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 mms, to show
85 * when page is mapped & limit reverse
97 atomic_t _count
; /* Usage count, see below. */
99 unsigned int active
; /* SLAB */
104 * Third double word block
106 * WARNING: bit 0 of the first word encode PageTail(). That means
107 * the rest users of the storage space MUST NOT use the bit to
108 * avoid collision and false-positive PageTail().
111 struct list_head lru
; /* Pageout list, eg. active_list
112 * protected by zone->lru_lock !
113 * Can be used as a generic list
116 struct { /* slub per cpu partial pages */
117 struct page
*next
; /* Next partial slab */
119 int pages
; /* Nr of partial slabs left */
120 int pobjects
; /* Approximate # of objects */
127 struct rcu_head rcu_head
; /* Used by SLAB
128 * when destroying via RCU
130 /* Tail pages of compound page */
132 unsigned long compound_head
; /* If bit zero is set */
134 /* First tail page only */
137 * On 64 bit system we have enough space in struct page
138 * to encode compound_dtor and compound_order with
139 * unsigned int. It can help compiler generate better or
140 * smaller code on some archtectures.
142 unsigned int compound_dtor
;
143 unsigned int compound_order
;
145 unsigned short int compound_dtor
;
146 unsigned short int compound_order
;
150 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && USE_SPLIT_PMD_PTLOCKS
152 unsigned long __pad
; /* do not overlay pmd_huge_pte
153 * with compound_head to avoid
154 * possible bit 0 collision.
156 pgtable_t pmd_huge_pte
; /* protected by page->ptl */
161 /* Remainder is not double word aligned */
163 unsigned long private; /* Mapping-private opaque data:
164 * usually used for buffer_heads
165 * if PagePrivate set; used for
166 * swp_entry_t if PageSwapCache;
167 * indicates order in the buddy
168 * system if PG_buddy is set.
170 #if USE_SPLIT_PTE_PTLOCKS
171 #if ALLOC_SPLIT_PTLOCKS
177 struct kmem_cache
*slab_cache
; /* SL[AU]B: Pointer to slab */
181 struct mem_cgroup
*mem_cgroup
;
185 * On machines where all RAM is mapped into kernel address space,
186 * we can simply calculate the virtual address. On machines with
187 * highmem some memory is mapped into kernel virtual memory
188 * dynamically, so we need a place to store that address.
189 * Note that this field could be 16 bits on x86 ... ;)
191 * Architectures with slow multiplication can define
192 * WANT_PAGE_VIRTUAL in asm/page.h
194 #if defined(WANT_PAGE_VIRTUAL)
195 void *virtual; /* Kernel virtual address (NULL if
196 not kmapped, ie. highmem) */
197 #endif /* WANT_PAGE_VIRTUAL */
199 #ifdef CONFIG_KMEMCHECK
201 * kmemcheck wants to track the status of each byte in a page; this
202 * is a pointer to such a status block. NULL if not tracked.
207 #ifdef LAST_CPUPID_NOT_IN_PAGE_FLAGS
212 * The struct page can be forced to be double word aligned so that atomic ops
213 * on double words work. The SLUB allocator can make use of such a feature.
215 #ifdef CONFIG_HAVE_ALIGNED_STRUCT_PAGE
216 __aligned(2 * sizeof(unsigned long))
222 #if (BITS_PER_LONG > 32) || (PAGE_SIZE >= 65536)
231 #define PAGE_FRAG_CACHE_MAX_SIZE __ALIGN_MASK(32768, ~PAGE_MASK)
232 #define PAGE_FRAG_CACHE_MAX_ORDER get_order(PAGE_FRAG_CACHE_MAX_SIZE)
234 struct page_frag_cache
{
236 #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
242 /* we maintain a pagecount bias, so that we dont dirty cache line
243 * containing page->_count every time we allocate a fragment.
245 unsigned int pagecnt_bias
;
249 typedef unsigned long vm_flags_t
;
252 * A region containing a mapping of a non-memory backed file under NOMMU
253 * conditions. These are held in a global tree and are pinned by the VMAs that
257 struct rb_node vm_rb
; /* link in global region tree */
258 vm_flags_t vm_flags
; /* VMA vm_flags */
259 unsigned long vm_start
; /* start address of region */
260 unsigned long vm_end
; /* region initialised to here */
261 unsigned long vm_top
; /* region allocated to here */
262 unsigned long vm_pgoff
; /* the offset in vm_file corresponding to vm_start */
263 struct file
*vm_file
; /* the backing file or NULL */
265 int vm_usage
; /* region usage count (access under nommu_region_sem) */
266 bool vm_icache_flushed
: 1; /* true if the icache has been flushed for
270 #ifdef CONFIG_USERFAULTFD
271 #define NULL_VM_UFFD_CTX ((struct vm_userfaultfd_ctx) { NULL, })
272 struct vm_userfaultfd_ctx
{
273 struct userfaultfd_ctx
*ctx
;
275 #else /* CONFIG_USERFAULTFD */
276 #define NULL_VM_UFFD_CTX ((struct vm_userfaultfd_ctx) {})
277 struct vm_userfaultfd_ctx
{};
278 #endif /* CONFIG_USERFAULTFD */
281 * This struct defines a memory VMM memory area. There is one of these
282 * per VM-area/task. A VM area is any part of the process virtual memory
283 * space that has a special rule for the page-fault handlers (ie a shared
284 * library, the executable area etc).
286 struct vm_area_struct
{
287 /* The first cache line has the info for VMA tree walking. */
289 unsigned long vm_start
; /* Our start address within vm_mm. */
290 unsigned long vm_end
; /* The first byte after our end address
293 /* linked list of VM areas per task, sorted by address */
294 struct vm_area_struct
*vm_next
, *vm_prev
;
296 struct rb_node vm_rb
;
299 * Largest free memory gap in bytes to the left of this VMA.
300 * Either between this VMA and vma->vm_prev, or between one of the
301 * VMAs below us in the VMA rbtree and its ->vm_prev. This helps
302 * get_unmapped_area find a free area of the right size.
304 unsigned long rb_subtree_gap
;
306 /* Second cache line starts here. */
308 struct mm_struct
*vm_mm
; /* The address space we belong to. */
309 pgprot_t vm_page_prot
; /* Access permissions of this VMA. */
310 unsigned long vm_flags
; /* Flags, see mm.h. */
313 * For areas with an address space and backing store,
314 * linkage into the address_space->i_mmap interval tree.
318 unsigned long rb_subtree_last
;
322 * A file's MAP_PRIVATE vma can be in both i_mmap tree and anon_vma
323 * list, after a COW of one of the file pages. A MAP_SHARED vma
324 * can only be in the i_mmap tree. An anonymous MAP_PRIVATE, stack
325 * or brk vma (with NULL file) can only be in an anon_vma list.
327 struct list_head anon_vma_chain
; /* Serialized by mmap_sem &
329 struct anon_vma
*anon_vma
; /* Serialized by page_table_lock */
331 /* Function pointers to deal with this struct. */
332 const struct vm_operations_struct
*vm_ops
;
334 /* Information about our backing store: */
335 unsigned long vm_pgoff
; /* Offset (within vm_file) in PAGE_SIZE
336 units, *not* PAGE_CACHE_SIZE */
337 struct file
* vm_file
; /* File we map to (can be NULL). */
338 void * vm_private_data
; /* was vm_pte (shared mem) */
341 struct vm_region
*vm_region
; /* NOMMU mapping region */
344 struct mempolicy
*vm_policy
; /* NUMA policy for the VMA */
346 struct vm_userfaultfd_ctx vm_userfaultfd_ctx
;
350 struct task_struct
*task
;
351 struct core_thread
*next
;
356 struct core_thread dumper
;
357 struct completion startup
;
361 MM_FILEPAGES
, /* Resident file mapping pages */
362 MM_ANONPAGES
, /* Resident anonymous pages */
363 MM_SWAPENTS
, /* Anonymous swap entries */
364 MM_SHMEMPAGES
, /* Resident shared memory pages */
368 #if USE_SPLIT_PTE_PTLOCKS && defined(CONFIG_MMU)
369 #define SPLIT_RSS_COUNTING
370 /* per-thread cached information, */
371 struct task_rss_stat
{
372 int events
; /* for synchronization threshold */
373 int count
[NR_MM_COUNTERS
];
375 #endif /* USE_SPLIT_PTE_PTLOCKS */
378 atomic_long_t count
[NR_MM_COUNTERS
];
383 struct vm_area_struct
*mmap
; /* list of VMAs */
384 struct rb_root mm_rb
;
385 u32 vmacache_seqnum
; /* per-thread vmacache */
387 unsigned long (*get_unmapped_area
) (struct file
*filp
,
388 unsigned long addr
, unsigned long len
,
389 unsigned long pgoff
, unsigned long flags
);
391 unsigned long mmap_base
; /* base of mmap area */
392 unsigned long mmap_legacy_base
; /* base of mmap area in bottom-up allocations */
393 unsigned long task_size
; /* size of task vm space */
394 unsigned long highest_vm_end
; /* highest vma end address */
396 atomic_t mm_users
; /* How many users with user space? */
397 atomic_t mm_count
; /* How many references to "struct mm_struct" (users count as 1) */
398 atomic_long_t nr_ptes
; /* PTE page table pages */
399 #if CONFIG_PGTABLE_LEVELS > 2
400 atomic_long_t nr_pmds
; /* PMD page table pages */
402 int map_count
; /* number of VMAs */
404 spinlock_t page_table_lock
; /* Protects page tables and some counters */
405 struct rw_semaphore mmap_sem
;
407 struct list_head mmlist
; /* List of maybe swapped mm's. These are globally strung
408 * together off init_mm.mmlist, and are protected
413 unsigned long hiwater_rss
; /* High-watermark of RSS usage */
414 unsigned long hiwater_vm
; /* High-water virtual memory usage */
416 unsigned long total_vm
; /* Total pages mapped */
417 unsigned long locked_vm
; /* Pages that have PG_mlocked set */
418 unsigned long pinned_vm
; /* Refcount permanently increased */
419 unsigned long data_vm
; /* VM_WRITE & ~VM_SHARED/GROWSDOWN */
420 unsigned long exec_vm
; /* VM_EXEC & ~VM_WRITE */
421 unsigned long stack_vm
; /* VM_GROWSUP/DOWN */
422 unsigned long def_flags
;
423 unsigned long start_code
, end_code
, start_data
, end_data
;
424 unsigned long start_brk
, brk
, start_stack
;
425 unsigned long arg_start
, arg_end
, env_start
, env_end
;
427 unsigned long saved_auxv
[AT_VECTOR_SIZE
]; /* for /proc/PID/auxv */
430 * Special counters, in some configurations protected by the
431 * page_table_lock, in other configurations by being atomic.
433 struct mm_rss_stat rss_stat
;
435 struct linux_binfmt
*binfmt
;
437 cpumask_var_t cpu_vm_mask_var
;
439 /* Architecture-specific MM context */
440 mm_context_t context
;
442 unsigned long flags
; /* Must use atomic bitops to access the bits */
444 struct core_state
*core_state
; /* coredumping support */
446 spinlock_t ioctx_lock
;
447 struct kioctx_table __rcu
*ioctx_table
;
451 * "owner" points to a task that is regarded as the canonical
452 * user/owner of this mm. All of the following must be true in
453 * order for it to be changed:
455 * current == mm->owner
457 * new_owner->mm == mm
458 * new_owner->alloc_lock is held
460 struct task_struct __rcu
*owner
;
463 /* store ref to file /proc/<pid>/exe symlink points to */
464 struct file __rcu
*exe_file
;
465 #ifdef CONFIG_MMU_NOTIFIER
466 struct mmu_notifier_mm
*mmu_notifier_mm
;
468 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
469 pgtable_t pmd_huge_pte
; /* protected by page_table_lock */
471 #ifdef CONFIG_CPUMASK_OFFSTACK
472 struct cpumask cpumask_allocation
;
474 #ifdef CONFIG_NUMA_BALANCING
476 * numa_next_scan is the next time that the PTEs will be marked
477 * pte_numa. NUMA hinting faults will gather statistics and migrate
478 * pages to new nodes if necessary.
480 unsigned long numa_next_scan
;
482 /* Restart point for scanning and setting pte_numa */
483 unsigned long numa_scan_offset
;
485 /* numa_scan_seq prevents two threads setting pte_numa */
488 #if defined(CONFIG_NUMA_BALANCING) || defined(CONFIG_COMPACTION)
490 * An operation with batched TLB flushing is going on. Anything that
491 * can move process memory needs to flush the TLB when moving a
492 * PROT_NONE or PROT_NUMA mapped page.
494 bool tlb_flush_pending
;
496 struct uprobes_state uprobes_state
;
497 #ifdef CONFIG_X86_INTEL_MPX
498 /* address of the bounds directory */
499 void __user
*bd_addr
;
501 #ifdef CONFIG_HUGETLB_PAGE
502 atomic_long_t hugetlb_usage
;
506 static inline void mm_init_cpumask(struct mm_struct
*mm
)
508 #ifdef CONFIG_CPUMASK_OFFSTACK
509 mm
->cpu_vm_mask_var
= &mm
->cpumask_allocation
;
511 cpumask_clear(mm
->cpu_vm_mask_var
);
514 /* Future-safe accessor for struct mm_struct's cpu_vm_mask. */
515 static inline cpumask_t
*mm_cpumask(struct mm_struct
*mm
)
517 return mm
->cpu_vm_mask_var
;
520 #if defined(CONFIG_NUMA_BALANCING) || defined(CONFIG_COMPACTION)
522 * Memory barriers to keep this state in sync are graciously provided by
523 * the page table locks, outside of which no page table modifications happen.
524 * The barriers below prevent the compiler from re-ordering the instructions
525 * around the memory barriers that are already present in the code.
527 static inline bool mm_tlb_flush_pending(struct mm_struct
*mm
)
530 return mm
->tlb_flush_pending
;
532 static inline void set_tlb_flush_pending(struct mm_struct
*mm
)
534 mm
->tlb_flush_pending
= true;
537 * Guarantee that the tlb_flush_pending store does not leak into the
538 * critical section updating the page tables
540 smp_mb__before_spinlock();
542 /* Clearing is done after a TLB flush, which also provides a barrier. */
543 static inline void clear_tlb_flush_pending(struct mm_struct
*mm
)
546 mm
->tlb_flush_pending
= false;
549 static inline bool mm_tlb_flush_pending(struct mm_struct
*mm
)
553 static inline void set_tlb_flush_pending(struct mm_struct
*mm
)
556 static inline void clear_tlb_flush_pending(struct mm_struct
*mm
)
561 struct vm_special_mapping
567 enum tlb_flush_reason
{
568 TLB_FLUSH_ON_TASK_SWITCH
,
569 TLB_REMOTE_SHOOTDOWN
,
571 TLB_LOCAL_MM_SHOOTDOWN
,
573 NR_TLB_FLUSH_REASONS
,
577 * A swap entry has to fit into a "unsigned long", as the entry is hidden
578 * in the "index" field of the swapper address space.
584 #endif /* _LINUX_MM_TYPES_H */