_MLOCKED_LRU,
4062 MF_MSG_CLEAN_MLOCKED_LRU,
4063 MF_MSG_DIRTY_UNEVICTABLE_LRU,
4064 MF_MSG_CLEAN_UNEVICTABLE_LRU,
4067 MF_MSG_TRUNCATED_LRU,
4074 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) || defined(CONFIG_HUGETLBFS)
4075 extern void clear_huge_page(struct page *page,
4076 unsigned long addr_hint,
4077 unsigned int pages_per_huge_page);
4078 int copy_user_large_folio(struct folio *dst, struct folio *src,
4079 unsigned long addr_hint,
4080 struct vm_area_struct *vma);
4081 long copy_folio_from_user(struct folio *dst_folio,
4082 const void __user *usr_src,
4083 bool allow_pagefault);
4086 * vma_is_special_huge - Are transhuge page-table entries considered special?
4087 * @vma: Pointer to the struct vm_area_struct to consider
4089 * Whether transhuge page-table entries are considered "special" following
4090 * the definition in vm_normal_page().
4092 * Return: true if transhuge page-table entries should be considered special,
4095 static inline bool vma_is_special_huge(const struct vm_area_struct *vma)
4097 return vma_is_dax(vma) || (vma->vm_file &&
4098 (vma->vm_flags & (VM_PFNMAP | VM_MIXEDMAP)));
4101 #endif /* CONFIG_TRANSPARENT_HUGEPAGE || CONFIG_HUGETLBFS */
4103 #if MAX_NUMNODES > 1
4104 void __init setup_nr_node_ids(void);
4106 static inline void setup_nr_node_ids(void) {}
4109 extern int memcmp_pages(struct page *page1, struct page *page2);
4111 static inline int pages_identical(struct page *page1, struct page *page2)
4113 return !memcmp_pages(page1, page2);
4116 #ifdef CONFIG_MAPPING_DIRTY_HELPERS
4117 unsigned long clean_record_shared_mapping_range(struct address_space *mapping,
4118 pgoff_t first_index, pgoff_t nr,
4119 pgoff_t bitmap_pgoff,
4120 unsigned long *bitmap,
4124 unsigned long wp_shared_mapping_range(struct address_space *mapping,
4125 pgoff_t first_index, pgoff_t nr);
4128 extern int sysctl_nr_trim_pages;
4130 #ifdef CONFIG_PRINTK
4131 void mem_dump_obj(void *object);
4133 static inline void mem_dump_obj(void *object) {}
4137 * seal_check_write - Check for F_SEAL_WRITE or F_SEAL_FUTURE_WRITE flags and
4139 * @seals: the seals to check
4140 * @vma: the vma to operate on
4142 * Check whether F_SEAL_WRITE or F_SEAL_FUTURE_WRITE are set; if so, do proper
4143 * check/handling on the vma flags. Return 0 if check pass, or <0 for errors.
4145 static inline int seal_check_write(int seals, struct vm_area_struct *vma)
4147 if (seals & (F_SEAL_WRITE | F_SEAL_FUTURE_WRITE)) {
4149 * New PROT_WRITE and MAP_SHARED mmaps are not allowed when
4150 * write seals are active.
4152 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_WRITE))
4156 * Since an F_SEAL_[FUTURE_]WRITE sealed memfd can be mapped as
4157 * MAP_SHARED and read-only, take care to not allow mprotect to
4158 * revert protections on such mappings. Do this only for shared
4159 * mappings. For private mappings, don't need to mask
4160 * VM_MAYWRITE as we still want them to be COW-writable.
4162 if (vma->vm_flags & VM_SHARED)
4163 vm_flags_clear(vma, VM_MAYWRITE);
4169 #ifdef CONFIG_ANON_VMA_NAME
4170 int madvise_set_anon_name(struct mm_struct *mm, unsigned long start,
4171 unsigned long len_in,
4172 struct anon_vma_name *anon_name);
4175 madvise_set_anon_name(struct mm_struct *mm, unsigned long start,
4176 unsigned long len_in, struct anon_vma_name *anon_name) {
4181 #ifdef CONFIG_UNACCEPTED_MEMORY
4183 bool range_contains_unaccepted_memory(phys_addr_t start, phys_addr_t end);
4184 void accept_memory(phys_addr_t start, phys_addr_t end);
4188 static inline bool range_contains_unaccepted_memory(phys_addr_t start,
4194 static inline void accept_memory(phys_addr_t start, phys_addr_t end)
4200 static inline bool pfn_is_unaccepted_memory(unsigned long pfn)
4202 phys_addr_t paddr = pfn << PAGE_SHIFT;
4204 return range_contains_unaccepted_memory(paddr, paddr + PAGE_SIZE);
4207 #endif /* _LINUX_MM_H */