1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef _LINUX_MM_TYPES_H
3 #define _LINUX_MM_TYPES_H
5 #include <linux/mm_types_task.h>
7 #include <linux/auxvec.h>
8 #include <linux/list.h>
9 #include <linux/spinlock.h>
10 #include <linux/rbtree.h>
11 #include <linux/rwsem.h>
12 #include <linux/completion.h>
13 #include <linux/cpumask.h>
14 #include <linux/uprobes.h>
15 #include <linux/page-flags-layout.h>
16 #include <linux/workqueue.h>
20 #ifndef AT_VECTOR_SIZE_ARCH
21 #define AT_VECTOR_SIZE_ARCH 0
23 #define AT_VECTOR_SIZE (2*(AT_VECTOR_SIZE_ARCH + AT_VECTOR_SIZE_BASE + 1))
30 * Each physical page in the system has a struct page associated with
31 * it to keep track of whatever it is we are using the page for at the
32 * moment. Note that we have no way to track which tasks are using
33 * a page, though if it is a pagecache page, rmap structures can tell us
36 * The objects in struct page are organized in double word blocks in
37 * order to allows us to use atomic double word operations on portions
38 * of struct page. That is currently only used by slub but the arrangement
39 * allows the use of atomic double word operations on the flags/mapping
40 * and lru list pointers also.
43 /* First double word block */
44 unsigned long flags; /* Atomic flags, some possibly
45 * updated asynchronously */
47 struct address_space *mapping; /* If low bit clear, points to
48 * inode address_space, or NULL.
49 * If page mapped as anonymous
50 * memory, low bit is set, and
51 * it points to anon_vma object:
52 * see PAGE_MAPPING_ANON below.
54 void *s_mem; /* slab first object */
55 atomic_t compound_mapcount; /* first tail page */
56 /* page_deferred_list().next -- second tail page */
59 /* Second double word */
61 pgoff_t index; /* Our offset within mapping. */
62 void *freelist; /* sl[aou]b first free object */
63 /* page_deferred_list().prev -- second tail page */
67 #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \
68 defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
69 /* Used for cmpxchg_double in slub */
70 unsigned long counters;
73 * Keep _refcount separate from slub cmpxchg_double data.
74 * As the rest of the double word is protected by slab_lock
75 * but _refcount is not.
83 * Count of ptes mapped in mms, to show when
84 * page is mapped & limit reverse map searches.
86 * Extra information about page type may be
87 * stored here for pages that are never mapped,
88 * in which case the value MUST BE <= -2.
89 * See page-flags.h for more details.
93 unsigned int active; /* SLAB */
102 * Usage count, *USE WRAPPER FUNCTION* when manual
103 * accounting. See page_ref.h
110 * Third double word block
112 * WARNING: bit 0 of the first word encode PageTail(). That means
113 * the rest users of the storage space MUST NOT use the bit to
114 * avoid collision and false-positive PageTail().
117 struct list_head lru; /* Pageout list, eg. active_list
118 * protected by zone_lru_lock !
119 * Can be used as a generic list
122 struct dev_pagemap *pgmap; /* ZONE_DEVICE pages are never on an
123 * lru or handled by a slab
124 * allocator, this points to the
125 * hosting device page map.
127 struct { /* slub per cpu partial pages */
128 struct page *next; /* Next partial slab */
130 int pages; /* Nr of partial slabs left */
131 int pobjects; /* Approximate # of objects */
138 struct rcu_head rcu_head; /* Used by SLAB
139 * when destroying via RCU
141 /* Tail pages of compound page */
143 unsigned long compound_head; /* If bit zero is set */
145 /* First tail page only */
148 * On 64 bit system we have enough space in struct page
149 * to encode compound_dtor and compound_order with
150 * unsigned int. It can help compiler generate better or
151 * smaller code on some archtectures.
153 unsigned int compound_dtor;
154 unsigned int compound_order;
156 unsigned short int compound_dtor;
157 unsigned short int compound_order;
161 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && USE_SPLIT_PMD_PTLOCKS
163 unsigned long __pad; /* do not overlay pmd_huge_pte
164 * with compound_head to avoid
165 * possible bit 0 collision.
167 pgtable_t pmd_huge_pte; /* protected by page->ptl */
172 /* Remainder is not double word aligned */
174 unsigned long private; /* Mapping-private opaque data:
175 * usually used for buffer_heads
176 * if PagePrivate set; used for
177 * swp_entry_t if PageSwapCache;
178 * indicates order in the buddy
179 * system if PG_buddy is set.
181 #if USE_SPLIT_PTE_PTLOCKS
182 #if ALLOC_SPLIT_PTLOCKS
188 struct kmem_cache *slab_cache; /* SL[AU]B: Pointer to slab */
192 struct mem_cgroup *mem_cgroup;
196 * On machines where all RAM is mapped into kernel address space,
197 * we can simply calculate the virtual address. On machines with
198 * highmem some memory is mapped into kernel virtual memory
199 * dynamically, so we need a place to store that address.
200 * Note that this field could be 16 bits on x86 ... ;)
202 * Architectures with slow multiplication can define
203 * WANT_PAGE_VIRTUAL in asm/page.h
205 #if defined(WANT_PAGE_VIRTUAL)
206 void *virtual; /* Kernel virtual address (NULL if
207 not kmapped, ie. highmem) */
208 #endif /* WANT_PAGE_VIRTUAL */
210 #ifdef LAST_CPUPID_NOT_IN_PAGE_FLAGS
215 * The struct page can be forced to be double word aligned so that atomic ops
216 * on double words work. The SLUB allocator can make use of such a feature.
218 #ifdef CONFIG_HAVE_ALIGNED_STRUCT_PAGE
219 __aligned(2 * sizeof(unsigned long))
223 #define PAGE_FRAG_CACHE_MAX_SIZE __ALIGN_MASK(32768, ~PAGE_MASK)
224 #define PAGE_FRAG_CACHE_MAX_ORDER get_order(PAGE_FRAG_CACHE_MAX_SIZE)
226 struct page_frag_cache {
228 #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
234 /* we maintain a pagecount bias, so that we dont dirty cache line
235 * containing page->_refcount every time we allocate a fragment.
237 unsigned int pagecnt_bias;
241 typedef unsigned long vm_flags_t;
243 static inline atomic_t *compound_mapcount_ptr(struct page *page)
245 return &page[1].compound_mapcount;
249 * A region containing a mapping of a non-memory backed file under NOMMU
250 * conditions. These are held in a global tree and are pinned by the VMAs that
254 struct rb_node vm_rb; /* link in global region tree */
255 vm_flags_t vm_flags; /* VMA vm_flags */
256 unsigned long vm_start; /* start address of region */
257 unsigned long vm_end; /* region initialised to here */
258 unsigned long vm_top; /* region allocated to here */
259 unsigned long vm_pgoff; /* the offset in vm_file corresponding to vm_start */
260 struct file *vm_file; /* the backing file or NULL */
262 int vm_usage; /* region usage count (access under nommu_region_sem) */
263 bool vm_icache_flushed : 1; /* true if the icache has been flushed for
267 #ifdef CONFIG_USERFAULTFD
268 #define NULL_VM_UFFD_CTX ((struct vm_userfaultfd_ctx) { NULL, })
269 struct vm_userfaultfd_ctx {
270 struct userfaultfd_ctx *ctx;
272 #else /* CONFIG_USERFAULTFD */
273 #define NULL_VM_UFFD_CTX ((struct vm_userfaultfd_ctx) {})
274 struct vm_userfaultfd_ctx {};
275 #endif /* CONFIG_USERFAULTFD */
278 * This struct defines a memory VMM memory area. There is one of these
279 * per VM-area/task. A VM area is any part of the process virtual memory
280 * space that has a special rule for the page-fault handlers (ie a shared
281 * library, the executable area etc).
283 struct vm_area_struct {
284 /* The first cache line has the info for VMA tree walking. */
286 unsigned long vm_start; /* Our start address within vm_mm. */
287 unsigned long vm_end; /* The first byte after our end address
290 /* linked list of VM areas per task, sorted by address */
291 struct vm_area_struct *vm_next, *vm_prev;
293 struct rb_node vm_rb;
296 * Largest free memory gap in bytes to the left of this VMA.
297 * Either between this VMA and vma->vm_prev, or between one of the
298 * VMAs below us in the VMA rbtree and its ->vm_prev. This helps
299 * get_unmapped_area find a free area of the right size.
301 unsigned long rb_subtree_gap;
303 /* Second cache line starts here. */
305 struct mm_struct *vm_mm; /* The address space we belong to. */
306 pgprot_t vm_page_prot; /* Access permissions of this VMA. */
307 unsigned long vm_flags; /* Flags, see mm.h. */
310 * For areas with an address space and backing store,
311 * linkage into the address_space->i_mmap interval tree.
315 unsigned long rb_subtree_last;
319 * A file's MAP_PRIVATE vma can be in both i_mmap tree and anon_vma
320 * list, after a COW of one of the file pages. A MAP_SHARED vma
321 * can only be in the i_mmap tree. An anonymous MAP_PRIVATE, stack
322 * or brk vma (with NULL file) can only be in an anon_vma list.
324 struct list_head anon_vma_chain; /* Serialized by mmap_sem &
326 struct anon_vma *anon_vma; /* Serialized by page_table_lock */
328 /* Function pointers to deal with this struct. */
329 const struct vm_operations_struct *vm_ops;
331 /* Information about our backing store: */
332 unsigned long vm_pgoff; /* Offset (within vm_file) in PAGE_SIZE
334 struct file * vm_file; /* File we map to (can be NULL). */
335 void * vm_private_data; /* was vm_pte (shared mem) */
337 atomic_long_t swap_readahead_info;
339 struct vm_region *vm_region; /* NOMMU mapping region */
342 struct mempolicy *vm_policy; /* NUMA policy for the VMA */
344 struct vm_userfaultfd_ctx vm_userfaultfd_ctx;
345 } __randomize_layout;
348 struct task_struct *task;
349 struct core_thread *next;
354 struct core_thread dumper;
355 struct completion startup;
360 struct vm_area_struct *mmap; /* list of VMAs */
361 struct rb_root mm_rb;
362 u64 vmacache_seqnum; /* per-thread vmacache */
364 unsigned long (*get_unmapped_area) (struct file *filp,
365 unsigned long addr, unsigned long len,
366 unsigned long pgoff, unsigned long flags);
368 unsigned long mmap_base; /* base of mmap area */
369 unsigned long mmap_legacy_base; /* base of mmap area in bottom-up allocations */
370 #ifdef CONFIG_HAVE_ARCH_COMPAT_MMAP_BASES
371 /* Base adresses for compatible mmap() */
372 unsigned long mmap_compat_base;
373 unsigned long mmap_compat_legacy_base;
375 unsigned long task_size; /* size of task vm space */
376 unsigned long highest_vm_end; /* highest vma end address */
380 * @mm_users: The number of users including userspace.
382 * Use mmget()/mmget_not_zero()/mmput() to modify. When this drops
383 * to 0 (i.e. when the task exits and there are no other temporary
384 * reference holders), we also release a reference on @mm_count
385 * (which may then free the &struct mm_struct if @mm_count also
391 * @mm_count: The number of references to &struct mm_struct
392 * (@mm_users count as 1).
394 * Use mmgrab()/mmdrop() to modify. When this drops to 0, the
395 * &struct mm_struct is freed.
399 atomic_long_t nr_ptes; /* PTE page table pages */
400 #if CONFIG_PGTABLE_LEVELS > 2
401 atomic_long_t nr_pmds; /* PMD page table pages */
403 int map_count; /* number of VMAs */
405 spinlock_t page_table_lock; /* Protects page tables and some counters */
406 struct rw_semaphore mmap_sem;
408 struct list_head mmlist; /* List of maybe swapped mm's. These are globally strung
409 * together off init_mm.mmlist, and are protected
414 unsigned long hiwater_rss; /* High-watermark of RSS usage */
415 unsigned long hiwater_vm; /* High-water virtual memory usage */
417 unsigned long total_vm; /* Total pages mapped */
418 unsigned long locked_vm; /* Pages that have PG_mlocked set */
419 unsigned long pinned_vm; /* Refcount permanently increased */
420 unsigned long data_vm; /* VM_WRITE & ~VM_SHARED & ~VM_STACK */
421 unsigned long exec_vm; /* VM_EXEC & ~VM_WRITE & ~VM_STACK */
422 unsigned long stack_vm; /* VM_STACK */
423 unsigned long def_flags;
424 unsigned long start_code, end_code, start_data, end_data;
425 unsigned long start_brk, brk, start_stack;
426 unsigned long arg_start, arg_end, env_start, env_end;
428 unsigned long saved_auxv[AT_VECTOR_SIZE]; /* for /proc/PID/auxv */
431 * Special counters, in some configurations protected by the
432 * page_table_lock, in other configurations by being atomic.
434 struct mm_rss_stat rss_stat;
436 struct linux_binfmt *binfmt;
438 cpumask_var_t cpu_vm_mask_var;
440 /* Architecture-specific MM context */
441 mm_context_t context;
443 unsigned long flags; /* Must use atomic bitops to access the bits */
445 struct core_state *core_state; /* coredumping support */
446 #ifdef CONFIG_MEMBARRIER
447 atomic_t membarrier_state;
450 spinlock_t ioctx_lock;
451 struct kioctx_table __rcu *ioctx_table;
455 * "owner" points to a task that is regarded as the canonical
456 * user/owner of this mm. All of the following must be true in
457 * order for it to be changed:
459 * current == mm->owner
461 * new_owner->mm == mm
462 * new_owner->alloc_lock is held
464 struct task_struct __rcu *owner;
466 struct user_namespace *user_ns;
468 /* store ref to file /proc/<pid>/exe symlink points to */
469 struct file __rcu *exe_file;
470 #ifdef CONFIG_MMU_NOTIFIER
471 struct mmu_notifier_mm *mmu_notifier_mm;
473 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
474 pgtable_t pmd_huge_pte; /* protected by page_table_lock */
476 #ifdef CONFIG_CPUMASK_OFFSTACK
477 struct cpumask cpumask_allocation;
479 #ifdef CONFIG_NUMA_BALANCING
481 * numa_next_scan is the next time that the PTEs will be marked
482 * pte_numa. NUMA hinting faults will gather statistics and migrate
483 * pages to new nodes if necessary.
485 unsigned long numa_next_scan;
487 /* Restart point for scanning and setting pte_numa */
488 unsigned long numa_scan_offset;
490 /* numa_scan_seq prevents two threads setting pte_numa */
494 * An operation with batched TLB flushing is going on. Anything that
495 * can move process memory needs to flush the TLB when moving a
496 * PROT_NONE or PROT_NUMA mapped page.
498 atomic_t tlb_flush_pending;
499 #ifdef CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH
500 /* See flush_tlb_batched_pending() */
501 bool tlb_flush_batched;
503 struct uprobes_state uprobes_state;
504 #ifdef CONFIG_HUGETLB_PAGE
505 atomic_long_t hugetlb_usage;
507 struct work_struct async_put_work;
509 #if IS_ENABLED(CONFIG_HMM)
510 /* HMM needs to track a few things per mm */
513 } __randomize_layout;
515 extern struct mm_struct init_mm;
517 static inline void mm_init_cpumask(struct mm_struct *mm)
519 #ifdef CONFIG_CPUMASK_OFFSTACK
520 mm->cpu_vm_mask_var = &mm->cpumask_allocation;
522 cpumask_clear(mm->cpu_vm_mask_var);
525 /* Future-safe accessor for struct mm_struct's cpu_vm_mask. */
526 static inline cpumask_t *mm_cpumask(struct mm_struct *mm)
528 return mm->cpu_vm_mask_var;
532 extern void tlb_gather_mmu(struct mmu_gather *tlb, struct mm_struct *mm,
533 unsigned long start, unsigned long end);
534 extern void tlb_finish_mmu(struct mmu_gather *tlb,
535 unsigned long start, unsigned long end);
537 static inline void init_tlb_flush_pending(struct mm_struct *mm)
539 atomic_set(&mm->tlb_flush_pending, 0);
542 static inline void inc_tlb_flush_pending(struct mm_struct *mm)
544 atomic_inc(&mm->tlb_flush_pending);
546 * The only time this value is relevant is when there are indeed pages
547 * to flush. And we'll only flush pages after changing them, which
550 * So the ordering here is:
552 * atomic_inc(&mm->tlb_flush_pending);
559 * mm_tlb_flush_pending();
564 * atomic_dec(&mm->tlb_flush_pending);
566 * Where the increment if constrained by the PTL unlock, it thus
567 * ensures that the increment is visible if the PTE modification is
568 * visible. After all, if there is no PTE modification, nobody cares
569 * about TLB flushes either.
571 * This very much relies on users (mm_tlb_flush_pending() and
572 * mm_tlb_flush_nested()) only caring about _specific_ PTEs (and
573 * therefore specific PTLs), because with SPLIT_PTE_PTLOCKS and RCpc
574 * locks (PPC) the unlock of one doesn't order against the lock of
577 * The decrement is ordered by the flush_tlb_range(), such that
578 * mm_tlb_flush_pending() will not return false unless all flushes have
583 static inline void dec_tlb_flush_pending(struct mm_struct *mm)
586 * See inc_tlb_flush_pending().
588 * This cannot be smp_mb__before_atomic() because smp_mb() simply does
589 * not order against TLB invalidate completion, which is what we need.
591 * Therefore we must rely on tlb_flush_*() to guarantee order.
593 atomic_dec(&mm->tlb_flush_pending);
596 static inline bool mm_tlb_flush_pending(struct mm_struct *mm)
599 * Must be called after having acquired the PTL; orders against that
600 * PTLs release and therefore ensures that if we observe the modified
601 * PTE we must also observe the increment from inc_tlb_flush_pending().
603 * That is, it only guarantees to return true if there is a flush
604 * pending for _this_ PTL.
606 return atomic_read(&mm->tlb_flush_pending);
609 static inline bool mm_tlb_flush_nested(struct mm_struct *mm)
612 * Similar to mm_tlb_flush_pending(), we must have acquired the PTL
613 * for which there is a TLB flush pending in order to guarantee
614 * we've seen both that PTE modification and the increment.
616 * (no requirement on actually still holding the PTL, that is irrelevant)
618 return atomic_read(&mm->tlb_flush_pending) > 1;
623 struct vm_special_mapping {
624 const char *name; /* The name, e.g. "[vdso]". */
627 * If .fault is not provided, this points to a
628 * NULL-terminated array of pages that back the special mapping.
630 * This must not be NULL unless .fault is provided.
635 * If non-NULL, then this is called to resolve page faults
636 * on the special mapping. If used, .pages is not checked.
638 int (*fault)(const struct vm_special_mapping *sm,
639 struct vm_area_struct *vma,
640 struct vm_fault *vmf);
642 int (*mremap)(const struct vm_special_mapping *sm,
643 struct vm_area_struct *new_vma);
646 enum tlb_flush_reason {
647 TLB_FLUSH_ON_TASK_SWITCH,
648 TLB_REMOTE_SHOOTDOWN,
650 TLB_LOCAL_MM_SHOOTDOWN,
652 NR_TLB_FLUSH_REASONS,
656 * A swap entry has to fit into a "unsigned long", as the entry is hidden
657 * in the "index" field of the swapper address space.
663 #endif /* _LINUX_MM_TYPES_H */