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 */
59 /* Second double word */
62 pgoff_t index
; /* Our offset within mapping. */
63 void *freelist
; /* sl[aou]b first free object */
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 _count separate from slub cmpxchg_double data.
74 * As the rest of the double word is protected by
75 * slab_lock but _count is not.
84 * Count of ptes mapped in
85 * mms, to show when page is
86 * mapped & limit reverse map
89 * Used also for tail pages
90 * refcounting instead of
91 * _count. Tail pages cannot
92 * be mapped and keeping the
93 * tail page _count zero at
94 * all times guarantees
95 * get_page_unless_zero() will
96 * never succeed on tail
106 int units
; /* SLOB */
108 atomic_t _count
; /* Usage count, see below. */
110 unsigned int active
; /* SLAB */
115 * Third double word block
117 * WARNING: bit 0 of the first word encode PageTail(). That means
118 * the rest users of the storage space MUST NOT use the bit to
119 * avoid collision and false-positive PageTail().
122 struct list_head lru
; /* Pageout list, eg. active_list
123 * protected by zone->lru_lock !
124 * Can be used as a generic list
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 CONFIG_KMEMCHECK
212 * kmemcheck wants to track the status of each byte in a page; this
213 * is a pointer to such a status block. NULL if not tracked.
218 #ifdef LAST_CPUPID_NOT_IN_PAGE_FLAGS
223 * The struct page can be forced to be double word aligned so that atomic ops
224 * on double words work. The SLUB allocator can make use of such a feature.
226 #ifdef CONFIG_HAVE_ALIGNED_STRUCT_PAGE
227 __aligned(2 * sizeof(unsigned long))
233 #if (BITS_PER_LONG > 32) || (PAGE_SIZE >= 65536)
242 #define PAGE_FRAG_CACHE_MAX_SIZE __ALIGN_MASK(32768, ~PAGE_MASK)
243 #define PAGE_FRAG_CACHE_MAX_ORDER get_order(PAGE_FRAG_CACHE_MAX_SIZE)
245 struct page_frag_cache
{
247 #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
253 /* we maintain a pagecount bias, so that we dont dirty cache line
254 * containing page->_count every time we allocate a fragment.
256 unsigned int pagecnt_bias
;
260 typedef unsigned long vm_flags_t
;
263 * A region containing a mapping of a non-memory backed file under NOMMU
264 * conditions. These are held in a global tree and are pinned by the VMAs that
268 struct rb_node vm_rb
; /* link in global region tree */
269 vm_flags_t vm_flags
; /* VMA vm_flags */
270 unsigned long vm_start
; /* start address of region */
271 unsigned long vm_end
; /* region initialised to here */
272 unsigned long vm_top
; /* region allocated to here */
273 unsigned long vm_pgoff
; /* the offset in vm_file corresponding to vm_start */
274 struct file
*vm_file
; /* the backing file or NULL */
276 int vm_usage
; /* region usage count (access under nommu_region_sem) */
277 bool vm_icache_flushed
: 1; /* true if the icache has been flushed for
281 #ifdef CONFIG_USERFAULTFD
282 #define NULL_VM_UFFD_CTX ((struct vm_userfaultfd_ctx) { NULL, })
283 struct vm_userfaultfd_ctx
{
284 struct userfaultfd_ctx
*ctx
;
286 #else /* CONFIG_USERFAULTFD */
287 #define NULL_VM_UFFD_CTX ((struct vm_userfaultfd_ctx) {})
288 struct vm_userfaultfd_ctx
{};
289 #endif /* CONFIG_USERFAULTFD */
292 * This struct defines a memory VMM memory area. There is one of these
293 * per VM-area/task. A VM area is any part of the process virtual memory
294 * space that has a special rule for the page-fault handlers (ie a shared
295 * library, the executable area etc).
297 struct vm_area_struct
{
298 /* The first cache line has the info for VMA tree walking. */
300 unsigned long vm_start
; /* Our start address within vm_mm. */
301 unsigned long vm_end
; /* The first byte after our end address
304 /* linked list of VM areas per task, sorted by address */
305 struct vm_area_struct
*vm_next
, *vm_prev
;
307 struct rb_node vm_rb
;
310 * Largest free memory gap in bytes to the left of this VMA.
311 * Either between this VMA and vma->vm_prev, or between one of the
312 * VMAs below us in the VMA rbtree and its ->vm_prev. This helps
313 * get_unmapped_area find a free area of the right size.
315 unsigned long rb_subtree_gap
;
317 /* Second cache line starts here. */
319 struct mm_struct
*vm_mm
; /* The address space we belong to. */
320 pgprot_t vm_page_prot
; /* Access permissions of this VMA. */
321 unsigned long vm_flags
; /* Flags, see mm.h. */
324 * For areas with an address space and backing store,
325 * linkage into the address_space->i_mmap interval tree.
329 unsigned long rb_subtree_last
;
333 * A file's MAP_PRIVATE vma can be in both i_mmap tree and anon_vma
334 * list, after a COW of one of the file pages. A MAP_SHARED vma
335 * can only be in the i_mmap tree. An anonymous MAP_PRIVATE, stack
336 * or brk vma (with NULL file) can only be in an anon_vma list.
338 struct list_head anon_vma_chain
; /* Serialized by mmap_sem &
340 struct anon_vma
*anon_vma
; /* Serialized by page_table_lock */
342 /* Function pointers to deal with this struct. */
343 const struct vm_operations_struct
*vm_ops
;
345 /* Information about our backing store: */
346 unsigned long vm_pgoff
; /* Offset (within vm_file) in PAGE_SIZE
347 units, *not* PAGE_CACHE_SIZE */
348 struct file
* vm_file
; /* File we map to (can be NULL). */
349 void * vm_private_data
; /* was vm_pte (shared mem) */
352 struct vm_region
*vm_region
; /* NOMMU mapping region */
355 struct mempolicy
*vm_policy
; /* NUMA policy for the VMA */
357 struct vm_userfaultfd_ctx vm_userfaultfd_ctx
;
361 struct task_struct
*task
;
362 struct core_thread
*next
;
367 struct core_thread dumper
;
368 struct completion startup
;
372 MM_FILEPAGES
, /* Resident file mapping pages */
373 MM_ANONPAGES
, /* Resident anonymous pages */
374 MM_SWAPENTS
, /* Anonymous swap entries */
375 MM_SHMEMPAGES
, /* Resident shared memory pages */
379 #if USE_SPLIT_PTE_PTLOCKS && defined(CONFIG_MMU)
380 #define SPLIT_RSS_COUNTING
381 /* per-thread cached information, */
382 struct task_rss_stat
{
383 int events
; /* for synchronization threshold */
384 int count
[NR_MM_COUNTERS
];
386 #endif /* USE_SPLIT_PTE_PTLOCKS */
389 atomic_long_t count
[NR_MM_COUNTERS
];
394 struct vm_area_struct
*mmap
; /* list of VMAs */
395 struct rb_root mm_rb
;
396 u32 vmacache_seqnum
; /* per-thread vmacache */
398 unsigned long (*get_unmapped_area
) (struct file
*filp
,
399 unsigned long addr
, unsigned long len
,
400 unsigned long pgoff
, unsigned long flags
);
402 unsigned long mmap_base
; /* base of mmap area */
403 unsigned long mmap_legacy_base
; /* base of mmap area in bottom-up allocations */
404 unsigned long task_size
; /* size of task vm space */
405 unsigned long highest_vm_end
; /* highest vma end address */
407 atomic_t mm_users
; /* How many users with user space? */
408 atomic_t mm_count
; /* How many references to "struct mm_struct" (users count as 1) */
409 atomic_long_t nr_ptes
; /* PTE page table pages */
410 #if CONFIG_PGTABLE_LEVELS > 2
411 atomic_long_t nr_pmds
; /* PMD page table pages */
413 int map_count
; /* number of VMAs */
415 spinlock_t page_table_lock
; /* Protects page tables and some counters */
416 struct rw_semaphore mmap_sem
;
418 struct list_head mmlist
; /* List of maybe swapped mm's. These are globally strung
419 * together off init_mm.mmlist, and are protected
424 unsigned long hiwater_rss
; /* High-watermark of RSS usage */
425 unsigned long hiwater_vm
; /* High-water virtual memory usage */
427 unsigned long total_vm
; /* Total pages mapped */
428 unsigned long locked_vm
; /* Pages that have PG_mlocked set */
429 unsigned long pinned_vm
; /* Refcount permanently increased */
430 unsigned long data_vm
; /* VM_WRITE & ~VM_SHARED/GROWSDOWN */
431 unsigned long exec_vm
; /* VM_EXEC & ~VM_WRITE */
432 unsigned long stack_vm
; /* VM_GROWSUP/DOWN */
433 unsigned long def_flags
;
434 unsigned long start_code
, end_code
, start_data
, end_data
;
435 unsigned long start_brk
, brk
, start_stack
;
436 unsigned long arg_start
, arg_end
, env_start
, env_end
;
438 unsigned long saved_auxv
[AT_VECTOR_SIZE
]; /* for /proc/PID/auxv */
441 * Special counters, in some configurations protected by the
442 * page_table_lock, in other configurations by being atomic.
444 struct mm_rss_stat rss_stat
;
446 struct linux_binfmt
*binfmt
;
448 cpumask_var_t cpu_vm_mask_var
;
450 /* Architecture-specific MM context */
451 mm_context_t context
;
453 unsigned long flags
; /* Must use atomic bitops to access the bits */
455 struct core_state
*core_state
; /* coredumping support */
457 spinlock_t ioctx_lock
;
458 struct kioctx_table __rcu
*ioctx_table
;
462 * "owner" points to a task that is regarded as the canonical
463 * user/owner of this mm. All of the following must be true in
464 * order for it to be changed:
466 * current == mm->owner
468 * new_owner->mm == mm
469 * new_owner->alloc_lock is held
471 struct task_struct __rcu
*owner
;
474 /* store ref to file /proc/<pid>/exe symlink points to */
475 struct file __rcu
*exe_file
;
476 #ifdef CONFIG_MMU_NOTIFIER
477 struct mmu_notifier_mm
*mmu_notifier_mm
;
479 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
480 pgtable_t pmd_huge_pte
; /* protected by page_table_lock */
482 #ifdef CONFIG_CPUMASK_OFFSTACK
483 struct cpumask cpumask_allocation
;
485 #ifdef CONFIG_NUMA_BALANCING
487 * numa_next_scan is the next time that the PTEs will be marked
488 * pte_numa. NUMA hinting faults will gather statistics and migrate
489 * pages to new nodes if necessary.
491 unsigned long numa_next_scan
;
493 /* Restart point for scanning and setting pte_numa */
494 unsigned long numa_scan_offset
;
496 /* numa_scan_seq prevents two threads setting pte_numa */
499 #if defined(CONFIG_NUMA_BALANCING) || defined(CONFIG_COMPACTION)
501 * An operation with batched TLB flushing is going on. Anything that
502 * can move process memory needs to flush the TLB when moving a
503 * PROT_NONE or PROT_NUMA mapped page.
505 bool tlb_flush_pending
;
507 struct uprobes_state uprobes_state
;
508 #ifdef CONFIG_X86_INTEL_MPX
509 /* address of the bounds directory */
510 void __user
*bd_addr
;
512 #ifdef CONFIG_HUGETLB_PAGE
513 atomic_long_t hugetlb_usage
;
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
;
531 #if defined(CONFIG_NUMA_BALANCING) || defined(CONFIG_COMPACTION)
533 * Memory barriers to keep this state in sync are graciously provided by
534 * the page table locks, outside of which no page table modifications happen.
535 * The barriers below prevent the compiler from re-ordering the instructions
536 * around the memory barriers that are already present in the code.
538 static inline bool mm_tlb_flush_pending(struct mm_struct
*mm
)
541 return mm
->tlb_flush_pending
;
543 static inline void set_tlb_flush_pending(struct mm_struct
*mm
)
545 mm
->tlb_flush_pending
= true;
548 * Guarantee that the tlb_flush_pending store does not leak into the
549 * critical section updating the page tables
551 smp_mb__before_spinlock();
553 /* Clearing is done after a TLB flush, which also provides a barrier. */
554 static inline void clear_tlb_flush_pending(struct mm_struct
*mm
)
557 mm
->tlb_flush_pending
= false;
560 static inline bool mm_tlb_flush_pending(struct mm_struct
*mm
)
564 static inline void set_tlb_flush_pending(struct mm_struct
*mm
)
567 static inline void clear_tlb_flush_pending(struct mm_struct
*mm
)
572 struct vm_special_mapping
578 enum tlb_flush_reason
{
579 TLB_FLUSH_ON_TASK_SWITCH
,
580 TLB_REMOTE_SHOOTDOWN
,
582 TLB_LOCAL_MM_SHOOTDOWN
,
584 NR_TLB_FLUSH_REASONS
,
588 * A swap entry has to fit into a "unsigned long", as the entry is hidden
589 * in the "index" field of the swapper address space.
595 #endif /* _LINUX_MM_TYPES_H */