1 #ifndef _LINUX_MM_TYPES_H
2 #define _LINUX_MM_TYPES_H
3
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/prio_tree.h>
10 #include <linux/rbtree.h>
11 #include <linux/rwsem.h>
12 #include <linux/completion.h>
13 #include <linux/cpumask.h>
14 #include <linux/page-debug-flags.h>
15 #include <asm/page.h>
16 #include <asm/mmu.h>
17
18 #ifndef AT_VECTOR_SIZE_ARCH
19 #define AT_VECTOR_SIZE_ARCH 0
20 #endif
21 #define AT_VECTOR_SIZE (2*(AT_VECTOR_SIZE_ARCH + AT_VECTOR_SIZE_BASE + 1))
22
23 struct address_space;
24
25 #define USE_SPLIT_PTLOCKS (NR_CPUS >= CONFIG_SPLIT_PTLOCK_CPUS)
26
27 /*
28 * Each physical page in the system has a struct page associated with
29 * it to keep track of whatever it is we are using the page for at the
30 * moment. Note that we have no way to track which tasks are using
31 * a page, though if it is a pagecache page, rmap structures can tell us
32 * who is mapping it.
33 *
34 * The objects in struct page are organized in double word blocks in
35 * order to allows us to use atomic double word operations on portions
36 * of struct page. That is currently only used by slub but the arrangement
37 * allows the use of atomic double word operations on the flags/mapping
38 * and lru list pointers also.
39 */
40 struct page {
41 /* First double word block */
42 unsigned long flags; /* Atomic flags, some possibly
43 * updated asynchronously */
44 struct address_space *mapping; /* If low bit clear, points to
45 * inode address_space, or NULL.
46 * If page mapped as anonymous
47 * memory, low bit is set, and
48 * it points to anon_vma object:
49 * see PAGE_MAPPING_ANON below.
50 */
51 /* Second double word */
52 struct {
53 union {
54 pgoff_t index; /* Our offset within mapping. */
55 void *freelist; /* slub first free object */
56 };
57
58 union {
59 #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \
60 defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
61 /* Used for cmpxchg_double in slub */
62 unsigned long counters;
63 #else
64 /*
65 * Keep _count separate from slub cmpxchg_double data.
66 * As the rest of the double word is protected by
67 * slab_lock but _count is not.
68 */
69 unsigned counters;
70 #endif
71
72 struct {
73
74 union {
75 /*
76 * Count of ptes mapped in
77 * mms, to show when page is
78 * mapped & limit reverse map
79 * searches.
80 *
81 * Used also for tail pages
82 * refcounting instead of
83 * _count. Tail pages cannot
84 * be mapped and keeping the
85 * tail page _count zero at
86 * all times guarantees
87 * get_page_unless_zero() will
88 * never succeed on tail
89 * pages.
90 */
91 atomic_t _mapcount;
92
93 struct {
94 unsigned inuse:16;
95 unsigned objects:15;
96 unsigned frozen:1;
97 };
98 };
99 atomic_t _count; /* Usage count, see below. */
100 };
101 };
102 };
103
104 /* Third double word block */
105 union {
106 struct list_head lru; /* Pageout list, eg. active_list
107 * protected by zone->lru_lock !
108 */
109 struct { /* slub per cpu partial pages */
110 struct page *next; /* Next partial slab */
111 #ifdef CONFIG_64BIT
112 int pages; /* Nr of partial slabs left */
113 int pobjects; /* Approximate # of objects */
114 #else
115 short int pages;
116 short int pobjects;
117 #endif
118 };
119 };
120
121 /* Remainder is not double word aligned */
122 union {
123 unsigned long private; /* Mapping-private opaque data:
124 * usually used for buffer_heads
125 * if PagePrivate set; used for
126 * swp_entry_t if PageSwapCache;
127 * indicates order in the buddy
128 * system if PG_buddy is set.
129 */
130 #if USE_SPLIT_PTLOCKS
131 spinlock_t ptl;
132 #endif
133 struct kmem_cache *slab; /* SLUB: Pointer to slab */
134 struct page *first_page; /* Compound tail pages */
135 };
136
137 /*
138 * On machines where all RAM is mapped into kernel address space,
139 * we can simply calculate the virtual address. On machines with
140 * highmem some memory is mapped into kernel virtual memory
141 * dynamically, so we need a place to store that address.
142 * Note that this field could be 16 bits on x86 ... ;)
143 *
144 * Architectures with slow multiplication can define
145 * WANT_PAGE_VIRTUAL in asm/page.h
146 */
147 #if defined(WANT_PAGE_VIRTUAL)
148 void *virtual; /* Kernel virtual address (NULL if
149 not kmapped, ie. highmem) */
150 #endif /* WANT_PAGE_VIRTUAL */
151 #ifdef CONFIG_WANT_PAGE_DEBUG_FLAGS
152 unsigned long debug_flags; /* Use atomic bitops on this */
153 #endif
154
155 #ifdef CONFIG_KMEMCHECK
156 /*
157 * kmemcheck wants to track the status of each byte in a page; this
158 * is a pointer to such a status block. NULL if not tracked.
159 */
160 void *shadow;
161 #endif
162 }
163 /*
164 * The struct page can be forced to be double word aligned so that atomic ops
165 * on double words work. The SLUB allocator can make use of such a feature.
166 */
167 #ifdef CONFIG_HAVE_ALIGNED_STRUCT_PAGE
168 __aligned(2 * sizeof(unsigned long))
169 #endif
170 ;
171
172 struct page_frag {
173 struct page *page;
174 #if (BITS_PER_LONG > 32) || (PAGE_SIZE >= 65536)
175 __u32 offset;
176 __u32 size;
177 #else
178 __u16 offset;
179 __u16 size;
180 #endif
181 };
182
183 typedef unsigned long __nocast vm_flags_t;
184
185 /*
186 * A region containing a mapping of a non-memory backed file under NOMMU
187 * conditions. These are held in a global tree and are pinned by the VMAs that
188 * map parts of them.
189 */
190 struct vm_region {
191 struct rb_node vm_rb; /* link in global region tree */
192 vm_flags_t vm_flags; /* VMA vm_flags */
193 unsigned long vm_start; /* start address of region */
194 unsigned long vm_end; /* region initialised to here */
195 unsigned long vm_top; /* region allocated to here */
196 unsigned long vm_pgoff; /* the offset in vm_file corresponding to vm_start */
197 struct file *vm_file; /* the backing file or NULL */
198
199 int vm_usage; /* region usage count (access under nommu_region_sem) */
200 bool vm_icache_flushed : 1; /* true if the icache has been flushed for
201 * this region */
202 };
203
204 /*
205 * This struct defines a memory VMM memory area. There is one of these
206 * per VM-area/task. A VM area is any part of the process virtual memory
207 * space that has a special rule for the page-fault handlers (ie a shared
208 * library, the executable area etc).
209 */
210 struct vm_area_struct {
211 struct mm_struct * vm_mm; /* The address space we belong to. */
212 unsigned long vm_start; /* Our start address within vm_mm. */
213 unsigned long vm_end; /* The first byte after our end address
214 within vm_mm. */
215
216 /* linked list of VM areas per task, sorted by address */
217 struct vm_area_struct *vm_next, *vm_prev;
218
219 pgprot_t vm_page_prot; /* Access permissions of this VMA. */
220 unsigned long vm_flags; /* Flags, see mm.h. */
221
222 struct rb_node vm_rb;
223
224 /*
225 * For areas with an address space and backing store,
226 * linkage into the address_space->i_mmap prio tree, or
227 * linkage to the list of like vmas hanging off its node, or
228 * linkage of vma in the address_space->i_mmap_nonlinear list.
229 */
230 union {
231 struct {
232 struct list_head list;
233 void *parent; /* aligns with prio_tree_node parent */
234 struct vm_area_struct *head;
235 } vm_set;
236
237 struct raw_prio_tree_node prio_tree_node;
238 } shared;
239
240 /*
241 * A file's MAP_PRIVATE vma can be in both i_mmap tree and anon_vma
242 * list, after a COW of one of the file pages. A MAP_SHARED vma
243 * can only be in the i_mmap tree. An anonymous MAP_PRIVATE, stack
244 * or brk vma (with NULL file) can only be in an anon_vma list.
245 */
246 struct list_head anon_vma_chain; /* Serialized by mmap_sem &
247 * page_table_lock */
248 struct anon_vma *anon_vma; /* Serialized by page_table_lock */
249
250 /* Function pointers to deal with this struct. */
251 const struct vm_operations_struct *vm_ops;
252
253 /* Information about our backing store: */
254 unsigned long vm_pgoff; /* Offset (within vm_file) in PAGE_SIZE
255 units, *not* PAGE_CACHE_SIZE */
256 struct file * vm_file; /* File we map to (can be NULL). */
257 void * vm_private_data; /* was vm_pte (shared mem) */
258
259 #ifndef CONFIG_MMU
260 struct vm_region *vm_region; /* NOMMU mapping region */
261 #endif
262 #ifdef CONFIG_NUMA
263 struct mempolicy *vm_policy; /* NUMA policy for the VMA */
264 #endif
265 };
266
267 struct core_thread {
268 struct task_struct *task;
269 struct core_thread *next;
270 };
271
272 struct core_state {
273 atomic_t nr_threads;
274 struct core_thread dumper;
275 struct completion startup;
276 };
277
278 enum {
279 MM_FILEPAGES,
280 MM_ANONPAGES,
281 MM_SWAPENTS,
282 NR_MM_COUNTERS
283 };
284
285 #if USE_SPLIT_PTLOCKS && defined(CONFIG_MMU)
286 #define SPLIT_RSS_COUNTING
287 /* per-thread cached information, */
288 struct task_rss_stat {
289 int events; /* for synchronization threshold */
290 int count[NR_MM_COUNTERS];
291 };
292 #endif /* USE_SPLIT_PTLOCKS */
293
294 struct mm_rss_stat {
295 atomic_long_t count[NR_MM_COUNTERS];
296 };
297
298 struct mm_struct {
299 struct vm_area_struct * mmap; /* list of VMAs */
300 struct rb_root mm_rb;
301 struct vm_area_struct * mmap_cache; /* last find_vma result */
302 #ifdef CONFIG_MMU
303 unsigned long (*get_unmapped_area) (struct file *filp,
304 unsigned long addr, unsigned long len,
305 unsigned long pgoff, unsigned long flags);
306 void (*unmap_area) (struct mm_struct *mm, unsigned long addr);
307 #endif
308 unsigned long mmap_base; /* base of mmap area */
309 unsigned long mmap_legacy_base; /* base of mmap area in bottom-up allocations */
310 unsigned long task_size; /* size of task vm space */
311 unsigned long cached_hole_size; /* if non-zero, the largest hole below free_area_cache */
312 unsigned long free_area_cache; /* first hole of size cached_hole_size or larger */
313 pgd_t * pgd;
314 atomic_t mm_users; /* How many users with user space? */
315 atomic_t mm_count; /* How many references to "struct mm_struct" (users count as 1) */
316 int map_count; /* number of VMAs */
317
318 spinlock_t page_table_lock; /* Protects page tables and some counters */
319 struct rw_semaphore mmap_sem;
320
321 struct list_head mmlist; /* List of maybe swapped mm's. These are globally strung
322 * together off init_mm.mmlist, and are protected
323 * by mmlist_lock
324 */
325
326
327 unsigned long hiwater_rss; /* High-watermark of RSS usage */
328 unsigned long hiwater_vm; /* High-water virtual memory usage */
329
330 unsigned long total_vm; /* Total pages mapped */
331 unsigned long locked_vm; /* Pages that have PG_mlocked set */
332 unsigned long pinned_vm; /* Refcount permanently increased */
333 unsigned long shared_vm; /* Shared pages (files) */
334 unsigned long exec_vm; /* VM_EXEC & ~VM_WRITE */
335 unsigned long stack_vm; /* VM_GROWSUP/DOWN */
336 unsigned long reserved_vm; /* VM_RESERVED|VM_IO pages */
337 unsigned long def_flags;
338 unsigned long nr_ptes; /* Page table pages */
339 unsigned long start_code, end_code, start_data, end_data;
340 unsigned long start_brk, brk, start_stack;
341 unsigned long arg_start, arg_end, env_start, env_end;
342
343 unsigned long saved_auxv[AT_VECTOR_SIZE]; /* for /proc/PID/auxv */
344
345 /*
346 * Special counters, in some configurations protected by the
347 * page_table_lock, in other configurations by being atomic.
348 */
349 struct mm_rss_stat rss_stat;
350
351 struct linux_binfmt *binfmt;
352
353 cpumask_var_t cpu_vm_mask_var;
354
355 /* Architecture-specific MM context */
356 mm_context_t context;
357
358 /* Swap token stuff */
359 /*
360 * Last value of global fault stamp as seen by this process.
361 * In other words, this value gives an indication of how long
362 * it has been since this task got the token.
363 * Look at mm/thrash.c
364 */
365 unsigned int faultstamp;
366 unsigned int token_priority;
367 unsigned int last_interval;
368
369 unsigned long flags; /* Must use atomic bitops to access the bits */
370
371 struct core_state *core_state; /* coredumping support */
372 #ifdef CONFIG_AIO
373 spinlock_t ioctx_lock;
374 struct hlist_head ioctx_list;
375 #endif
376 #ifdef CONFIG_MM_OWNER
377 /*
378 * "owner" points to a task that is regarded as the canonical
379 * user/owner of this mm. All of the following must be true in
380 * order for it to be changed:
381 *
382 * current == mm->owner
383 * current->mm != mm
384 * new_owner->mm == mm
385 * new_owner->alloc_lock is held
386 */
387 struct task_struct __rcu *owner;
388 #endif
389
390 /* store ref to file /proc/<pid>/exe symlink points to */
391 struct file *exe_file;
392 unsigned long num_exe_file_vmas;
393 #ifdef CONFIG_MMU_NOTIFIER
394 struct mmu_notifier_mm *mmu_notifier_mm;
395 #endif
396 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
397 pgtable_t pmd_huge_pte; /* protected by page_table_lock */
398 #endif
399 #ifdef CONFIG_CPUMASK_OFFSTACK
400 struct cpumask cpumask_allocation;
401 #endif
402 };
403
mm_init_cpumask(struct mm_struct * mm)404 static inline void mm_init_cpumask(struct mm_struct *mm)
405 {
406 #ifdef CONFIG_CPUMASK_OFFSTACK
407 mm->cpu_vm_mask_var = &mm->cpumask_allocation;
408 #endif
409 }
410
411 /* Future-safe accessor for struct mm_struct's cpu_vm_mask. */
mm_cpumask(struct mm_struct * mm)412 static inline cpumask_t *mm_cpumask(struct mm_struct *mm)
413 {
414 return mm->cpu_vm_mask_var;
415 }
416
417 #endif /* _LINUX_MM_TYPES_H */
418