1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef _LINUX_MM_TYPES_H
3 #define _LINUX_MM_TYPES_H
5 #include <linux/mm_types_task.h>
7 #include <linux/auxvec.h>
8 #include <linux/list.h>
9 #include <linux/spinlock.h>
10 #include <linux/rbtree.h>
11 #include <linux/rwsem.h>
12 #include <linux/completion.h>
13 #include <linux/cpumask.h>
14 #include <linux/uprobes.h>
15 #include <linux/page-flags-layout.h>
16 #include <linux/workqueue.h>
20 #ifndef AT_VECTOR_SIZE_ARCH
21 #define AT_VECTOR_SIZE_ARCH 0
23 #define AT_VECTOR_SIZE (2*(AT_VECTOR_SIZE_ARCH + AT_VECTOR_SIZE_BASE + 1))
30 * Each physical page in the system has a struct page associated with
31 * it to keep track of whatever it is we are using the page for at the
32 * moment. Note that we have no way to track which tasks are using
33 * a page, though if it is a pagecache page, rmap structures can tell us
36 * The objects in struct page are organized in double word blocks in
37 * order to allows us to use atomic double word operations on portions
38 * of struct page. That is currently only used by slub but the arrangement
39 * allows the use of atomic double word operations on the flags/mapping
40 * and lru list pointers also.
43 /* First double word block */
44 unsigned long flags
; /* Atomic flags, some possibly
45 * updated asynchronously */
47 struct address_space
*mapping
; /* If low bit clear, points to
48 * inode address_space, or NULL.
49 * If page mapped as anonymous
50 * memory, low bit is set, and
51 * it points to anon_vma object
52 * or KSM private structure. See
53 * PAGE_MAPPING_ANON and
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 */
63 pgoff_t index
; /* Our offset within mapping. */
64 void *freelist
; /* sl[aou]b first free object */
65 /* page_deferred_list().prev -- second tail page */
69 #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \
70 defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
71 /* Used for cmpxchg_double in slub */
72 unsigned long counters
;
75 * Keep _refcount separate from slub cmpxchg_double data.
76 * As the rest of the double word is protected by slab_lock
77 * but _refcount is not.
85 * Count of ptes mapped in mms, to show when
86 * page is mapped & limit reverse map searches.
88 * Extra information about page type may be
89 * stored here for pages that are never mapped,
90 * in which case the value MUST BE <= -2.
91 * See page-flags.h for more details.
95 unsigned int active
; /* SLAB */
101 int units
; /* SLOB */
104 * Usage count, *USE WRAPPER FUNCTION* when manual
105 * accounting. See page_ref.h
112 * Third double word block
114 * WARNING: bit 0 of the first word encode PageTail(). That means
115 * the rest users of the storage space MUST NOT use the bit to
116 * avoid collision and false-positive PageTail().
119 struct list_head lru
; /* Pageout list, eg. active_list
120 * protected by zone_lru_lock !
121 * Can be used as a generic list
124 struct dev_pagemap
*pgmap
; /* ZONE_DEVICE pages are never on an
125 * lru or handled by a slab
126 * allocator, this points to the
127 * hosting device page map.
129 struct { /* slub per cpu partial pages */
130 struct page
*next
; /* Next partial slab */
132 int pages
; /* Nr of partial slabs left */
133 int pobjects
; /* Approximate # of objects */
140 struct rcu_head rcu_head
; /* Used by SLAB
141 * when destroying via RCU
143 /* Tail pages of compound page */
145 unsigned long compound_head
; /* If bit zero is set */
147 /* First tail page only */
150 * On 64 bit system we have enough space in struct page
151 * to encode compound_dtor and compound_order with
152 * unsigned int. It can help compiler generate better or
153 * smaller code on some archtectures.
155 unsigned int compound_dtor
;
156 unsigned int compound_order
;
158 unsigned short int compound_dtor
;
159 unsigned short int compound_order
;
163 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && USE_SPLIT_PMD_PTLOCKS
165 unsigned long __pad
; /* do not overlay pmd_huge_pte
166 * with compound_head to avoid
167 * possible bit 0 collision.
169 pgtable_t pmd_huge_pte
; /* protected by page->ptl */
174 /* Remainder is not double word aligned */
176 unsigned long private; /* Mapping-private opaque data:
177 * usually used for buffer_heads
178 * if PagePrivate set; used for
179 * swp_entry_t if PageSwapCache;
180 * indicates order in the buddy
181 * system if PG_buddy is set.
183 #if USE_SPLIT_PTE_PTLOCKS
184 #if ALLOC_SPLIT_PTLOCKS
190 struct kmem_cache
*slab_cache
; /* SL[AU]B: Pointer to slab */
194 struct mem_cgroup
*mem_cgroup
;
198 * On machines where all RAM is mapped into kernel address space,
199 * we can simply calculate the virtual address. On machines with
200 * highmem some memory is mapped into kernel virtual memory
201 * dynamically, so we need a place to store that address.
202 * Note that this field could be 16 bits on x86 ... ;)
204 * Architectures with slow multiplication can define
205 * WANT_PAGE_VIRTUAL in asm/page.h
207 #if defined(WANT_PAGE_VIRTUAL)
208 void *virtual; /* Kernel virtual address (NULL if
209 not kmapped, ie. highmem) */
210 #endif /* WANT_PAGE_VIRTUAL */
212 #ifdef LAST_CPUPID_NOT_IN_PAGE_FLAGS
217 * The struct page can be forced to be double word aligned so that atomic ops
218 * on double words work. The SLUB allocator can make use of such a feature.
220 #ifdef CONFIG_HAVE_ALIGNED_STRUCT_PAGE
221 __aligned(2 * sizeof(unsigned long))
225 #define PAGE_FRAG_CACHE_MAX_SIZE __ALIGN_MASK(32768, ~PAGE_MASK)
226 #define PAGE_FRAG_CACHE_MAX_ORDER get_order(PAGE_FRAG_CACHE_MAX_SIZE)
228 struct page_frag_cache
{
230 #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
236 /* we maintain a pagecount bias, so that we dont dirty cache line
237 * containing page->_refcount every time we allocate a fragment.
239 unsigned int pagecnt_bias
;
243 typedef unsigned long vm_flags_t
;
246 * A region containing a mapping of a non-memory backed file under NOMMU
247 * conditions. These are held in a global tree and are pinned by the VMAs that
251 struct rb_node vm_rb
; /* link in global region tree */
252 vm_flags_t vm_flags
; /* VMA vm_flags */
253 unsigned long vm_start
; /* start address of region */
254 unsigned long vm_end
; /* region initialised to here */
255 unsigned long vm_top
; /* region allocated to here */
256 unsigned long vm_pgoff
; /* the offset in vm_file corresponding to vm_start */
257 struct file
*vm_file
; /* the backing file or NULL */
259 int vm_usage
; /* region usage count (access under nommu_region_sem) */
260 bool vm_icache_flushed
: 1; /* true if the icache has been flushed for
264 #ifdef CONFIG_USERFAULTFD
265 #define NULL_VM_UFFD_CTX ((struct vm_userfaultfd_ctx) { NULL, })
266 struct vm_userfaultfd_ctx
{
267 struct userfaultfd_ctx
*ctx
;
269 #else /* CONFIG_USERFAULTFD */
270 #define NULL_VM_UFFD_CTX ((struct vm_userfaultfd_ctx) {})
271 struct vm_userfaultfd_ctx
{};
272 #endif /* CONFIG_USERFAULTFD */
275 * This struct defines a memory VMM memory area. There is one of these
276 * per VM-area/task. A VM area is any part of the process virtual memory
277 * space that has a special rule for the page-fault handlers (ie a shared
278 * library, the executable area etc).
280 struct vm_area_struct
{
281 /* The first cache line has the info for VMA tree walking. */
283 unsigned long vm_start
; /* Our start address within vm_mm. */
284 unsigned long vm_end
; /* The first byte after our end address
287 /* linked list of VM areas per task, sorted by address */
288 struct vm_area_struct
*vm_next
, *vm_prev
;
290 struct rb_node vm_rb
;
293 * Largest free memory gap in bytes to the left of this VMA.
294 * Either between this VMA and vma->vm_prev, or between one of the
295 * VMAs below us in the VMA rbtree and its ->vm_prev. This helps
296 * get_unmapped_area find a free area of the right size.
298 unsigned long rb_subtree_gap
;
300 /* Second cache line starts here. */
302 struct mm_struct
*vm_mm
; /* The address space we belong to. */
303 pgprot_t vm_page_prot
; /* Access permissions of this VMA. */
304 unsigned long vm_flags
; /* Flags, see mm.h. */
307 * For areas with an address space and backing store,
308 * linkage into the address_space->i_mmap interval tree.
312 unsigned long rb_subtree_last
;
316 * A file's MAP_PRIVATE vma can be in both i_mmap tree and anon_vma
317 * list, after a COW of one of the file pages. A MAP_SHARED vma
318 * can only be in the i_mmap tree. An anonymous MAP_PRIVATE, stack
319 * or brk vma (with NULL file) can only be in an anon_vma list.
321 struct list_head anon_vma_chain
; /* Serialized by mmap_sem &
323 struct anon_vma
*anon_vma
; /* Serialized by page_table_lock */
325 /* Function pointers to deal with this struct. */
326 const struct vm_operations_struct
*vm_ops
;
328 /* Information about our backing store: */
329 unsigned long vm_pgoff
; /* Offset (within vm_file) in PAGE_SIZE
331 struct file
* vm_file
; /* File we map to (can be NULL). */
332 void * vm_private_data
; /* was vm_pte (shared mem) */
334 atomic_long_t swap_readahead_info
;
336 struct vm_region
*vm_region
; /* NOMMU mapping region */
339 struct mempolicy
*vm_policy
; /* NUMA policy for the VMA */
341 struct vm_userfaultfd_ctx vm_userfaultfd_ctx
;
342 } __randomize_layout
;
345 struct task_struct
*task
;
346 struct core_thread
*next
;
351 struct core_thread dumper
;
352 struct completion startup
;
357 struct vm_area_struct
*mmap
; /* list of VMAs */
358 struct rb_root mm_rb
;
359 u32 vmacache_seqnum
; /* per-thread vmacache */
361 unsigned long (*get_unmapped_area
) (struct file
*filp
,
362 unsigned long addr
, unsigned long len
,
363 unsigned long pgoff
, unsigned long flags
);
365 unsigned long mmap_base
; /* base of mmap area */
366 unsigned long mmap_legacy_base
; /* base of mmap area in bottom-up allocations */
367 #ifdef CONFIG_HAVE_ARCH_COMPAT_MMAP_BASES
368 /* Base adresses for compatible mmap() */
369 unsigned long mmap_compat_base
;
370 unsigned long mmap_compat_legacy_base
;
372 unsigned long task_size
; /* size of task vm space */
373 unsigned long highest_vm_end
; /* highest vma end address */
377 * @mm_users: The number of users including userspace.
379 * Use mmget()/mmget_not_zero()/mmput() to modify. When this drops
380 * to 0 (i.e. when the task exits and there are no other temporary
381 * reference holders), we also release a reference on @mm_count
382 * (which may then free the &struct mm_struct if @mm_count also
388 * @mm_count: The number of references to &struct mm_struct
389 * (@mm_users count as 1).
391 * Use mmgrab()/mmdrop() to modify. When this drops to 0, the
392 * &struct mm_struct is freed.
397 atomic_long_t pgtables_bytes
; /* PTE page table pages */
399 int map_count
; /* number of VMAs */
401 spinlock_t page_table_lock
; /* Protects page tables and some counters */
402 struct rw_semaphore mmap_sem
;
404 struct list_head mmlist
; /* List of maybe swapped mm's. These are globally strung
405 * together off init_mm.mmlist, and are protected
410 unsigned long hiwater_rss
; /* High-watermark of RSS usage */
411 unsigned long hiwater_vm
; /* High-water virtual memory usage */
413 unsigned long total_vm
; /* Total pages mapped */
414 unsigned long locked_vm
; /* Pages that have PG_mlocked set */
415 unsigned long pinned_vm
; /* Refcount permanently increased */
416 unsigned long data_vm
; /* VM_WRITE & ~VM_SHARED & ~VM_STACK */
417 unsigned long exec_vm
; /* VM_EXEC & ~VM_WRITE & ~VM_STACK */
418 unsigned long stack_vm
; /* VM_STACK */
419 unsigned long def_flags
;
420 unsigned long start_code
, end_code
, start_data
, end_data
;
421 unsigned long start_brk
, brk
, start_stack
;
422 unsigned long arg_start
, arg_end
, env_start
, env_end
;
424 unsigned long saved_auxv
[AT_VECTOR_SIZE
]; /* for /proc/PID/auxv */
427 * Special counters, in some configurations protected by the
428 * page_table_lock, in other configurations by being atomic.
430 struct mm_rss_stat rss_stat
;
432 struct linux_binfmt
*binfmt
;
434 cpumask_var_t cpu_vm_mask_var
;
436 /* Architecture-specific MM context */
437 mm_context_t context
;
439 unsigned long flags
; /* Must use atomic bitops to access the bits */
441 struct core_state
*core_state
; /* coredumping support */
442 #ifdef CONFIG_MEMBARRIER
443 atomic_t membarrier_state
;
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
;
462 struct user_namespace
*user_ns
;
464 /* store ref to file /proc/<pid>/exe symlink points to */
465 struct file __rcu
*exe_file
;
466 #ifdef CONFIG_MMU_NOTIFIER
467 struct mmu_notifier_mm
*mmu_notifier_mm
;
469 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
470 pgtable_t pmd_huge_pte
; /* protected by page_table_lock */
472 #ifdef CONFIG_CPUMASK_OFFSTACK
473 struct cpumask cpumask_allocation
;
475 #ifdef CONFIG_NUMA_BALANCING
477 * numa_next_scan is the next time that the PTEs will be marked
478 * pte_numa. NUMA hinting faults will gather statistics and migrate
479 * pages to new nodes if necessary.
481 unsigned long numa_next_scan
;
483 /* Restart point for scanning and setting pte_numa */
484 unsigned long numa_scan_offset
;
486 /* numa_scan_seq prevents two threads setting pte_numa */
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 atomic_t tlb_flush_pending
;
495 #ifdef CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH
496 /* See flush_tlb_batched_pending() */
497 bool tlb_flush_batched
;
499 struct uprobes_state uprobes_state
;
500 #ifdef CONFIG_HUGETLB_PAGE
501 atomic_long_t hugetlb_usage
;
503 struct work_struct async_put_work
;
505 #if IS_ENABLED(CONFIG_HMM)
506 /* HMM needs to track a few things per mm */
509 } __randomize_layout
;
511 extern struct mm_struct init_mm
;
513 static inline void mm_init_cpumask(struct mm_struct
*mm
)
515 #ifdef CONFIG_CPUMASK_OFFSTACK
516 mm
->cpu_vm_mask_var
= &mm
->cpumask_allocation
;
518 cpumask_clear(mm
->cpu_vm_mask_var
);
521 /* Future-safe accessor for struct mm_struct's cpu_vm_mask. */
522 static inline cpumask_t
*mm_cpumask(struct mm_struct
*mm
)
524 return mm
->cpu_vm_mask_var
;
528 extern void tlb_gather_mmu(struct mmu_gather
*tlb
, struct mm_struct
*mm
,
529 unsigned long start
, unsigned long end
);
530 extern void tlb_finish_mmu(struct mmu_gather
*tlb
,
531 unsigned long start
, unsigned long end
);
533 static inline void init_tlb_flush_pending(struct mm_struct
*mm
)
535 atomic_set(&mm
->tlb_flush_pending
, 0);
538 static inline void inc_tlb_flush_pending(struct mm_struct
*mm
)
540 atomic_inc(&mm
->tlb_flush_pending
);
542 * The only time this value is relevant is when there are indeed pages
543 * to flush. And we'll only flush pages after changing them, which
546 * So the ordering here is:
548 * atomic_inc(&mm->tlb_flush_pending);
555 * mm_tlb_flush_pending();
560 * atomic_dec(&mm->tlb_flush_pending);
562 * Where the increment if constrained by the PTL unlock, it thus
563 * ensures that the increment is visible if the PTE modification is
564 * visible. After all, if there is no PTE modification, nobody cares
565 * about TLB flushes either.
567 * This very much relies on users (mm_tlb_flush_pending() and
568 * mm_tlb_flush_nested()) only caring about _specific_ PTEs (and
569 * therefore specific PTLs), because with SPLIT_PTE_PTLOCKS and RCpc
570 * locks (PPC) the unlock of one doesn't order against the lock of
573 * The decrement is ordered by the flush_tlb_range(), such that
574 * mm_tlb_flush_pending() will not return false unless all flushes have
579 static inline void dec_tlb_flush_pending(struct mm_struct
*mm
)
582 * See inc_tlb_flush_pending().
584 * This cannot be smp_mb__before_atomic() because smp_mb() simply does
585 * not order against TLB invalidate completion, which is what we need.
587 * Therefore we must rely on tlb_flush_*() to guarantee order.
589 atomic_dec(&mm
->tlb_flush_pending
);
592 static inline bool mm_tlb_flush_pending(struct mm_struct
*mm
)
595 * Must be called after having acquired the PTL; orders against that
596 * PTLs release and therefore ensures that if we observe the modified
597 * PTE we must also observe the increment from inc_tlb_flush_pending().
599 * That is, it only guarantees to return true if there is a flush
600 * pending for _this_ PTL.
602 return atomic_read(&mm
->tlb_flush_pending
);
605 static inline bool mm_tlb_flush_nested(struct mm_struct
*mm
)
608 * Similar to mm_tlb_flush_pending(), we must have acquired the PTL
609 * for which there is a TLB flush pending in order to guarantee
610 * we've seen both that PTE modification and the increment.
612 * (no requirement on actually still holding the PTL, that is irrelevant)
614 return atomic_read(&mm
->tlb_flush_pending
) > 1;
619 struct vm_special_mapping
{
620 const char *name
; /* The name, e.g. "[vdso]". */
623 * If .fault is not provided, this points to a
624 * NULL-terminated array of pages that back the special mapping.
626 * This must not be NULL unless .fault is provided.
631 * If non-NULL, then this is called to resolve page faults
632 * on the special mapping. If used, .pages is not checked.
634 int (*fault
)(const struct vm_special_mapping
*sm
,
635 struct vm_area_struct
*vma
,
636 struct vm_fault
*vmf
);
638 int (*mremap
)(const struct vm_special_mapping
*sm
,
639 struct vm_area_struct
*new_vma
);
642 enum tlb_flush_reason
{
643 TLB_FLUSH_ON_TASK_SWITCH
,
644 TLB_REMOTE_SHOOTDOWN
,
646 TLB_LOCAL_MM_SHOOTDOWN
,
648 NR_TLB_FLUSH_REASONS
,
652 * A swap entry has to fit into a "unsigned long", as the entry is hidden
653 * in the "index" field of the swapper address space.
659 #endif /* _LINUX_MM_TYPES_H */