1 #ifndef _LINUX_MM_TYPES_H
2 #define _LINUX_MM_TYPES_H
4 #include <linux/auxvec.h>
5 #include <linux/types.h>
6 #include <linux/threads.h>
7 #include <linux/list.h>
8 #include <linux/spinlock.h>
9 #include <linux/rbtree.h>
10 #include <linux/rwsem.h>
11 #include <linux/completion.h>
12 #include <linux/cpumask.h>
13 #include <linux/uprobes.h>
14 #include <linux/page-flags-layout.h>
15 #include <linux/workqueue.h>
19 #ifndef AT_VECTOR_SIZE_ARCH
20 #define AT_VECTOR_SIZE_ARCH 0
22 #define AT_VECTOR_SIZE (2*(AT_VECTOR_SIZE_ARCH + AT_VECTOR_SIZE_BASE + 1))
27 #define USE_SPLIT_PTE_PTLOCKS (NR_CPUS >= CONFIG_SPLIT_PTLOCK_CPUS)
28 #define USE_SPLIT_PMD_PTLOCKS (USE_SPLIT_PTE_PTLOCKS && \
29 IS_ENABLED(CONFIG_ARCH_ENABLE_SPLIT_PMD_PTLOCK))
30 #define ALLOC_SPLIT_PTLOCKS (SPINLOCK_SIZE > BITS_PER_LONG/8)
33 * Each physical page in the system has a struct page associated with
34 * it to keep track of whatever it is we are using the page for at the
35 * moment. Note that we have no way to track which tasks are using
36 * a page, though if it is a pagecache page, rmap structures can tell us
39 * The objects in struct page are organized in double word blocks in
40 * order to allows us to use atomic double word operations on portions
41 * of struct page. That is currently only used by slub but the arrangement
42 * allows the use of atomic double word operations on the flags/mapping
43 * and lru list pointers also.
46 /* First double word block */
47 unsigned long flags
; /* Atomic flags, some possibly
48 * updated asynchronously */
50 struct address_space
*mapping
; /* If low bit clear, points to
51 * inode address_space, or NULL.
52 * If page mapped as anonymous
53 * memory, low bit is set, and
54 * it points to anon_vma object:
55 * see PAGE_MAPPING_ANON below.
57 void *s_mem
; /* slab first object */
58 atomic_t compound_mapcount
; /* first tail page */
59 /* page_deferred_list().next -- second tail page */
62 /* Second double word */
64 pgoff_t index
; /* Our offset within mapping. */
65 void *freelist
; /* sl[aou]b first free object */
66 /* page_deferred_list().prev -- second tail page */
70 #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \
71 defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
72 /* Used for cmpxchg_double in slub */
73 unsigned long counters
;
76 * Keep _refcount separate from slub cmpxchg_double data.
77 * As the rest of the double word is protected by slab_lock
78 * but _refcount is not.
86 * Count of ptes mapped in mms, to show when
87 * page is mapped & limit reverse map searches.
89 * Extra information about page type may be
90 * stored here for pages that are never mapped,
91 * in which case the value MUST BE <= -2.
92 * See page-flags.h for more details.
96 unsigned int active
; /* SLAB */
102 int units
; /* SLOB */
105 * Usage count, *USE WRAPPER FUNCTION* when manual
106 * accounting. See page_ref.h
113 * Third double word block
115 * WARNING: bit 0 of the first word encode PageTail(). That means
116 * the rest users of the storage space MUST NOT use the bit to
117 * avoid collision and false-positive PageTail().
120 struct list_head lru
; /* Pageout list, eg. active_list
121 * protected by zone_lru_lock !
122 * Can be used as a generic list
125 struct dev_pagemap
*pgmap
; /* ZONE_DEVICE pages are never on an
126 * lru or handled by a slab
127 * allocator, this points to the
128 * hosting device page map.
130 struct { /* slub per cpu partial pages */
131 struct page
*next
; /* Next partial slab */
133 int pages
; /* Nr of partial slabs left */
134 int pobjects
; /* Approximate # of objects */
141 struct rcu_head rcu_head
; /* Used by SLAB
142 * when destroying via RCU
144 /* Tail pages of compound page */
146 unsigned long compound_head
; /* If bit zero is set */
148 /* First tail page only */
151 * On 64 bit system we have enough space in struct page
152 * to encode compound_dtor and compound_order with
153 * unsigned int. It can help compiler generate better or
154 * smaller code on some archtectures.
156 unsigned int compound_dtor
;
157 unsigned int compound_order
;
159 unsigned short int compound_dtor
;
160 unsigned short int compound_order
;
164 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && USE_SPLIT_PMD_PTLOCKS
166 unsigned long __pad
; /* do not overlay pmd_huge_pte
167 * with compound_head to avoid
168 * possible bit 0 collision.
170 pgtable_t pmd_huge_pte
; /* protected by page->ptl */
175 /* Remainder is not double word aligned */
177 unsigned long private; /* Mapping-private opaque data:
178 * usually used for buffer_heads
179 * if PagePrivate set; used for
180 * swp_entry_t if PageSwapCache;
181 * indicates order in the buddy
182 * system if PG_buddy is set.
184 #if USE_SPLIT_PTE_PTLOCKS
185 #if ALLOC_SPLIT_PTLOCKS
191 struct kmem_cache
*slab_cache
; /* SL[AU]B: Pointer to slab */
195 struct mem_cgroup
*mem_cgroup
;
199 * On machines where all RAM is mapped into kernel address space,
200 * we can simply calculate the virtual address. On machines with
201 * highmem some memory is mapped into kernel virtual memory
202 * dynamically, so we need a place to store that address.
203 * Note that this field could be 16 bits on x86 ... ;)
205 * Architectures with slow multiplication can define
206 * WANT_PAGE_VIRTUAL in asm/page.h
208 #if defined(WANT_PAGE_VIRTUAL)
209 void *virtual; /* Kernel virtual address (NULL if
210 not kmapped, ie. highmem) */
211 #endif /* WANT_PAGE_VIRTUAL */
213 #ifdef CONFIG_KMEMCHECK
215 * kmemcheck wants to track the status of each byte in a page; this
216 * is a pointer to such a status block. NULL if not tracked.
221 #ifdef LAST_CPUPID_NOT_IN_PAGE_FLAGS
226 * The struct page can be forced to be double word aligned so that atomic ops
227 * on double words work. The SLUB allocator can make use of such a feature.
229 #ifdef CONFIG_HAVE_ALIGNED_STRUCT_PAGE
230 __aligned(2 * sizeof(unsigned long))
236 #if (BITS_PER_LONG > 32) || (PAGE_SIZE >= 65536)
245 #define PAGE_FRAG_CACHE_MAX_SIZE __ALIGN_MASK(32768, ~PAGE_MASK)
246 #define PAGE_FRAG_CACHE_MAX_ORDER get_order(PAGE_FRAG_CACHE_MAX_SIZE)
248 struct page_frag_cache
{
250 #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
256 /* we maintain a pagecount bias, so that we dont dirty cache line
257 * containing page->_refcount every time we allocate a fragment.
259 unsigned int pagecnt_bias
;
263 typedef unsigned long vm_flags_t
;
266 * A region containing a mapping of a non-memory backed file under NOMMU
267 * conditions. These are held in a global tree and are pinned by the VMAs that
271 struct rb_node vm_rb
; /* link in global region tree */
272 vm_flags_t vm_flags
; /* VMA vm_flags */
273 unsigned long vm_start
; /* start address of region */
274 unsigned long vm_end
; /* region initialised to here */
275 unsigned long vm_top
; /* region allocated to here */
276 unsigned long vm_pgoff
; /* the offset in vm_file corresponding to vm_start */
277 struct file
*vm_file
; /* the backing file or NULL */
279 int vm_usage
; /* region usage count (access under nommu_region_sem) */
280 bool vm_icache_flushed
: 1; /* true if the icache has been flushed for
284 #ifdef CONFIG_USERFAULTFD
285 #define NULL_VM_UFFD_CTX ((struct vm_userfaultfd_ctx) { NULL, })
286 struct vm_userfaultfd_ctx
{
287 struct userfaultfd_ctx
*ctx
;
289 #else /* CONFIG_USERFAULTFD */
290 #define NULL_VM_UFFD_CTX ((struct vm_userfaultfd_ctx) {})
291 struct vm_userfaultfd_ctx
{};
292 #endif /* CONFIG_USERFAULTFD */
295 * This struct defines a memory VMM memory area. There is one of these
296 * per VM-area/task. A VM area is any part of the process virtual memory
297 * space that has a special rule for the page-fault handlers (ie a shared
298 * library, the executable area etc).
300 struct vm_area_struct
{
301 /* The first cache line has the info for VMA tree walking. */
303 unsigned long vm_start
; /* Our start address within vm_mm. */
304 unsigned long vm_end
; /* The first byte after our end address
307 /* linked list of VM areas per task, sorted by address */
308 struct vm_area_struct
*vm_next
, *vm_prev
;
310 struct rb_node vm_rb
;
313 * Largest free memory gap in bytes to the left of this VMA.
314 * Either between this VMA and vma->vm_prev, or between one of the
315 * VMAs below us in the VMA rbtree and its ->vm_prev. This helps
316 * get_unmapped_area find a free area of the right size.
318 unsigned long rb_subtree_gap
;
320 /* Second cache line starts here. */
322 struct mm_struct
*vm_mm
; /* The address space we belong to. */
323 pgprot_t vm_page_prot
; /* Access permissions of this VMA. */
324 unsigned long vm_flags
; /* Flags, see mm.h. */
327 * For areas with an address space and backing store,
328 * linkage into the address_space->i_mmap interval tree.
332 unsigned long rb_subtree_last
;
336 * A file's MAP_PRIVATE vma can be in both i_mmap tree and anon_vma
337 * list, after a COW of one of the file pages. A MAP_SHARED vma
338 * can only be in the i_mmap tree. An anonymous MAP_PRIVATE, stack
339 * or brk vma (with NULL file) can only be in an anon_vma list.
341 struct list_head anon_vma_chain
; /* Serialized by mmap_sem &
343 struct anon_vma
*anon_vma
; /* Serialized by page_table_lock */
345 /* Function pointers to deal with this struct. */
346 const struct vm_operations_struct
*vm_ops
;
348 /* Information about our backing store: */
349 unsigned long vm_pgoff
; /* Offset (within vm_file) in PAGE_SIZE
351 struct file
* vm_file
; /* File we map to (can be NULL). */
352 void * vm_private_data
; /* was vm_pte (shared mem) */
355 struct vm_region
*vm_region
; /* NOMMU mapping region */
358 struct mempolicy
*vm_policy
; /* NUMA policy for the VMA */
360 struct vm_userfaultfd_ctx vm_userfaultfd_ctx
;
364 struct task_struct
*task
;
365 struct core_thread
*next
;
370 struct core_thread dumper
;
371 struct completion startup
;
375 MM_FILEPAGES
, /* Resident file mapping pages */
376 MM_ANONPAGES
, /* Resident anonymous pages */
377 MM_SWAPENTS
, /* Anonymous swap entries */
378 MM_SHMEMPAGES
, /* Resident shared memory pages */
382 #if USE_SPLIT_PTE_PTLOCKS && defined(CONFIG_MMU)
383 #define SPLIT_RSS_COUNTING
384 /* per-thread cached information, */
385 struct task_rss_stat
{
386 int events
; /* for synchronization threshold */
387 int count
[NR_MM_COUNTERS
];
389 #endif /* USE_SPLIT_PTE_PTLOCKS */
392 atomic_long_t count
[NR_MM_COUNTERS
];
397 struct vm_area_struct
*mmap
; /* list of VMAs */
398 struct rb_root mm_rb
;
399 u32 vmacache_seqnum
; /* per-thread vmacache */
401 unsigned long (*get_unmapped_area
) (struct file
*filp
,
402 unsigned long addr
, unsigned long len
,
403 unsigned long pgoff
, unsigned long flags
);
405 unsigned long mmap_base
; /* base of mmap area */
406 unsigned long mmap_legacy_base
; /* base of mmap area in bottom-up allocations */
407 unsigned long task_size
; /* size of task vm space */
408 unsigned long highest_vm_end
; /* highest vma end address */
410 atomic_t mm_users
; /* How many users with user space? */
411 atomic_t mm_count
; /* How many references to "struct mm_struct" (users count as 1) */
412 atomic_long_t nr_ptes
; /* PTE page table pages */
413 #if CONFIG_PGTABLE_LEVELS > 2
414 atomic_long_t nr_pmds
; /* PMD page table pages */
416 int map_count
; /* number of VMAs */
418 spinlock_t page_table_lock
; /* Protects page tables and some counters */
419 struct rw_semaphore mmap_sem
;
421 struct list_head mmlist
; /* List of maybe swapped mm's. These are globally strung
422 * together off init_mm.mmlist, and are protected
427 unsigned long hiwater_rss
; /* High-watermark of RSS usage */
428 unsigned long hiwater_vm
; /* High-water virtual memory usage */
430 unsigned long total_vm
; /* Total pages mapped */
431 unsigned long locked_vm
; /* Pages that have PG_mlocked set */
432 unsigned long pinned_vm
; /* Refcount permanently increased */
433 unsigned long data_vm
; /* VM_WRITE & ~VM_SHARED & ~VM_STACK */
434 unsigned long exec_vm
; /* VM_EXEC & ~VM_WRITE & ~VM_STACK */
435 unsigned long stack_vm
; /* VM_STACK */
436 unsigned long def_flags
;
437 unsigned long start_code
, end_code
, start_data
, end_data
;
438 unsigned long start_brk
, brk
, start_stack
;
439 unsigned long arg_start
, arg_end
, env_start
, env_end
;
441 unsigned long saved_auxv
[AT_VECTOR_SIZE
]; /* for /proc/PID/auxv */
444 * Special counters, in some configurations protected by the
445 * page_table_lock, in other configurations by being atomic.
447 struct mm_rss_stat rss_stat
;
449 struct linux_binfmt
*binfmt
;
451 cpumask_var_t cpu_vm_mask_var
;
453 /* Architecture-specific MM context */
454 mm_context_t context
;
456 unsigned long flags
; /* Must use atomic bitops to access the bits */
458 struct core_state
*core_state
; /* coredumping support */
460 spinlock_t ioctx_lock
;
461 struct kioctx_table __rcu
*ioctx_table
;
465 * "owner" points to a task that is regarded as the canonical
466 * user/owner of this mm. All of the following must be true in
467 * order for it to be changed:
469 * current == mm->owner
471 * new_owner->mm == mm
472 * new_owner->alloc_lock is held
474 struct task_struct __rcu
*owner
;
476 struct user_namespace
*user_ns
;
478 /* store ref to file /proc/<pid>/exe symlink points to */
479 struct file __rcu
*exe_file
;
480 #ifdef CONFIG_MMU_NOTIFIER
481 struct mmu_notifier_mm
*mmu_notifier_mm
;
483 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
484 pgtable_t pmd_huge_pte
; /* protected by page_table_lock */
486 #ifdef CONFIG_CPUMASK_OFFSTACK
487 struct cpumask cpumask_allocation
;
489 #ifdef CONFIG_NUMA_BALANCING
491 * numa_next_scan is the next time that the PTEs will be marked
492 * pte_numa. NUMA hinting faults will gather statistics and migrate
493 * pages to new nodes if necessary.
495 unsigned long numa_next_scan
;
497 /* Restart point for scanning and setting pte_numa */
498 unsigned long numa_scan_offset
;
500 /* numa_scan_seq prevents two threads setting pte_numa */
503 #if defined(CONFIG_NUMA_BALANCING) || defined(CONFIG_COMPACTION)
505 * An operation with batched TLB flushing is going on. Anything that
506 * can move process memory needs to flush the TLB when moving a
507 * PROT_NONE or PROT_NUMA mapped page.
509 bool tlb_flush_pending
;
511 struct uprobes_state uprobes_state
;
512 #ifdef CONFIG_HUGETLB_PAGE
513 atomic_long_t hugetlb_usage
;
515 struct work_struct async_put_work
;
518 static inline void mm_init_cpumask(struct mm_struct
*mm
)
520 #ifdef CONFIG_CPUMASK_OFFSTACK
521 mm
->cpu_vm_mask_var
= &mm
->cpumask_allocation
;
523 cpumask_clear(mm
->cpu_vm_mask_var
);
526 /* Future-safe accessor for struct mm_struct's cpu_vm_mask. */
527 static inline cpumask_t
*mm_cpumask(struct mm_struct
*mm
)
529 return mm
->cpu_vm_mask_var
;
532 #if defined(CONFIG_NUMA_BALANCING) || defined(CONFIG_COMPACTION)
534 * Memory barriers to keep this state in sync are graciously provided by
535 * the page table locks, outside of which no page table modifications happen.
536 * The barriers below prevent the compiler from re-ordering the instructions
537 * around the memory barriers that are already present in the code.
539 static inline bool mm_tlb_flush_pending(struct mm_struct
*mm
)
542 return mm
->tlb_flush_pending
;
544 static inline void set_tlb_flush_pending(struct mm_struct
*mm
)
546 mm
->tlb_flush_pending
= true;
549 * Guarantee that the tlb_flush_pending store does not leak into the
550 * critical section updating the page tables
552 smp_mb__before_spinlock();
554 /* Clearing is done after a TLB flush, which also provides a barrier. */
555 static inline void clear_tlb_flush_pending(struct mm_struct
*mm
)
558 mm
->tlb_flush_pending
= false;
561 static inline bool mm_tlb_flush_pending(struct mm_struct
*mm
)
565 static inline void set_tlb_flush_pending(struct mm_struct
*mm
)
568 static inline void clear_tlb_flush_pending(struct mm_struct
*mm
)
575 struct vm_special_mapping
{
576 const char *name
; /* The name, e.g. "[vdso]". */
579 * If .fault is not provided, this points to a
580 * NULL-terminated array of pages that back the special mapping.
582 * This must not be NULL unless .fault is provided.
587 * If non-NULL, then this is called to resolve page faults
588 * on the special mapping. If used, .pages is not checked.
590 int (*fault
)(const struct vm_special_mapping
*sm
,
591 struct vm_area_struct
*vma
,
592 struct vm_fault
*vmf
);
594 int (*mremap
)(const struct vm_special_mapping
*sm
,
595 struct vm_area_struct
*new_vma
);
598 enum tlb_flush_reason
{
599 TLB_FLUSH_ON_TASK_SWITCH
,
600 TLB_REMOTE_SHOOTDOWN
,
602 TLB_LOCAL_MM_SHOOTDOWN
,
604 NR_TLB_FLUSH_REASONS
,
608 * A swap entry has to fit into a "unsigned long", as the entry is hidden
609 * in the "index" field of the swapper address space.
615 #endif /* _LINUX_MM_TYPES_H */