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 _refcount separate from slub cmpxchg_double
77 * data. As the rest of the double word is protected by
78 * slab_lock but _refcount is not.
87 * Count of ptes mapped in mms, to show
88 * when page is mapped & limit reverse
101 * Usage count, *USE WRAPPER FUNCTION*
102 * when manual accounting. See page_ref.h
106 unsigned int active
; /* SLAB */
111 * Third double word block
113 * WARNING: bit 0 of the first word encode PageTail(). That means
114 * the rest users of the storage space MUST NOT use the bit to
115 * avoid collision and false-positive PageTail().
118 struct list_head lru
; /* Pageout list, eg. active_list
119 * protected by zone->lru_lock !
120 * Can be used as a generic list
123 struct dev_pagemap
*pgmap
; /* ZONE_DEVICE pages are never on an
124 * lru or handled by a slab
125 * allocator, this points to the
126 * hosting device page map.
128 struct { /* slub per cpu partial pages */
129 struct page
*next
; /* Next partial slab */
131 int pages
; /* Nr of partial slabs left */
132 int pobjects
; /* Approximate # of objects */
139 struct rcu_head rcu_head
; /* Used by SLAB
140 * when destroying via RCU
142 /* Tail pages of compound page */
144 unsigned long compound_head
; /* If bit zero is set */
146 /* First tail page only */
149 * On 64 bit system we have enough space in struct page
150 * to encode compound_dtor and compound_order with
151 * unsigned int. It can help compiler generate better or
152 * smaller code on some archtectures.
154 unsigned int compound_dtor
;
155 unsigned int compound_order
;
157 unsigned short int compound_dtor
;
158 unsigned short int compound_order
;
162 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && USE_SPLIT_PMD_PTLOCKS
164 unsigned long __pad
; /* do not overlay pmd_huge_pte
165 * with compound_head to avoid
166 * possible bit 0 collision.
168 pgtable_t pmd_huge_pte
; /* protected by page->ptl */
173 /* Remainder is not double word aligned */
175 unsigned long private; /* Mapping-private opaque data:
176 * usually used for buffer_heads
177 * if PagePrivate set; used for
178 * swp_entry_t if PageSwapCache;
179 * indicates order in the buddy
180 * system if PG_buddy is set.
182 #if USE_SPLIT_PTE_PTLOCKS
183 #if ALLOC_SPLIT_PTLOCKS
189 struct kmem_cache
*slab_cache
; /* SL[AU]B: Pointer to slab */
193 struct mem_cgroup
*mem_cgroup
;
197 * On machines where all RAM is mapped into kernel address space,
198 * we can simply calculate the virtual address. On machines with
199 * highmem some memory is mapped into kernel virtual memory
200 * dynamically, so we need a place to store that address.
201 * Note that this field could be 16 bits on x86 ... ;)
203 * Architectures with slow multiplication can define
204 * WANT_PAGE_VIRTUAL in asm/page.h
206 #if defined(WANT_PAGE_VIRTUAL)
207 void *virtual; /* Kernel virtual address (NULL if
208 not kmapped, ie. highmem) */
209 #endif /* WANT_PAGE_VIRTUAL */
211 #ifdef CONFIG_KMEMCHECK
213 * kmemcheck wants to track the status of each byte in a page; this
214 * is a pointer to such a status block. NULL if not tracked.
219 #ifdef LAST_CPUPID_NOT_IN_PAGE_FLAGS
224 * The struct page can be forced to be double word aligned so that atomic ops
225 * on double words work. The SLUB allocator can make use of such a feature.
227 #ifdef CONFIG_HAVE_ALIGNED_STRUCT_PAGE
228 __aligned(2 * sizeof(unsigned long))
234 #if (BITS_PER_LONG > 32) || (PAGE_SIZE >= 65536)
243 #define PAGE_FRAG_CACHE_MAX_SIZE __ALIGN_MASK(32768, ~PAGE_MASK)
244 #define PAGE_FRAG_CACHE_MAX_ORDER get_order(PAGE_FRAG_CACHE_MAX_SIZE)
246 struct page_frag_cache
{
248 #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
254 /* we maintain a pagecount bias, so that we dont dirty cache line
255 * containing page->_refcount every time we allocate a fragment.
257 unsigned int pagecnt_bias
;
261 typedef unsigned long vm_flags_t
;
264 * A region containing a mapping of a non-memory backed file under NOMMU
265 * conditions. These are held in a global tree and are pinned by the VMAs that
269 struct rb_node vm_rb
; /* link in global region tree */
270 vm_flags_t vm_flags
; /* VMA vm_flags */
271 unsigned long vm_start
; /* start address of region */
272 unsigned long vm_end
; /* region initialised to here */
273 unsigned long vm_top
; /* region allocated to here */
274 unsigned long vm_pgoff
; /* the offset in vm_file corresponding to vm_start */
275 struct file
*vm_file
; /* the backing file or NULL */
277 int vm_usage
; /* region usage count (access under nommu_region_sem) */
278 bool vm_icache_flushed
: 1; /* true if the icache has been flushed for
282 #ifdef CONFIG_USERFAULTFD
283 #define NULL_VM_UFFD_CTX ((struct vm_userfaultfd_ctx) { NULL, })
284 struct vm_userfaultfd_ctx
{
285 struct userfaultfd_ctx
*ctx
;
287 #else /* CONFIG_USERFAULTFD */
288 #define NULL_VM_UFFD_CTX ((struct vm_userfaultfd_ctx) {})
289 struct vm_userfaultfd_ctx
{};
290 #endif /* CONFIG_USERFAULTFD */
293 * This struct defines a memory VMM memory area. There is one of these
294 * per VM-area/task. A VM area is any part of the process virtual memory
295 * space that has a special rule for the page-fault handlers (ie a shared
296 * library, the executable area etc).
298 struct vm_area_struct
{
299 /* The first cache line has the info for VMA tree walking. */
301 unsigned long vm_start
; /* Our start address within vm_mm. */
302 unsigned long vm_end
; /* The first byte after our end address
305 /* linked list of VM areas per task, sorted by address */
306 struct vm_area_struct
*vm_next
, *vm_prev
;
308 struct rb_node vm_rb
;
311 * Largest free memory gap in bytes to the left of this VMA.
312 * Either between this VMA and vma->vm_prev, or between one of the
313 * VMAs below us in the VMA rbtree and its ->vm_prev. This helps
314 * get_unmapped_area find a free area of the right size.
316 unsigned long rb_subtree_gap
;
318 /* Second cache line starts here. */
320 struct mm_struct
*vm_mm
; /* The address space we belong to. */
321 pgprot_t vm_page_prot
; /* Access permissions of this VMA. */
322 unsigned long vm_flags
; /* Flags, see mm.h. */
325 * For areas with an address space and backing store,
326 * linkage into the address_space->i_mmap interval tree.
330 unsigned long rb_subtree_last
;
334 * A file's MAP_PRIVATE vma can be in both i_mmap tree and anon_vma
335 * list, after a COW of one of the file pages. A MAP_SHARED vma
336 * can only be in the i_mmap tree. An anonymous MAP_PRIVATE, stack
337 * or brk vma (with NULL file) can only be in an anon_vma list.
339 struct list_head anon_vma_chain
; /* Serialized by mmap_sem &
341 struct anon_vma
*anon_vma
; /* Serialized by page_table_lock */
343 /* Function pointers to deal with this struct. */
344 const struct vm_operations_struct
*vm_ops
;
346 /* Information about our backing store: */
347 unsigned long vm_pgoff
; /* Offset (within vm_file) in PAGE_SIZE
349 struct file
* vm_file
; /* File we map to (can be NULL). */
350 void * vm_private_data
; /* was vm_pte (shared mem) */
353 struct vm_region
*vm_region
; /* NOMMU mapping region */
356 struct mempolicy
*vm_policy
; /* NUMA policy for the VMA */
358 struct vm_userfaultfd_ctx vm_userfaultfd_ctx
;
362 struct task_struct
*task
;
363 struct core_thread
*next
;
368 struct core_thread dumper
;
369 struct completion startup
;
373 MM_FILEPAGES
, /* Resident file mapping pages */
374 MM_ANONPAGES
, /* Resident anonymous pages */
375 MM_SWAPENTS
, /* Anonymous swap entries */
376 MM_SHMEMPAGES
, /* Resident shared memory pages */
380 #if USE_SPLIT_PTE_PTLOCKS && defined(CONFIG_MMU)
381 #define SPLIT_RSS_COUNTING
382 /* per-thread cached information, */
383 struct task_rss_stat
{
384 int events
; /* for synchronization threshold */
385 int count
[NR_MM_COUNTERS
];
387 #endif /* USE_SPLIT_PTE_PTLOCKS */
390 atomic_long_t count
[NR_MM_COUNTERS
];
395 struct vm_area_struct
*mmap
; /* list of VMAs */
396 struct rb_root mm_rb
;
397 u32 vmacache_seqnum
; /* per-thread vmacache */
399 unsigned long (*get_unmapped_area
) (struct file
*filp
,
400 unsigned long addr
, unsigned long len
,
401 unsigned long pgoff
, unsigned long flags
);
403 unsigned long mmap_base
; /* base of mmap area */
404 unsigned long mmap_legacy_base
; /* base of mmap area in bottom-up allocations */
405 unsigned long task_size
; /* size of task vm space */
406 unsigned long highest_vm_end
; /* highest vma end address */
408 atomic_t mm_users
; /* How many users with user space? */
409 atomic_t mm_count
; /* How many references to "struct mm_struct" (users count as 1) */
410 atomic_long_t nr_ptes
; /* PTE page table pages */
411 #if CONFIG_PGTABLE_LEVELS > 2
412 atomic_long_t nr_pmds
; /* PMD page table pages */
414 int map_count
; /* number of VMAs */
416 spinlock_t page_table_lock
; /* Protects page tables and some counters */
417 struct rw_semaphore mmap_sem
;
419 struct list_head mmlist
; /* List of maybe swapped mm's. These are globally strung
420 * together off init_mm.mmlist, and are protected
425 unsigned long hiwater_rss
; /* High-watermark of RSS usage */
426 unsigned long hiwater_vm
; /* High-water virtual memory usage */
428 unsigned long total_vm
; /* Total pages mapped */
429 unsigned long locked_vm
; /* Pages that have PG_mlocked set */
430 unsigned long pinned_vm
; /* Refcount permanently increased */
431 unsigned long data_vm
; /* VM_WRITE & ~VM_SHARED & ~VM_STACK */
432 unsigned long exec_vm
; /* VM_EXEC & ~VM_WRITE & ~VM_STACK */
433 unsigned long stack_vm
; /* VM_STACK */
434 unsigned long def_flags
;
435 unsigned long start_code
, end_code
, start_data
, end_data
;
436 unsigned long start_brk
, brk
, start_stack
;
437 unsigned long arg_start
, arg_end
, env_start
, env_end
;
439 unsigned long saved_auxv
[AT_VECTOR_SIZE
]; /* for /proc/PID/auxv */
442 * Special counters, in some configurations protected by the
443 * page_table_lock, in other configurations by being atomic.
445 struct mm_rss_stat rss_stat
;
447 struct linux_binfmt
*binfmt
;
449 cpumask_var_t cpu_vm_mask_var
;
451 /* Architecture-specific MM context */
452 mm_context_t context
;
454 unsigned long flags
; /* Must use atomic bitops to access the bits */
456 struct core_state
*core_state
; /* coredumping support */
458 spinlock_t ioctx_lock
;
459 struct kioctx_table __rcu
*ioctx_table
;
463 * "owner" points to a task that is regarded as the canonical
464 * user/owner of this mm. All of the following must be true in
465 * order for it to be changed:
467 * current == mm->owner
469 * new_owner->mm == mm
470 * new_owner->alloc_lock is held
472 struct task_struct __rcu
*owner
;
475 /* store ref to file /proc/<pid>/exe symlink points to */
476 struct file __rcu
*exe_file
;
477 #ifdef CONFIG_MMU_NOTIFIER
478 struct mmu_notifier_mm
*mmu_notifier_mm
;
480 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
481 pgtable_t pmd_huge_pte
; /* protected by page_table_lock */
483 #ifdef CONFIG_CPUMASK_OFFSTACK
484 struct cpumask cpumask_allocation
;
486 #ifdef CONFIG_NUMA_BALANCING
488 * numa_next_scan is the next time that the PTEs will be marked
489 * pte_numa. NUMA hinting faults will gather statistics and migrate
490 * pages to new nodes if necessary.
492 unsigned long numa_next_scan
;
494 /* Restart point for scanning and setting pte_numa */
495 unsigned long numa_scan_offset
;
497 /* numa_scan_seq prevents two threads setting pte_numa */
500 #if defined(CONFIG_NUMA_BALANCING) || defined(CONFIG_COMPACTION)
502 * An operation with batched TLB flushing is going on. Anything that
503 * can move process memory needs to flush the TLB when moving a
504 * PROT_NONE or PROT_NUMA mapped page.
506 bool tlb_flush_pending
;
508 struct uprobes_state uprobes_state
;
509 #ifdef CONFIG_X86_INTEL_MPX
510 /* address of the bounds directory */
511 void __user
*bd_addr
;
513 #ifdef CONFIG_HUGETLB_PAGE
514 atomic_long_t hugetlb_usage
;
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
);
595 enum tlb_flush_reason
{
596 TLB_FLUSH_ON_TASK_SWITCH
,
597 TLB_REMOTE_SHOOTDOWN
,
599 TLB_LOCAL_MM_SHOOTDOWN
,
601 NR_TLB_FLUSH_REASONS
,
605 * A swap entry has to fit into a "unsigned long", as the entry is hidden
606 * in the "index" field of the swapper address space.
612 #endif /* _LINUX_MM_TYPES_H */