]>
Commit | Line | Data |
---|---|---|
b2441318 | 1 | /* SPDX-License-Identifier: GPL-2.0 */ |
5b99cd0e HC |
2 | #ifndef _LINUX_MM_TYPES_H |
3 | #define _LINUX_MM_TYPES_H | |
4 | ||
2e58f173 IM |
5 | #include <linux/mm_types_task.h> |
6 | ||
4f9a58d7 | 7 | #include <linux/auxvec.h> |
5b99cd0e HC |
8 | #include <linux/list.h> |
9 | #include <linux/spinlock.h> | |
c92ff1bd MS |
10 | #include <linux/rbtree.h> |
11 | #include <linux/rwsem.h> | |
12 | #include <linux/completion.h> | |
cddb8a5c | 13 | #include <linux/cpumask.h> |
d4b3b638 | 14 | #include <linux/uprobes.h> |
bbeae5b0 | 15 | #include <linux/page-flags-layout.h> |
ec8d7c14 | 16 | #include <linux/workqueue.h> |
2e58f173 | 17 | |
c92ff1bd | 18 | #include <asm/mmu.h> |
5b99cd0e | 19 | |
4f9a58d7 OH |
20 | #ifndef AT_VECTOR_SIZE_ARCH |
21 | #define AT_VECTOR_SIZE_ARCH 0 | |
22 | #endif | |
23 | #define AT_VECTOR_SIZE (2*(AT_VECTOR_SIZE_ARCH + AT_VECTOR_SIZE_BASE + 1)) | |
24 | ||
1c8f4220 | 25 | |
5b99cd0e | 26 | struct address_space; |
1306a85a | 27 | struct mem_cgroup; |
5b99cd0e HC |
28 | |
29 | /* | |
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 | |
97b4a671 | 34 | * who is mapping it. |
be50015d | 35 | * |
97b4a671 MW |
36 | * If you allocate the page using alloc_pages(), you can use some of the |
37 | * space in struct page for your own purposes. The five words in the main | |
38 | * union are available, except for bit 0 of the first word which must be | |
39 | * kept clear. Many users use this word to store a pointer to an object | |
40 | * which is guaranteed to be aligned. If you use the same storage as | |
41 | * page->mapping, you must restore it to NULL before freeing the page. | |
be50015d | 42 | * |
97b4a671 MW |
43 | * If your page will not be mapped to userspace, you can also use the four |
44 | * bytes in the mapcount union, but you must call page_mapcount_reset() | |
45 | * before freeing it. | |
46 | * | |
47 | * If you want to use the refcount field, it must be used in such a way | |
48 | * that other CPUs temporarily incrementing and then decrementing the | |
49 | * refcount does not cause problems. On receiving the page from | |
50 | * alloc_pages(), the refcount will be positive. | |
51 | * | |
52 | * If you allocate pages of order > 0, you can use some of the fields | |
53 | * in each subpage, but you may need to restore some of their values | |
54 | * afterwards. | |
fc9bb8c7 | 55 | * |
4cf7c8bf MW |
56 | * SLUB uses cmpxchg_double() to atomically update its freelist and |
57 | * counters. That requires that freelist & counters be adjacent and | |
58 | * double-word aligned. We align all struct pages to double-word | |
59 | * boundaries, and ensure that 'freelist' is aligned within the | |
60 | * struct. | |
5b99cd0e | 61 | */ |
e20df2c6 MW |
62 | #ifdef CONFIG_HAVE_ALIGNED_STRUCT_PAGE |
63 | #define _struct_page_alignment __aligned(2 * sizeof(unsigned long)) | |
64 | #else | |
0dd4da5b | 65 | #define _struct_page_alignment |
7d27a04b | 66 | #endif |
e20df2c6 | 67 | |
5b99cd0e HC |
68 | struct page { |
69 | unsigned long flags; /* Atomic flags, some possibly | |
70 | * updated asynchronously */ | |
b7ccc7f8 | 71 | /* |
4da1984e MW |
72 | * Five words (20/40 bytes) are available in this union. |
73 | * WARNING: bit 0 of the first word is used for PageTail(). That | |
74 | * means the other users of this union MUST NOT use the bit to | |
b7ccc7f8 MW |
75 | * avoid collision and false-positive PageTail(). |
76 | */ | |
8456a648 | 77 | union { |
66a6ffd2 | 78 | struct { /* Page cache and anonymous pages */ |
4da1984e MW |
79 | /** |
80 | * @lru: Pageout list, eg. active_list protected by | |
f4b7e272 | 81 | * pgdat->lru_lock. Sometimes used as a generic list |
4da1984e MW |
82 | * by the page owner. |
83 | */ | |
84 | struct list_head lru; | |
66a6ffd2 MW |
85 | /* See page-flags.h for PAGE_MAPPING_FLAGS */ |
86 | struct address_space *mapping; | |
87 | pgoff_t index; /* Our offset within mapping. */ | |
88 | /** | |
89 | * @private: Mapping-private opaque data. | |
90 | * Usually used for buffer_heads if PagePrivate. | |
91 | * Used for swp_entry_t if PageSwapCache. | |
92 | * Indicates order in the buddy system if PageBuddy. | |
93 | */ | |
94 | unsigned long private; | |
95 | }; | |
c25fff71 JDB |
96 | struct { /* page_pool used by netstack */ |
97 | /** | |
98 | * @dma_addr: might require a 64-bit value even on | |
99 | * 32-bit architectures. | |
100 | */ | |
101 | dma_addr_t dma_addr; | |
102 | }; | |
66a6ffd2 | 103 | struct { /* slab, slob and slub */ |
4da1984e | 104 | union { |
3e05617c | 105 | struct list_head slab_list; |
4da1984e MW |
106 | struct { /* Partial pages */ |
107 | struct page *next; | |
108 | #ifdef CONFIG_64BIT | |
109 | int pages; /* Nr of pages left */ | |
110 | int pobjects; /* Approximate count */ | |
111 | #else | |
112 | short int pages; | |
113 | short int pobjects; | |
114 | #endif | |
115 | }; | |
116 | }; | |
66a6ffd2 MW |
117 | struct kmem_cache *slab_cache; /* not slob */ |
118 | /* Double-word boundary */ | |
119 | void *freelist; /* first free object */ | |
120 | union { | |
121 | void *s_mem; /* slab: first object */ | |
122 | unsigned long counters; /* SLUB */ | |
123 | struct { /* SLUB */ | |
124 | unsigned inuse:16; | |
125 | unsigned objects:15; | |
126 | unsigned frozen:1; | |
127 | }; | |
128 | }; | |
129 | }; | |
4da1984e MW |
130 | struct { /* Tail pages of compound page */ |
131 | unsigned long compound_head; /* Bit zero is set */ | |
132 | ||
133 | /* First tail page only */ | |
134 | unsigned char compound_dtor; | |
135 | unsigned char compound_order; | |
136 | atomic_t compound_mapcount; | |
1378a5ee | 137 | unsigned int compound_nr; /* 1 << compound_order */ |
4da1984e MW |
138 | }; |
139 | struct { /* Second tail page of compound page */ | |
140 | unsigned long _compound_pad_1; /* compound_head */ | |
47e29d32 | 141 | atomic_t hpage_pinned_refcount; |
87eaceb3 | 142 | /* For both global and memcg */ |
4da1984e MW |
143 | struct list_head deferred_list; |
144 | }; | |
66a6ffd2 | 145 | struct { /* Page table pages */ |
4da1984e MW |
146 | unsigned long _pt_pad_1; /* compound_head */ |
147 | pgtable_t pmd_huge_pte; /* protected by page->ptl */ | |
66a6ffd2 | 148 | unsigned long _pt_pad_2; /* mapping */ |
4231aba0 NP |
149 | union { |
150 | struct mm_struct *pt_mm; /* x86 pgds only */ | |
151 | atomic_t pt_frag_refcount; /* powerpc */ | |
152 | }; | |
7d27a04b | 153 | #if ALLOC_SPLIT_PTLOCKS |
66a6ffd2 | 154 | spinlock_t *ptl; |
7d27a04b | 155 | #else |
66a6ffd2 | 156 | spinlock_t ptl; |
7d27a04b | 157 | #endif |
7d27a04b | 158 | }; |
50e7fbc3 MW |
159 | struct { /* ZONE_DEVICE pages */ |
160 | /** @pgmap: Points to the hosting device page map. */ | |
161 | struct dev_pagemap *pgmap; | |
8a164fef | 162 | void *zone_device_data; |
76470ccd RC |
163 | /* |
164 | * ZONE_DEVICE private pages are counted as being | |
165 | * mapped so the next 3 words hold the mapping, index, | |
166 | * and private fields from the source anonymous or | |
167 | * page cache page while the page is migrated to device | |
168 | * private memory. | |
169 | * ZONE_DEVICE MEMORY_DEVICE_FS_DAX pages also | |
170 | * use the mapping, index, and private fields when | |
171 | * pmem backed DAX files are mapped. | |
172 | */ | |
50e7fbc3 | 173 | }; |
4da1984e MW |
174 | |
175 | /** @rcu_head: You can use this to free a page by RCU. */ | |
176 | struct rcu_head rcu_head; | |
7d27a04b MW |
177 | }; |
178 | ||
b21999da MW |
179 | union { /* This union is 4 bytes in size. */ |
180 | /* | |
181 | * If the page can be mapped to userspace, encodes the number | |
182 | * of times this page is referenced by a page table. | |
183 | */ | |
184 | atomic_t _mapcount; | |
185 | ||
6e292b9b MW |
186 | /* |
187 | * If the page is neither PageSlab nor mappable to userspace, | |
188 | * the value stored here may help determine what this page | |
189 | * is used for. See page-flags.h for a list of page types | |
190 | * which are currently stored here. | |
191 | */ | |
192 | unsigned int page_type; | |
193 | ||
ca9c88c7 | 194 | unsigned int active; /* SLAB */ |
ca9c88c7 | 195 | int units; /* SLOB */ |
81819f0f | 196 | }; |
fc9bb8c7 | 197 | |
b21999da MW |
198 | /* Usage count. *DO NOT USE DIRECTLY*. See page_ref.h */ |
199 | atomic_t _refcount; | |
200 | ||
1306a85a | 201 | #ifdef CONFIG_MEMCG |
286e04b8 RG |
202 | union { |
203 | struct mem_cgroup *mem_cgroup; | |
204 | struct obj_cgroup **obj_cgroups; | |
205 | }; | |
1306a85a JW |
206 | #endif |
207 | ||
5b99cd0e HC |
208 | /* |
209 | * On machines where all RAM is mapped into kernel address space, | |
210 | * we can simply calculate the virtual address. On machines with | |
211 | * highmem some memory is mapped into kernel virtual memory | |
212 | * dynamically, so we need a place to store that address. | |
213 | * Note that this field could be 16 bits on x86 ... ;) | |
214 | * | |
215 | * Architectures with slow multiplication can define | |
216 | * WANT_PAGE_VIRTUAL in asm/page.h | |
217 | */ | |
218 | #if defined(WANT_PAGE_VIRTUAL) | |
219 | void *virtual; /* Kernel virtual address (NULL if | |
220 | not kmapped, ie. highmem) */ | |
221 | #endif /* WANT_PAGE_VIRTUAL */ | |
dfec072e | 222 | |
90572890 PZ |
223 | #ifdef LAST_CPUPID_NOT_IN_PAGE_FLAGS |
224 | int _last_cpupid; | |
57e0a030 | 225 | #endif |
e20df2c6 | 226 | } _struct_page_alignment; |
5b99cd0e | 227 | |
169226f7 YS |
228 | static inline atomic_t *compound_mapcount_ptr(struct page *page) |
229 | { | |
230 | return &page[1].compound_mapcount; | |
231 | } | |
232 | ||
47e29d32 JH |
233 | static inline atomic_t *compound_pincount_ptr(struct page *page) |
234 | { | |
235 | return &page[2].hpage_pinned_refcount; | |
236 | } | |
237 | ||
d1402fc7 LG |
238 | /* |
239 | * Used for sizing the vmemmap region on some architectures | |
240 | */ | |
241 | #define STRUCT_PAGE_MAX_SHIFT (order_base_2(sizeof(struct page))) | |
242 | ||
b63ae8ca AD |
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) | |
245 | ||
b03641af | 246 | #define page_private(page) ((page)->private) |
60e65a6f GJ |
247 | |
248 | static inline void set_page_private(struct page *page, unsigned long private) | |
249 | { | |
250 | page->private = private; | |
251 | } | |
b03641af | 252 | |
b63ae8ca AD |
253 | struct page_frag_cache { |
254 | void * va; | |
255 | #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE) | |
256 | __u16 offset; | |
257 | __u16 size; | |
258 | #else | |
259 | __u32 offset; | |
260 | #endif | |
261 | /* we maintain a pagecount bias, so that we dont dirty cache line | |
0139aa7b | 262 | * containing page->_refcount every time we allocate a fragment. |
b63ae8ca AD |
263 | */ |
264 | unsigned int pagecnt_bias; | |
265 | bool pfmemalloc; | |
266 | }; | |
267 | ||
64b990d2 | 268 | typedef unsigned long vm_flags_t; |
ca16d140 | 269 | |
8feae131 DH |
270 | /* |
271 | * A region containing a mapping of a non-memory backed file under NOMMU | |
272 | * conditions. These are held in a global tree and are pinned by the VMAs that | |
273 | * map parts of them. | |
274 | */ | |
275 | struct vm_region { | |
276 | struct rb_node vm_rb; /* link in global region tree */ | |
ca16d140 | 277 | vm_flags_t vm_flags; /* VMA vm_flags */ |
8feae131 DH |
278 | unsigned long vm_start; /* start address of region */ |
279 | unsigned long vm_end; /* region initialised to here */ | |
dd8632a1 | 280 | unsigned long vm_top; /* region allocated to here */ |
8feae131 DH |
281 | unsigned long vm_pgoff; /* the offset in vm_file corresponding to vm_start */ |
282 | struct file *vm_file; /* the backing file or NULL */ | |
283 | ||
1e2ae599 | 284 | int vm_usage; /* region usage count (access under nommu_region_sem) */ |
cfe79c00 MF |
285 | bool vm_icache_flushed : 1; /* true if the icache has been flushed for |
286 | * this region */ | |
8feae131 DH |
287 | }; |
288 | ||
745f234b AA |
289 | #ifdef CONFIG_USERFAULTFD |
290 | #define NULL_VM_UFFD_CTX ((struct vm_userfaultfd_ctx) { NULL, }) | |
291 | struct vm_userfaultfd_ctx { | |
292 | struct userfaultfd_ctx *ctx; | |
293 | }; | |
294 | #else /* CONFIG_USERFAULTFD */ | |
295 | #define NULL_VM_UFFD_CTX ((struct vm_userfaultfd_ctx) {}) | |
296 | struct vm_userfaultfd_ctx {}; | |
297 | #endif /* CONFIG_USERFAULTFD */ | |
298 | ||
c92ff1bd | 299 | /* |
552657b7 | 300 | * This struct describes a virtual memory area. There is one of these |
301 | * per VM-area/task. A VM area is any part of the process virtual memory | |
c92ff1bd MS |
302 | * space that has a special rule for the page-fault handlers (ie a shared |
303 | * library, the executable area etc). | |
304 | */ | |
305 | struct vm_area_struct { | |
e4c6bfd2 RR |
306 | /* The first cache line has the info for VMA tree walking. */ |
307 | ||
c92ff1bd MS |
308 | unsigned long vm_start; /* Our start address within vm_mm. */ |
309 | unsigned long vm_end; /* The first byte after our end address | |
310 | within vm_mm. */ | |
311 | ||
312 | /* linked list of VM areas per task, sorted by address */ | |
297c5eee | 313 | struct vm_area_struct *vm_next, *vm_prev; |
c92ff1bd | 314 | |
c92ff1bd MS |
315 | struct rb_node vm_rb; |
316 | ||
d3737187 ML |
317 | /* |
318 | * Largest free memory gap in bytes to the left of this VMA. | |
319 | * Either between this VMA and vma->vm_prev, or between one of the | |
320 | * VMAs below us in the VMA rbtree and its ->vm_prev. This helps | |
321 | * get_unmapped_area find a free area of the right size. | |
322 | */ | |
323 | unsigned long rb_subtree_gap; | |
324 | ||
e4c6bfd2 RR |
325 | /* Second cache line starts here. */ |
326 | ||
327 | struct mm_struct *vm_mm; /* The address space we belong to. */ | |
574c5b3d TH |
328 | |
329 | /* | |
330 | * Access permissions of this VMA. | |
331 | * See vmf_insert_mixed_prot() for discussion. | |
332 | */ | |
333 | pgprot_t vm_page_prot; | |
e4c6bfd2 RR |
334 | unsigned long vm_flags; /* Flags, see mm.h. */ |
335 | ||
c92ff1bd MS |
336 | /* |
337 | * For areas with an address space and backing store, | |
27ba0644 | 338 | * linkage into the address_space->i_mmap interval tree. |
c92ff1bd | 339 | */ |
ac51b934 KS |
340 | struct { |
341 | struct rb_node rb; | |
342 | unsigned long rb_subtree_last; | |
c92ff1bd MS |
343 | } shared; |
344 | ||
345 | /* | |
346 | * A file's MAP_PRIVATE vma can be in both i_mmap tree and anon_vma | |
347 | * list, after a COW of one of the file pages. A MAP_SHARED vma | |
348 | * can only be in the i_mmap tree. An anonymous MAP_PRIVATE, stack | |
349 | * or brk vma (with NULL file) can only be in an anon_vma list. | |
350 | */ | |
c1e8d7c6 | 351 | struct list_head anon_vma_chain; /* Serialized by mmap_lock & |
5beb4930 | 352 | * page_table_lock */ |
c92ff1bd MS |
353 | struct anon_vma *anon_vma; /* Serialized by page_table_lock */ |
354 | ||
355 | /* Function pointers to deal with this struct. */ | |
f0f37e2f | 356 | const struct vm_operations_struct *vm_ops; |
c92ff1bd MS |
357 | |
358 | /* Information about our backing store: */ | |
359 | unsigned long vm_pgoff; /* Offset (within vm_file) in PAGE_SIZE | |
ea1754a0 | 360 | units */ |
c92ff1bd MS |
361 | struct file * vm_file; /* File we map to (can be NULL). */ |
362 | void * vm_private_data; /* was vm_pte (shared mem) */ | |
c92ff1bd | 363 | |
219f8a2e | 364 | #ifdef CONFIG_SWAP |
ec560175 | 365 | atomic_long_t swap_readahead_info; |
219f8a2e | 366 | #endif |
c92ff1bd | 367 | #ifndef CONFIG_MMU |
8feae131 | 368 | struct vm_region *vm_region; /* NOMMU mapping region */ |
c92ff1bd MS |
369 | #endif |
370 | #ifdef CONFIG_NUMA | |
371 | struct mempolicy *vm_policy; /* NUMA policy for the VMA */ | |
372 | #endif | |
745f234b | 373 | struct vm_userfaultfd_ctx vm_userfaultfd_ctx; |
3859a271 | 374 | } __randomize_layout; |
c92ff1bd | 375 | |
b564daf8 ON |
376 | struct core_thread { |
377 | struct task_struct *task; | |
378 | struct core_thread *next; | |
379 | }; | |
380 | ||
32ecb1f2 | 381 | struct core_state { |
c5f1cc8c | 382 | atomic_t nr_threads; |
b564daf8 | 383 | struct core_thread dumper; |
32ecb1f2 ON |
384 | struct completion startup; |
385 | }; | |
386 | ||
db446a08 | 387 | struct kioctx_table; |
c92ff1bd | 388 | struct mm_struct { |
c1a2f7f0 RR |
389 | struct { |
390 | struct vm_area_struct *mmap; /* list of VMAs */ | |
391 | struct rb_root mm_rb; | |
7a9cdebd | 392 | u64 vmacache_seqnum; /* per-thread vmacache */ |
efc1a3b1 | 393 | #ifdef CONFIG_MMU |
c1a2f7f0 | 394 | unsigned long (*get_unmapped_area) (struct file *filp, |
c92ff1bd MS |
395 | unsigned long addr, unsigned long len, |
396 | unsigned long pgoff, unsigned long flags); | |
efc1a3b1 | 397 | #endif |
c1a2f7f0 RR |
398 | unsigned long mmap_base; /* base of mmap area */ |
399 | unsigned long mmap_legacy_base; /* base of mmap area in bottom-up allocations */ | |
1b028f78 | 400 | #ifdef CONFIG_HAVE_ARCH_COMPAT_MMAP_BASES |
c1a2f7f0 RR |
401 | /* Base adresses for compatible mmap() */ |
402 | unsigned long mmap_compat_base; | |
403 | unsigned long mmap_compat_legacy_base; | |
1b028f78 | 404 | #endif |
c1a2f7f0 RR |
405 | unsigned long task_size; /* size of task vm space */ |
406 | unsigned long highest_vm_end; /* highest vma end address */ | |
407 | pgd_t * pgd; | |
408 | ||
227a4aad MD |
409 | #ifdef CONFIG_MEMBARRIER |
410 | /** | |
411 | * @membarrier_state: Flags controlling membarrier behavior. | |
412 | * | |
413 | * This field is close to @pgd to hopefully fit in the same | |
414 | * cache-line, which needs to be touched by switch_mm(). | |
415 | */ | |
416 | atomic_t membarrier_state; | |
417 | #endif | |
418 | ||
c1a2f7f0 RR |
419 | /** |
420 | * @mm_users: The number of users including userspace. | |
421 | * | |
422 | * Use mmget()/mmget_not_zero()/mmput() to modify. When this | |
423 | * drops to 0 (i.e. when the task exits and there are no other | |
424 | * temporary reference holders), we also release a reference on | |
425 | * @mm_count (which may then free the &struct mm_struct if | |
426 | * @mm_count also drops to 0). | |
427 | */ | |
428 | atomic_t mm_users; | |
429 | ||
430 | /** | |
431 | * @mm_count: The number of references to &struct mm_struct | |
432 | * (@mm_users count as 1). | |
433 | * | |
434 | * Use mmgrab()/mmdrop() to modify. When this drops to 0, the | |
435 | * &struct mm_struct is freed. | |
436 | */ | |
437 | atomic_t mm_count; | |
b279ddc3 | 438 | |
c4812909 | 439 | #ifdef CONFIG_MMU |
c1a2f7f0 | 440 | atomic_long_t pgtables_bytes; /* PTE page table pages */ |
5a3fbef3 | 441 | #endif |
c1a2f7f0 | 442 | int map_count; /* number of VMAs */ |
481b4bb5 | 443 | |
c1a2f7f0 RR |
444 | spinlock_t page_table_lock; /* Protects page tables and some |
445 | * counters | |
446 | */ | |
da1c55f1 | 447 | struct rw_semaphore mmap_lock; |
c92ff1bd | 448 | |
c1a2f7f0 RR |
449 | struct list_head mmlist; /* List of maybe swapped mm's. These |
450 | * are globally strung together off | |
451 | * init_mm.mmlist, and are protected | |
452 | * by mmlist_lock | |
453 | */ | |
c92ff1bd | 454 | |
c92ff1bd | 455 | |
c1a2f7f0 RR |
456 | unsigned long hiwater_rss; /* High-watermark of RSS usage */ |
457 | unsigned long hiwater_vm; /* High-water virtual memory usage */ | |
c92ff1bd | 458 | |
c1a2f7f0 RR |
459 | unsigned long total_vm; /* Total pages mapped */ |
460 | unsigned long locked_vm; /* Pages that have PG_mlocked set */ | |
70f8a3ca | 461 | atomic64_t pinned_vm; /* Refcount permanently increased */ |
c1a2f7f0 RR |
462 | unsigned long data_vm; /* VM_WRITE & ~VM_SHARED & ~VM_STACK */ |
463 | unsigned long exec_vm; /* VM_EXEC & ~VM_WRITE & ~VM_STACK */ | |
464 | unsigned long stack_vm; /* VM_STACK */ | |
465 | unsigned long def_flags; | |
88aa7cc6 | 466 | |
c1a2f7f0 RR |
467 | spinlock_t arg_lock; /* protect the below fields */ |
468 | unsigned long start_code, end_code, start_data, end_data; | |
469 | unsigned long start_brk, brk, start_stack; | |
470 | unsigned long arg_start, arg_end, env_start, env_end; | |
c92ff1bd | 471 | |
c1a2f7f0 | 472 | unsigned long saved_auxv[AT_VECTOR_SIZE]; /* for /proc/PID/auxv */ |
c92ff1bd | 473 | |
c1a2f7f0 RR |
474 | /* |
475 | * Special counters, in some configurations protected by the | |
476 | * page_table_lock, in other configurations by being atomic. | |
477 | */ | |
478 | struct mm_rss_stat rss_stat; | |
801460d0 | 479 | |
c1a2f7f0 | 480 | struct linux_binfmt *binfmt; |
6345d24d | 481 | |
c1a2f7f0 RR |
482 | /* Architecture-specific MM context */ |
483 | mm_context_t context; | |
c92ff1bd | 484 | |
c1a2f7f0 | 485 | unsigned long flags; /* Must use atomic bitops to access */ |
c92ff1bd | 486 | |
c1a2f7f0 | 487 | struct core_state *core_state; /* coredumping support */ |
227a4aad | 488 | |
858f0993 | 489 | #ifdef CONFIG_AIO |
c1a2f7f0 RR |
490 | spinlock_t ioctx_lock; |
491 | struct kioctx_table __rcu *ioctx_table; | |
858f0993 | 492 | #endif |
f98bafa0 | 493 | #ifdef CONFIG_MEMCG |
c1a2f7f0 RR |
494 | /* |
495 | * "owner" points to a task that is regarded as the canonical | |
496 | * user/owner of this mm. All of the following must be true in | |
497 | * order for it to be changed: | |
498 | * | |
499 | * current == mm->owner | |
500 | * current->mm != mm | |
501 | * new_owner->mm == mm | |
502 | * new_owner->alloc_lock is held | |
503 | */ | |
504 | struct task_struct __rcu *owner; | |
78fb7466 | 505 | #endif |
c1a2f7f0 | 506 | struct user_namespace *user_ns; |
925d1c40 | 507 | |
c1a2f7f0 RR |
508 | /* store ref to file /proc/<pid>/exe symlink points to */ |
509 | struct file __rcu *exe_file; | |
cddb8a5c | 510 | #ifdef CONFIG_MMU_NOTIFIER |
984cfe4e | 511 | struct mmu_notifier_subscriptions *notifier_subscriptions; |
e7a00c45 | 512 | #endif |
e009bb30 | 513 | #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS |
c1a2f7f0 | 514 | pgtable_t pmd_huge_pte; /* protected by page_table_lock */ |
cbee9f88 PZ |
515 | #endif |
516 | #ifdef CONFIG_NUMA_BALANCING | |
c1a2f7f0 RR |
517 | /* |
518 | * numa_next_scan is the next time that the PTEs will be marked | |
519 | * pte_numa. NUMA hinting faults will gather statistics and | |
520 | * migrate pages to new nodes if necessary. | |
521 | */ | |
522 | unsigned long numa_next_scan; | |
cbee9f88 | 523 | |
c1a2f7f0 RR |
524 | /* Restart point for scanning and setting pte_numa */ |
525 | unsigned long numa_scan_offset; | |
6e5fb223 | 526 | |
c1a2f7f0 RR |
527 | /* numa_scan_seq prevents two threads setting pte_numa */ |
528 | int numa_scan_seq; | |
20841405 | 529 | #endif |
c1a2f7f0 RR |
530 | /* |
531 | * An operation with batched TLB flushing is going on. Anything | |
532 | * that can move process memory needs to flush the TLB when | |
533 | * moving a PROT_NONE or PROT_NUMA mapped page. | |
534 | */ | |
535 | atomic_t tlb_flush_pending; | |
3ea27719 | 536 | #ifdef CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH |
c1a2f7f0 RR |
537 | /* See flush_tlb_batched_pending() */ |
538 | bool tlb_flush_batched; | |
6345d24d | 539 | #endif |
c1a2f7f0 | 540 | struct uprobes_state uprobes_state; |
5d317b2b | 541 | #ifdef CONFIG_HUGETLB_PAGE |
c1a2f7f0 | 542 | atomic_long_t hugetlb_usage; |
5d317b2b | 543 | #endif |
c1a2f7f0 | 544 | struct work_struct async_put_work; |
c1a2f7f0 RR |
545 | } __randomize_layout; |
546 | ||
547 | /* | |
548 | * The mm_cpumask needs to be at the end of mm_struct, because it | |
549 | * is dynamically sized based on nr_cpu_ids. | |
550 | */ | |
551 | unsigned long cpu_bitmap[]; | |
552 | }; | |
c92ff1bd | 553 | |
abe722a1 IM |
554 | extern struct mm_struct init_mm; |
555 | ||
c1a2f7f0 | 556 | /* Pointer magic because the dynamic array size confuses some compilers. */ |
6345d24d LT |
557 | static inline void mm_init_cpumask(struct mm_struct *mm) |
558 | { | |
c1a2f7f0 RR |
559 | unsigned long cpu_bitmap = (unsigned long)mm; |
560 | ||
561 | cpu_bitmap += offsetof(struct mm_struct, cpu_bitmap); | |
562 | cpumask_clear((struct cpumask *)cpu_bitmap); | |
6345d24d LT |
563 | } |
564 | ||
45e575ab | 565 | /* Future-safe accessor for struct mm_struct's cpu_vm_mask. */ |
de03c72c KM |
566 | static inline cpumask_t *mm_cpumask(struct mm_struct *mm) |
567 | { | |
c1a2f7f0 | 568 | return (struct cpumask *)&mm->cpu_bitmap; |
de03c72c | 569 | } |
45e575ab | 570 | |
56236a59 MK |
571 | struct mmu_gather; |
572 | extern void tlb_gather_mmu(struct mmu_gather *tlb, struct mm_struct *mm, | |
573 | unsigned long start, unsigned long end); | |
574 | extern void tlb_finish_mmu(struct mmu_gather *tlb, | |
575 | unsigned long start, unsigned long end); | |
576 | ||
16af97dc | 577 | static inline void init_tlb_flush_pending(struct mm_struct *mm) |
20841405 | 578 | { |
16af97dc | 579 | atomic_set(&mm->tlb_flush_pending, 0); |
20841405 | 580 | } |
16af97dc NA |
581 | |
582 | static inline void inc_tlb_flush_pending(struct mm_struct *mm) | |
20841405 | 583 | { |
16af97dc | 584 | atomic_inc(&mm->tlb_flush_pending); |
af2c1401 | 585 | /* |
8b1b436d PZ |
586 | * The only time this value is relevant is when there are indeed pages |
587 | * to flush. And we'll only flush pages after changing them, which | |
588 | * requires the PTL. | |
589 | * | |
590 | * So the ordering here is: | |
591 | * | |
040cca3a | 592 | * atomic_inc(&mm->tlb_flush_pending); |
8b1b436d PZ |
593 | * spin_lock(&ptl); |
594 | * ... | |
595 | * set_pte_at(); | |
596 | * spin_unlock(&ptl); | |
597 | * | |
598 | * spin_lock(&ptl) | |
599 | * mm_tlb_flush_pending(); | |
600 | * .... | |
601 | * spin_unlock(&ptl); | |
602 | * | |
603 | * flush_tlb_range(); | |
040cca3a | 604 | * atomic_dec(&mm->tlb_flush_pending); |
8b1b436d | 605 | * |
0e709703 PZ |
606 | * Where the increment if constrained by the PTL unlock, it thus |
607 | * ensures that the increment is visible if the PTE modification is | |
608 | * visible. After all, if there is no PTE modification, nobody cares | |
609 | * about TLB flushes either. | |
610 | * | |
611 | * This very much relies on users (mm_tlb_flush_pending() and | |
612 | * mm_tlb_flush_nested()) only caring about _specific_ PTEs (and | |
613 | * therefore specific PTLs), because with SPLIT_PTE_PTLOCKS and RCpc | |
614 | * locks (PPC) the unlock of one doesn't order against the lock of | |
615 | * another PTL. | |
616 | * | |
617 | * The decrement is ordered by the flush_tlb_range(), such that | |
618 | * mm_tlb_flush_pending() will not return false unless all flushes have | |
619 | * completed. | |
af2c1401 | 620 | */ |
20841405 | 621 | } |
16af97dc | 622 | |
16af97dc | 623 | static inline void dec_tlb_flush_pending(struct mm_struct *mm) |
20841405 | 624 | { |
0a2c4048 | 625 | /* |
0e709703 PZ |
626 | * See inc_tlb_flush_pending(). |
627 | * | |
628 | * This cannot be smp_mb__before_atomic() because smp_mb() simply does | |
629 | * not order against TLB invalidate completion, which is what we need. | |
630 | * | |
631 | * Therefore we must rely on tlb_flush_*() to guarantee order. | |
0a2c4048 | 632 | */ |
16af97dc | 633 | atomic_dec(&mm->tlb_flush_pending); |
20841405 | 634 | } |
20841405 | 635 | |
0e709703 PZ |
636 | static inline bool mm_tlb_flush_pending(struct mm_struct *mm) |
637 | { | |
638 | /* | |
639 | * Must be called after having acquired the PTL; orders against that | |
640 | * PTLs release and therefore ensures that if we observe the modified | |
641 | * PTE we must also observe the increment from inc_tlb_flush_pending(). | |
642 | * | |
643 | * That is, it only guarantees to return true if there is a flush | |
644 | * pending for _this_ PTL. | |
645 | */ | |
646 | return atomic_read(&mm->tlb_flush_pending); | |
647 | } | |
648 | ||
649 | static inline bool mm_tlb_flush_nested(struct mm_struct *mm) | |
650 | { | |
651 | /* | |
652 | * Similar to mm_tlb_flush_pending(), we must have acquired the PTL | |
653 | * for which there is a TLB flush pending in order to guarantee | |
654 | * we've seen both that PTE modification and the increment. | |
655 | * | |
656 | * (no requirement on actually still holding the PTL, that is irrelevant) | |
657 | */ | |
658 | return atomic_read(&mm->tlb_flush_pending) > 1; | |
659 | } | |
660 | ||
f872f540 AL |
661 | struct vm_fault; |
662 | ||
3d353901 SJ |
663 | /** |
664 | * typedef vm_fault_t - Return type for page fault handlers. | |
665 | * | |
666 | * Page fault handlers return a bitmask of %VM_FAULT values. | |
667 | */ | |
668 | typedef __bitwise unsigned int vm_fault_t; | |
669 | ||
670 | /** | |
671 | * enum vm_fault_reason - Page fault handlers return a bitmask of | |
672 | * these values to tell the core VM what happened when handling the | |
673 | * fault. Used to decide whether a process gets delivered SIGBUS or | |
674 | * just gets major/minor fault counters bumped up. | |
675 | * | |
676 | * @VM_FAULT_OOM: Out Of Memory | |
677 | * @VM_FAULT_SIGBUS: Bad access | |
678 | * @VM_FAULT_MAJOR: Page read from storage | |
679 | * @VM_FAULT_WRITE: Special case for get_user_pages | |
680 | * @VM_FAULT_HWPOISON: Hit poisoned small page | |
681 | * @VM_FAULT_HWPOISON_LARGE: Hit poisoned large page. Index encoded | |
682 | * in upper bits | |
683 | * @VM_FAULT_SIGSEGV: segmentation fault | |
684 | * @VM_FAULT_NOPAGE: ->fault installed the pte, not return page | |
685 | * @VM_FAULT_LOCKED: ->fault locked the returned page | |
686 | * @VM_FAULT_RETRY: ->fault blocked, must retry | |
687 | * @VM_FAULT_FALLBACK: huge page fault failed, fall back to small | |
688 | * @VM_FAULT_DONE_COW: ->fault has fully handled COW | |
689 | * @VM_FAULT_NEEDDSYNC: ->fault did not modify page tables and needs | |
690 | * fsync() to complete (for synchronous page faults | |
691 | * in DAX) | |
692 | * @VM_FAULT_HINDEX_MASK: mask HINDEX value | |
693 | * | |
694 | */ | |
695 | enum vm_fault_reason { | |
696 | VM_FAULT_OOM = (__force vm_fault_t)0x000001, | |
697 | VM_FAULT_SIGBUS = (__force vm_fault_t)0x000002, | |
698 | VM_FAULT_MAJOR = (__force vm_fault_t)0x000004, | |
699 | VM_FAULT_WRITE = (__force vm_fault_t)0x000008, | |
700 | VM_FAULT_HWPOISON = (__force vm_fault_t)0x000010, | |
701 | VM_FAULT_HWPOISON_LARGE = (__force vm_fault_t)0x000020, | |
702 | VM_FAULT_SIGSEGV = (__force vm_fault_t)0x000040, | |
703 | VM_FAULT_NOPAGE = (__force vm_fault_t)0x000100, | |
704 | VM_FAULT_LOCKED = (__force vm_fault_t)0x000200, | |
705 | VM_FAULT_RETRY = (__force vm_fault_t)0x000400, | |
706 | VM_FAULT_FALLBACK = (__force vm_fault_t)0x000800, | |
707 | VM_FAULT_DONE_COW = (__force vm_fault_t)0x001000, | |
708 | VM_FAULT_NEEDDSYNC = (__force vm_fault_t)0x002000, | |
709 | VM_FAULT_HINDEX_MASK = (__force vm_fault_t)0x0f0000, | |
710 | }; | |
711 | ||
712 | /* Encode hstate index for a hwpoisoned large page */ | |
713 | #define VM_FAULT_SET_HINDEX(x) ((__force vm_fault_t)((x) << 16)) | |
fcae96ff | 714 | #define VM_FAULT_GET_HINDEX(x) (((__force unsigned int)(x) >> 16) & 0xf) |
3d353901 SJ |
715 | |
716 | #define VM_FAULT_ERROR (VM_FAULT_OOM | VM_FAULT_SIGBUS | \ | |
717 | VM_FAULT_SIGSEGV | VM_FAULT_HWPOISON | \ | |
718 | VM_FAULT_HWPOISON_LARGE | VM_FAULT_FALLBACK) | |
719 | ||
720 | #define VM_FAULT_RESULT_TRACE \ | |
721 | { VM_FAULT_OOM, "OOM" }, \ | |
722 | { VM_FAULT_SIGBUS, "SIGBUS" }, \ | |
723 | { VM_FAULT_MAJOR, "MAJOR" }, \ | |
724 | { VM_FAULT_WRITE, "WRITE" }, \ | |
725 | { VM_FAULT_HWPOISON, "HWPOISON" }, \ | |
726 | { VM_FAULT_HWPOISON_LARGE, "HWPOISON_LARGE" }, \ | |
727 | { VM_FAULT_SIGSEGV, "SIGSEGV" }, \ | |
728 | { VM_FAULT_NOPAGE, "NOPAGE" }, \ | |
729 | { VM_FAULT_LOCKED, "LOCKED" }, \ | |
730 | { VM_FAULT_RETRY, "RETRY" }, \ | |
731 | { VM_FAULT_FALLBACK, "FALLBACK" }, \ | |
732 | { VM_FAULT_DONE_COW, "DONE_COW" }, \ | |
733 | { VM_FAULT_NEEDDSYNC, "NEEDDSYNC" } | |
734 | ||
f872f540 AL |
735 | struct vm_special_mapping { |
736 | const char *name; /* The name, e.g. "[vdso]". */ | |
737 | ||
738 | /* | |
739 | * If .fault is not provided, this points to a | |
740 | * NULL-terminated array of pages that back the special mapping. | |
741 | * | |
742 | * This must not be NULL unless .fault is provided. | |
743 | */ | |
a62c34bd | 744 | struct page **pages; |
f872f540 AL |
745 | |
746 | /* | |
747 | * If non-NULL, then this is called to resolve page faults | |
748 | * on the special mapping. If used, .pages is not checked. | |
749 | */ | |
b3ec9f33 SJ |
750 | vm_fault_t (*fault)(const struct vm_special_mapping *sm, |
751 | struct vm_area_struct *vma, | |
752 | struct vm_fault *vmf); | |
b059a453 DS |
753 | |
754 | int (*mremap)(const struct vm_special_mapping *sm, | |
755 | struct vm_area_struct *new_vma); | |
a62c34bd AL |
756 | }; |
757 | ||
d17d8f9d DH |
758 | enum tlb_flush_reason { |
759 | TLB_FLUSH_ON_TASK_SWITCH, | |
760 | TLB_REMOTE_SHOOTDOWN, | |
761 | TLB_LOCAL_SHOOTDOWN, | |
762 | TLB_LOCAL_MM_SHOOTDOWN, | |
5b74283a | 763 | TLB_REMOTE_SEND_IPI, |
d17d8f9d DH |
764 | NR_TLB_FLUSH_REASONS, |
765 | }; | |
766 | ||
bd6dace7 TH |
767 | /* |
768 | * A swap entry has to fit into a "unsigned long", as the entry is hidden | |
769 | * in the "index" field of the swapper address space. | |
770 | */ | |
771 | typedef struct { | |
772 | unsigned long val; | |
773 | } swp_entry_t; | |
774 | ||
5b99cd0e | 775 | #endif /* _LINUX_MM_TYPES_H */ |