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>
18 #ifndef AT_VECTOR_SIZE_ARCH
19 #define AT_VECTOR_SIZE_ARCH 0
21 #define AT_VECTOR_SIZE (2*(AT_VECTOR_SIZE_ARCH + AT_VECTOR_SIZE_BASE + 1))
26 #define USE_SPLIT_PTE_PTLOCKS (NR_CPUS >= CONFIG_SPLIT_PTLOCK_CPUS)
27 #define USE_SPLIT_PMD_PTLOCKS (USE_SPLIT_PTE_PTLOCKS && \
28 IS_ENABLED(CONFIG_ARCH_ENABLE_SPLIT_PMD_PTLOCK))
29 #define ALLOC_SPLIT_PTLOCKS (SPINLOCK_SIZE > BITS_PER_LONG/8)
32 * Each physical page in the system has a struct page associated with
33 * it to keep track of whatever it is we are using the page for at the
34 * moment. Note that we have no way to track which tasks are using
35 * a page, though if it is a pagecache page, rmap structures can tell us
38 * The objects in struct page are organized in double word blocks in
39 * order to allows us to use atomic double word operations on portions
40 * of struct page. That is currently only used by slub but the arrangement
41 * allows the use of atomic double word operations on the flags/mapping
42 * and lru list pointers also.
45 /* First double word block */
46 unsigned long flags; /* Atomic flags, some possibly
47 * updated asynchronously */
49 struct address_space *mapping; /* If low bit clear, points to
50 * inode address_space, or NULL.
51 * If page mapped as anonymous
52 * memory, low bit is set, and
53 * it points to anon_vma object:
54 * see PAGE_MAPPING_ANON below.
56 void *s_mem; /* slab first object */
57 atomic_t compound_mapcount; /* first tail page */
58 /* page_deferred_list().next -- second tail page */
61 /* 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 _count separate from slub cmpxchg_double data.
77 * As the rest of the double word is protected by
78 * slab_lock but _count is not.
87 * Count of ptes mapped in mms, to show
88 * when page is mapped & limit reverse
100 atomic_t _count; /* Usage count, see below. */
102 unsigned int active; /* SLAB */
107 * Third double word block
109 * WARNING: bit 0 of the first word encode PageTail(). That means
110 * the rest users of the storage space MUST NOT use the bit to
111 * avoid collision and false-positive PageTail().
114 struct list_head lru; /* Pageout list, eg. active_list
115 * protected by zone->lru_lock !
116 * Can be used as a generic list
119 struct { /* slub per cpu partial pages */
120 struct page *next; /* Next partial slab */
122 int pages; /* Nr of partial slabs left */
123 int pobjects; /* Approximate # of objects */
130 struct rcu_head rcu_head; /* Used by SLAB
131 * when destroying via RCU
133 /* Tail pages of compound page */
135 unsigned long compound_head; /* If bit zero is set */
137 /* First tail page only */
140 * On 64 bit system we have enough space in struct page
141 * to encode compound_dtor and compound_order with
142 * unsigned int. It can help compiler generate better or
143 * smaller code on some archtectures.
145 unsigned int compound_dtor;
146 unsigned int compound_order;
148 unsigned short int compound_dtor;
149 unsigned short int compound_order;
153 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && USE_SPLIT_PMD_PTLOCKS
155 unsigned long __pad; /* do not overlay pmd_huge_pte
156 * with compound_head to avoid
157 * possible bit 0 collision.
159 pgtable_t pmd_huge_pte; /* protected by page->ptl */
164 /* Remainder is not double word aligned */
166 unsigned long private; /* Mapping-private opaque data:
167 * usually used for buffer_heads
168 * if PagePrivate set; used for
169 * swp_entry_t if PageSwapCache;
170 * indicates order in the buddy
171 * system if PG_buddy is set.
173 #if USE_SPLIT_PTE_PTLOCKS
174 #if ALLOC_SPLIT_PTLOCKS
180 struct kmem_cache *slab_cache; /* SL[AU]B: Pointer to slab */
184 struct mem_cgroup *mem_cgroup;
188 * On machines where all RAM is mapped into kernel address space,
189 * we can simply calculate the virtual address. On machines with
190 * highmem some memory is mapped into kernel virtual memory
191 * dynamically, so we need a place to store that address.
192 * Note that this field could be 16 bits on x86 ... ;)
194 * Architectures with slow multiplication can define
195 * WANT_PAGE_VIRTUAL in asm/page.h
197 #if defined(WANT_PAGE_VIRTUAL)
198 void *virtual; /* Kernel virtual address (NULL if
199 not kmapped, ie. highmem) */
200 #endif /* WANT_PAGE_VIRTUAL */
202 #ifdef CONFIG_KMEMCHECK
204 * kmemcheck wants to track the status of each byte in a page; this
205 * is a pointer to such a status block. NULL if not tracked.
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))
225 #if (BITS_PER_LONG > 32) || (PAGE_SIZE >= 65536)
234 #define PAGE_FRAG_CACHE_MAX_SIZE __ALIGN_MASK(32768, ~PAGE_MASK)
235 #define PAGE_FRAG_CACHE_MAX_ORDER get_order(PAGE_FRAG_CACHE_MAX_SIZE)
237 struct page_frag_cache {
239 #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
245 /* we maintain a pagecount bias, so that we dont dirty cache line
246 * containing page->_count every time we allocate a fragment.
248 unsigned int pagecnt_bias;
252 typedef unsigned long vm_flags_t;
255 * A region containing a mapping of a non-memory backed file under NOMMU
256 * conditions. These are held in a global tree and are pinned by the VMAs that
260 struct rb_node vm_rb; /* link in global region tree */
261 vm_flags_t vm_flags; /* VMA vm_flags */
262 unsigned long vm_start; /* start address of region */
263 unsigned long vm_end; /* region initialised to here */
264 unsigned long vm_top; /* region allocated to here */
265 unsigned long vm_pgoff; /* the offset in vm_file corresponding to vm_start */
266 struct file *vm_file; /* the backing file or NULL */
268 int vm_usage; /* region usage count (access under nommu_region_sem) */
269 bool vm_icache_flushed : 1; /* true if the icache has been flushed for
273 #ifdef CONFIG_USERFAULTFD
274 #define NULL_VM_UFFD_CTX ((struct vm_userfaultfd_ctx) { NULL, })
275 struct vm_userfaultfd_ctx {
276 struct userfaultfd_ctx *ctx;
278 #else /* CONFIG_USERFAULTFD */
279 #define NULL_VM_UFFD_CTX ((struct vm_userfaultfd_ctx) {})
280 struct vm_userfaultfd_ctx {};
281 #endif /* CONFIG_USERFAULTFD */
284 * This struct defines a memory VMM memory area. There is one of these
285 * per VM-area/task. A VM area is any part of the process virtual memory
286 * space that has a special rule for the page-fault handlers (ie a shared
287 * library, the executable area etc).
289 struct vm_area_struct {
290 /* The first cache line has the info for VMA tree walking. */
292 unsigned long vm_start; /* Our start address within vm_mm. */
293 unsigned long vm_end; /* The first byte after our end address
296 /* linked list of VM areas per task, sorted by address */
297 struct vm_area_struct *vm_next, *vm_prev;
299 struct rb_node vm_rb;
302 * Largest free memory gap in bytes to the left of this VMA.
303 * Either between this VMA and vma->vm_prev, or between one of the
304 * VMAs below us in the VMA rbtree and its ->vm_prev. This helps
305 * get_unmapped_area find a free area of the right size.
307 unsigned long rb_subtree_gap;
309 /* Second cache line starts here. */
311 struct mm_struct *vm_mm; /* The address space we belong to. */
312 pgprot_t vm_page_prot; /* Access permissions of this VMA. */
313 unsigned long vm_flags; /* Flags, see mm.h. */
316 * For areas with an address space and backing store,
317 * linkage into the address_space->i_mmap interval tree.
321 unsigned long rb_subtree_last;
325 * A file's MAP_PRIVATE vma can be in both i_mmap tree and anon_vma
326 * list, after a COW of one of the file pages. A MAP_SHARED vma
327 * can only be in the i_mmap tree. An anonymous MAP_PRIVATE, stack
328 * or brk vma (with NULL file) can only be in an anon_vma list.
330 struct list_head anon_vma_chain; /* Serialized by mmap_sem &
332 struct anon_vma *anon_vma; /* Serialized by page_table_lock */
334 /* Function pointers to deal with this struct. */
335 const struct vm_operations_struct *vm_ops;
337 /* Information about our backing store: */
338 unsigned long vm_pgoff; /* Offset (within vm_file) in PAGE_SIZE
339 units, *not* PAGE_CACHE_SIZE */
340 struct file * vm_file; /* File we map to (can be NULL). */
341 void * vm_private_data; /* was vm_pte (shared mem) */
344 struct vm_region *vm_region; /* NOMMU mapping region */
347 struct mempolicy *vm_policy; /* NUMA policy for the VMA */
349 struct vm_userfaultfd_ctx vm_userfaultfd_ctx;
353 struct task_struct *task;
354 struct core_thread *next;
359 struct core_thread dumper;
360 struct completion startup;
370 #if USE_SPLIT_PTE_PTLOCKS && defined(CONFIG_MMU)
371 #define SPLIT_RSS_COUNTING
372 /* per-thread cached information, */
373 struct task_rss_stat {
374 int events; /* for synchronization threshold */
375 int count[NR_MM_COUNTERS];
377 #endif /* USE_SPLIT_PTE_PTLOCKS */
380 atomic_long_t count[NR_MM_COUNTERS];
385 struct vm_area_struct *mmap; /* list of VMAs */
386 struct rb_root mm_rb;
387 u32 vmacache_seqnum; /* per-thread vmacache */
389 unsigned long (*get_unmapped_area) (struct file *filp,
390 unsigned long addr, unsigned long len,
391 unsigned long pgoff, unsigned long flags);
393 unsigned long mmap_base; /* base of mmap area */
394 unsigned long mmap_legacy_base; /* base of mmap area in bottom-up allocations */
395 unsigned long task_size; /* size of task vm space */
396 unsigned long highest_vm_end; /* highest vma end address */
398 atomic_t mm_users; /* How many users with user space? */
399 atomic_t mm_count; /* How many references to "struct mm_struct" (users count as 1) */
400 atomic_long_t nr_ptes; /* PTE page table pages */
401 #if CONFIG_PGTABLE_LEVELS > 2
402 atomic_long_t nr_pmds; /* PMD page table pages */
404 int map_count; /* number of VMAs */
406 spinlock_t page_table_lock; /* Protects page tables and some counters */
407 struct rw_semaphore mmap_sem;
409 struct list_head mmlist; /* List of maybe swapped mm's. These are globally strung
410 * together off init_mm.mmlist, and are protected
415 unsigned long hiwater_rss; /* High-watermark of RSS usage */
416 unsigned long hiwater_vm; /* High-water virtual memory usage */
418 unsigned long total_vm; /* Total pages mapped */
419 unsigned long locked_vm; /* Pages that have PG_mlocked set */
420 unsigned long pinned_vm; /* Refcount permanently increased */
421 unsigned long shared_vm; /* Shared pages (files) */
422 unsigned long exec_vm; /* VM_EXEC & ~VM_WRITE */
423 unsigned long stack_vm; /* VM_GROWSUP/DOWN */
424 unsigned long def_flags;
425 unsigned long start_code, end_code, start_data, end_data;
426 unsigned long start_brk, brk, start_stack;
427 unsigned long arg_start, arg_end, env_start, env_end;
429 unsigned long saved_auxv[AT_VECTOR_SIZE]; /* for /proc/PID/auxv */
432 * Special counters, in some configurations protected by the
433 * page_table_lock, in other configurations by being atomic.
435 struct mm_rss_stat rss_stat;
437 struct linux_binfmt *binfmt;
439 cpumask_var_t cpu_vm_mask_var;
441 /* Architecture-specific MM context */
442 mm_context_t context;
444 unsigned long flags; /* Must use atomic bitops to access the bits */
446 struct core_state *core_state; /* coredumping support */
448 spinlock_t ioctx_lock;
449 struct kioctx_table __rcu *ioctx_table;
453 * "owner" points to a task that is regarded as the canonical
454 * user/owner of this mm. All of the following must be true in
455 * order for it to be changed:
457 * current == mm->owner
459 * new_owner->mm == mm
460 * new_owner->alloc_lock is held
462 struct task_struct __rcu *owner;
465 /* store ref to file /proc/<pid>/exe symlink points to */
466 struct file __rcu *exe_file;
467 #ifdef CONFIG_MMU_NOTIFIER
468 struct mmu_notifier_mm *mmu_notifier_mm;
470 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
471 pgtable_t pmd_huge_pte; /* protected by page_table_lock */
473 #ifdef CONFIG_CPUMASK_OFFSTACK
474 struct cpumask cpumask_allocation;
476 #ifdef CONFIG_NUMA_BALANCING
478 * numa_next_scan is the next time that the PTEs will be marked
479 * pte_numa. NUMA hinting faults will gather statistics and migrate
480 * pages to new nodes if necessary.
482 unsigned long numa_next_scan;
484 /* Restart point for scanning and setting pte_numa */
485 unsigned long numa_scan_offset;
487 /* numa_scan_seq prevents two threads setting pte_numa */
490 #if defined(CONFIG_NUMA_BALANCING) || defined(CONFIG_COMPACTION)
492 * An operation with batched TLB flushing is going on. Anything that
493 * can move process memory needs to flush the TLB when moving a
494 * PROT_NONE or PROT_NUMA mapped page.
496 bool tlb_flush_pending;
498 struct uprobes_state uprobes_state;
499 #ifdef CONFIG_X86_INTEL_MPX
500 /* address of the bounds directory */
501 void __user *bd_addr;
503 #ifdef CONFIG_HUGETLB_PAGE
504 atomic_long_t hugetlb_usage;
508 static inline void mm_init_cpumask(struct mm_struct *mm)
510 #ifdef CONFIG_CPUMASK_OFFSTACK
511 mm->cpu_vm_mask_var = &mm->cpumask_allocation;
513 cpumask_clear(mm->cpu_vm_mask_var);
516 /* Future-safe accessor for struct mm_struct's cpu_vm_mask. */
517 static inline cpumask_t *mm_cpumask(struct mm_struct *mm)
519 return mm->cpu_vm_mask_var;
522 #if defined(CONFIG_NUMA_BALANCING) || defined(CONFIG_COMPACTION)
524 * Memory barriers to keep this state in sync are graciously provided by
525 * the page table locks, outside of which no page table modifications happen.
526 * The barriers below prevent the compiler from re-ordering the instructions
527 * around the memory barriers that are already present in the code.
529 static inline bool mm_tlb_flush_pending(struct mm_struct *mm)
532 return mm->tlb_flush_pending;
534 static inline void set_tlb_flush_pending(struct mm_struct *mm)
536 mm->tlb_flush_pending = true;
539 * Guarantee that the tlb_flush_pending store does not leak into the
540 * critical section updating the page tables
542 smp_mb__before_spinlock();
544 /* Clearing is done after a TLB flush, which also provides a barrier. */
545 static inline void clear_tlb_flush_pending(struct mm_struct *mm)
548 mm->tlb_flush_pending = false;
551 static inline bool mm_tlb_flush_pending(struct mm_struct *mm)
555 static inline void set_tlb_flush_pending(struct mm_struct *mm)
558 static inline void clear_tlb_flush_pending(struct mm_struct *mm)
563 struct vm_special_mapping
569 enum tlb_flush_reason {
570 TLB_FLUSH_ON_TASK_SWITCH,
571 TLB_REMOTE_SHOOTDOWN,
573 TLB_LOCAL_MM_SHOOTDOWN,
575 NR_TLB_FLUSH_REASONS,
579 * A swap entry has to fit into a "unsigned long", as the entry is hidden
580 * in the "index" field of the swapper address space.
586 #endif /* _LINUX_MM_TYPES_H */