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/page-debug-flags.h>
14 #include <linux/uprobes.h>
15 #include <linux/page-flags-layout.h>
19 #ifndef AT_VECTOR_SIZE_ARCH
20 #define AT_VECTOR_SIZE_ARCH 0
22 #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))
31 * Each physical page in the system has a struct page associated with
32 * it to keep track of whatever it is we are using the page for at the
33 * moment. Note that we have no way to track which tasks are using
34 * a page, though if it is a pagecache page, rmap structures can tell us
37 * The objects in struct page are organized in double word blocks in
38 * order to allows us to use atomic double word operations on portions
39 * of struct page. That is currently only used by slub but the arrangement
40 * allows the use of atomic double word operations on the flags/mapping
41 * and lru list pointers also.
44 /* First double word block */
45 unsigned long flags; /* Atomic flags, some possibly
46 * 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 * see PAGE_MAPPING_ANON below.
54 /* Second double word */
57 pgoff_t index; /* Our offset within mapping. */
58 void *freelist; /* slub/slob first free object */
59 bool pfmemalloc; /* If set by the page allocator,
60 * ALLOC_NO_WATERMARKS was set
61 * and the low watermark was not
62 * met implying that the system
63 * is under some pressure. The
64 * caller should try ensure
65 * this page is only used to
68 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && USE_SPLIT_PMD_PTLOCKS
69 pgtable_t pmd_huge_pte; /* protected by page->ptl */
74 #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \
75 defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
76 /* Used for cmpxchg_double in slub */
77 unsigned long counters;
80 * Keep _count separate from slub cmpxchg_double data.
81 * As the rest of the double word is protected by
82 * slab_lock but _count is not.
91 * Count of ptes mapped in
92 * mms, to show when page is
93 * mapped & limit reverse map
96 * Used also for tail pages
97 * refcounting instead of
98 * _count. Tail pages cannot
99 * be mapped and keeping the
100 * tail page _count zero at
101 * all times guarantees
102 * get_page_unless_zero() will
103 * never succeed on tail
113 int units; /* SLOB */
115 atomic_t _count; /* Usage count, see below. */
120 /* Third double word block */
122 struct list_head lru; /* Pageout list, eg. active_list
123 * protected by zone->lru_lock !
125 struct { /* slub per cpu partial pages */
126 struct page *next; /* Next partial slab */
128 int pages; /* Nr of partial slabs left */
129 int pobjects; /* Approximate # of objects */
136 struct list_head list; /* slobs list of pages */
137 struct slab *slab_page; /* slab fields */
140 /* Remainder is not double word aligned */
142 unsigned long private; /* Mapping-private opaque data:
143 * usually used for buffer_heads
144 * if PagePrivate set; used for
145 * swp_entry_t if PageSwapCache;
146 * indicates order in the buddy
147 * system if PG_buddy is set.
149 #if USE_SPLIT_PTE_PTLOCKS
150 #if BLOATED_SPINLOCKS
156 struct kmem_cache *slab_cache; /* SL[AU]B: Pointer to slab */
157 struct page *first_page; /* Compound tail pages */
161 * On machines where all RAM is mapped into kernel address space,
162 * we can simply calculate the virtual address. On machines with
163 * highmem some memory is mapped into kernel virtual memory
164 * dynamically, so we need a place to store that address.
165 * Note that this field could be 16 bits on x86 ... ;)
167 * Architectures with slow multiplication can define
168 * WANT_PAGE_VIRTUAL in asm/page.h
170 #if defined(WANT_PAGE_VIRTUAL)
171 void *virtual; /* Kernel virtual address (NULL if
172 not kmapped, ie. highmem) */
173 #endif /* WANT_PAGE_VIRTUAL */
174 #ifdef CONFIG_WANT_PAGE_DEBUG_FLAGS
175 unsigned long debug_flags; /* Use atomic bitops on this */
178 #ifdef CONFIG_KMEMCHECK
180 * kmemcheck wants to track the status of each byte in a page; this
181 * is a pointer to such a status block. NULL if not tracked.
186 #ifdef LAST_CPUPID_NOT_IN_PAGE_FLAGS
191 * The struct page can be forced to be double word aligned so that atomic ops
192 * on double words work. The SLUB allocator can make use of such a feature.
194 #ifdef CONFIG_HAVE_ALIGNED_STRUCT_PAGE
195 __aligned(2 * sizeof(unsigned long))
201 #if (BITS_PER_LONG > 32) || (PAGE_SIZE >= 65536)
210 typedef unsigned long __nocast vm_flags_t;
213 * A region containing a mapping of a non-memory backed file under NOMMU
214 * conditions. These are held in a global tree and are pinned by the VMAs that
218 struct rb_node vm_rb; /* link in global region tree */
219 vm_flags_t vm_flags; /* VMA vm_flags */
220 unsigned long vm_start; /* start address of region */
221 unsigned long vm_end; /* region initialised to here */
222 unsigned long vm_top; /* region allocated to here */
223 unsigned long vm_pgoff; /* the offset in vm_file corresponding to vm_start */
224 struct file *vm_file; /* the backing file or NULL */
226 int vm_usage; /* region usage count (access under nommu_region_sem) */
227 bool vm_icache_flushed : 1; /* true if the icache has been flushed for
232 * This struct defines a memory VMM memory area. There is one of these
233 * per VM-area/task. A VM area is any part of the process virtual memory
234 * space that has a special rule for the page-fault handlers (ie a shared
235 * library, the executable area etc).
237 struct vm_area_struct {
238 /* The first cache line has the info for VMA tree walking. */
240 unsigned long vm_start; /* Our start address within vm_mm. */
241 unsigned long vm_end; /* The first byte after our end address
244 /* linked list of VM areas per task, sorted by address */
245 struct vm_area_struct *vm_next, *vm_prev;
247 struct rb_node vm_rb;
250 * Largest free memory gap in bytes to the left of this VMA.
251 * Either between this VMA and vma->vm_prev, or between one of the
252 * VMAs below us in the VMA rbtree and its ->vm_prev. This helps
253 * get_unmapped_area find a free area of the right size.
255 unsigned long rb_subtree_gap;
257 /* Second cache line starts here. */
259 struct mm_struct *vm_mm; /* The address space we belong to. */
260 pgprot_t vm_page_prot; /* Access permissions of this VMA. */
261 unsigned long vm_flags; /* Flags, see mm.h. */
264 * For areas with an address space and backing store,
265 * linkage into the address_space->i_mmap interval tree, or
266 * linkage of vma in the address_space->i_mmap_nonlinear list.
271 unsigned long rb_subtree_last;
273 struct list_head nonlinear;
277 * A file's MAP_PRIVATE vma can be in both i_mmap tree and anon_vma
278 * list, after a COW of one of the file pages. A MAP_SHARED vma
279 * can only be in the i_mmap tree. An anonymous MAP_PRIVATE, stack
280 * or brk vma (with NULL file) can only be in an anon_vma list.
282 struct list_head anon_vma_chain; /* Serialized by mmap_sem &
284 struct anon_vma *anon_vma; /* Serialized by page_table_lock */
286 /* Function pointers to deal with this struct. */
287 const struct vm_operations_struct *vm_ops;
289 /* Information about our backing store: */
290 unsigned long vm_pgoff; /* Offset (within vm_file) in PAGE_SIZE
291 units, *not* PAGE_CACHE_SIZE */
292 struct file * vm_file; /* File we map to (can be NULL). */
293 void * vm_private_data; /* was vm_pte (shared mem) */
296 struct vm_region *vm_region; /* NOMMU mapping region */
299 struct mempolicy *vm_policy; /* NUMA policy for the VMA */
304 struct task_struct *task;
305 struct core_thread *next;
310 struct core_thread dumper;
311 struct completion startup;
321 #if USE_SPLIT_PTE_PTLOCKS && defined(CONFIG_MMU)
322 #define SPLIT_RSS_COUNTING
323 /* per-thread cached information, */
324 struct task_rss_stat {
325 int events; /* for synchronization threshold */
326 int count[NR_MM_COUNTERS];
328 #endif /* USE_SPLIT_PTE_PTLOCKS */
331 atomic_long_t count[NR_MM_COUNTERS];
336 struct vm_area_struct * mmap; /* list of VMAs */
337 struct rb_root mm_rb;
338 struct vm_area_struct * mmap_cache; /* last find_vma result */
340 unsigned long (*get_unmapped_area) (struct file *filp,
341 unsigned long addr, unsigned long len,
342 unsigned long pgoff, unsigned long flags);
344 unsigned long mmap_base; /* base of mmap area */
345 unsigned long mmap_legacy_base; /* base of mmap area in bottom-up allocations */
346 unsigned long task_size; /* size of task vm space */
347 unsigned long highest_vm_end; /* highest vma end address */
349 atomic_t mm_users; /* How many users with user space? */
350 atomic_t mm_count; /* How many references to "struct mm_struct" (users count as 1) */
351 atomic_long_t nr_ptes; /* Page table pages */
352 int map_count; /* number of VMAs */
354 spinlock_t page_table_lock; /* Protects page tables and some counters */
355 struct rw_semaphore mmap_sem;
357 struct list_head mmlist; /* List of maybe swapped mm's. These are globally strung
358 * together off init_mm.mmlist, and are protected
363 unsigned long hiwater_rss; /* High-watermark of RSS usage */
364 unsigned long hiwater_vm; /* High-water virtual memory usage */
366 unsigned long total_vm; /* Total pages mapped */
367 unsigned long locked_vm; /* Pages that have PG_mlocked set */
368 unsigned long pinned_vm; /* Refcount permanently increased */
369 unsigned long shared_vm; /* Shared pages (files) */
370 unsigned long exec_vm; /* VM_EXEC & ~VM_WRITE */
371 unsigned long stack_vm; /* VM_GROWSUP/DOWN */
372 unsigned long def_flags;
373 unsigned long start_code, end_code, start_data, end_data;
374 unsigned long start_brk, brk, start_stack;
375 unsigned long arg_start, arg_end, env_start, env_end;
377 unsigned long saved_auxv[AT_VECTOR_SIZE]; /* for /proc/PID/auxv */
380 * Special counters, in some configurations protected by the
381 * page_table_lock, in other configurations by being atomic.
383 struct mm_rss_stat rss_stat;
385 struct linux_binfmt *binfmt;
387 cpumask_var_t cpu_vm_mask_var;
389 /* Architecture-specific MM context */
390 mm_context_t context;
392 unsigned long flags; /* Must use atomic bitops to access the bits */
394 struct core_state *core_state; /* coredumping support */
396 spinlock_t ioctx_lock;
397 struct kioctx_table __rcu *ioctx_table;
399 #ifdef CONFIG_MM_OWNER
401 * "owner" points to a task that is regarded as the canonical
402 * user/owner of this mm. All of the following must be true in
403 * order for it to be changed:
405 * current == mm->owner
407 * new_owner->mm == mm
408 * new_owner->alloc_lock is held
410 struct task_struct __rcu *owner;
413 /* store ref to file /proc/<pid>/exe symlink points to */
414 struct file *exe_file;
415 #ifdef CONFIG_MMU_NOTIFIER
416 struct mmu_notifier_mm *mmu_notifier_mm;
418 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
419 pgtable_t pmd_huge_pte; /* protected by page_table_lock */
421 #ifdef CONFIG_CPUMASK_OFFSTACK
422 struct cpumask cpumask_allocation;
424 #ifdef CONFIG_NUMA_BALANCING
426 * numa_next_scan is the next time that the PTEs will be marked
427 * pte_numa. NUMA hinting faults will gather statistics and migrate
428 * pages to new nodes if necessary.
430 unsigned long numa_next_scan;
432 /* Restart point for scanning and setting pte_numa */
433 unsigned long numa_scan_offset;
435 /* numa_scan_seq prevents two threads setting pte_numa */
438 struct uprobes_state uprobes_state;
441 static inline void mm_init_cpumask(struct mm_struct *mm)
443 #ifdef CONFIG_CPUMASK_OFFSTACK
444 mm->cpu_vm_mask_var = &mm->cpumask_allocation;
448 /* Future-safe accessor for struct mm_struct's cpu_vm_mask. */
449 static inline cpumask_t *mm_cpumask(struct mm_struct *mm)
451 return mm->cpu_vm_mask_var;
454 #endif /* _LINUX_MM_TYPES_H */