]>
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 | ||
5b99cd0e | 25 | struct address_space; |
1306a85a | 26 | struct mem_cgroup; |
133ff0ea | 27 | struct hmm; |
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 | |
34 | * who is mapping it. | |
fc9bb8c7 CL |
35 | * |
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. | |
5b99cd0e HC |
41 | */ |
42 | struct page { | |
fc9bb8c7 | 43 | /* First double word block */ |
5b99cd0e HC |
44 | unsigned long flags; /* Atomic flags, some possibly |
45 | * updated asynchronously */ | |
8456a648 JK |
46 | union { |
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 | |
41710443 CD |
51 | * it points to anon_vma object |
52 | * or KSM private structure. See | |
53 | * PAGE_MAPPING_ANON and | |
54 | * PAGE_MAPPING_KSM. | |
8456a648 JK |
55 | */ |
56 | void *s_mem; /* slab first object */ | |
53f9263b | 57 | atomic_t compound_mapcount; /* first tail page */ |
9a982250 | 58 | /* page_deferred_list().next -- second tail page */ |
8456a648 JK |
59 | }; |
60 | ||
fc9bb8c7 | 61 | /* Second double word */ |
99691add VD |
62 | union { |
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 */ | |
66 | }; | |
013e8963 | 67 | |
99691add | 68 | union { |
abca7c49 PS |
69 | #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \ |
70 | defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE) | |
99691add VD |
71 | /* Used for cmpxchg_double in slub */ |
72 | unsigned long counters; | |
abca7c49 | 73 | #else |
99691add VD |
74 | /* |
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. | |
78 | */ | |
79 | unsigned counters; | |
abca7c49 | 80 | #endif |
99691add | 81 | struct { |
013e8963 | 82 | |
99691add | 83 | union { |
0139aa7b | 84 | /* |
99691add VD |
85 | * Count of ptes mapped in mms, to show when |
86 | * page is mapped & limit reverse map searches. | |
632c0a1a VD |
87 | * |
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. | |
0139aa7b | 92 | */ |
99691add VD |
93 | atomic_t _mapcount; |
94 | ||
95 | unsigned int active; /* SLAB */ | |
96 | struct { /* SLUB */ | |
97 | unsigned inuse:16; | |
98 | unsigned objects:15; | |
99 | unsigned frozen:1; | |
100 | }; | |
101 | int units; /* SLOB */ | |
fc9bb8c7 | 102 | }; |
99691add VD |
103 | /* |
104 | * Usage count, *USE WRAPPER FUNCTION* when manual | |
105 | * accounting. See page_ref.h | |
106 | */ | |
107 | atomic_t _refcount; | |
39b26464 | 108 | }; |
81819f0f | 109 | }; |
fc9bb8c7 | 110 | |
1d798ca3 KS |
111 | /* |
112 | * Third double word block | |
113 | * | |
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(). | |
117 | */ | |
49e22585 CL |
118 | union { |
119 | struct list_head lru; /* Pageout list, eg. active_list | |
a52633d8 | 120 | * protected by zone_lru_lock ! |
34bf6ef9 DH |
121 | * Can be used as a generic list |
122 | * by the page owner. | |
fc9bb8c7 | 123 | */ |
5c2c2587 DW |
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. | |
128 | */ | |
49e22585 CL |
129 | struct { /* slub per cpu partial pages */ |
130 | struct page *next; /* Next partial slab */ | |
131 | #ifdef CONFIG_64BIT | |
132 | int pages; /* Nr of partial slabs left */ | |
133 | int pobjects; /* Approximate # of objects */ | |
134 | #else | |
135 | short int pages; | |
136 | short int pobjects; | |
137 | #endif | |
138 | }; | |
b8c24c4a | 139 | |
68126702 JK |
140 | struct rcu_head rcu_head; /* Used by SLAB |
141 | * when destroying via RCU | |
142 | */ | |
1d798ca3 | 143 | /* Tail pages of compound page */ |
e4b294c2 | 144 | struct { |
1d798ca3 KS |
145 | unsigned long compound_head; /* If bit zero is set */ |
146 | ||
147 | /* First tail page only */ | |
1965c8b7 KS |
148 | #ifdef CONFIG_64BIT |
149 | /* | |
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. | |
154 | */ | |
155 | unsigned int compound_dtor; | |
156 | unsigned int compound_order; | |
157 | #else | |
f1e61557 KS |
158 | unsigned short int compound_dtor; |
159 | unsigned short int compound_order; | |
1965c8b7 | 160 | #endif |
e4b294c2 KS |
161 | }; |
162 | ||
7aa555bf | 163 | #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && USE_SPLIT_PMD_PTLOCKS |
1d798ca3 KS |
164 | struct { |
165 | unsigned long __pad; /* do not overlay pmd_huge_pte | |
166 | * with compound_head to avoid | |
167 | * possible bit 0 collision. | |
168 | */ | |
169 | pgtable_t pmd_huge_pte; /* protected by page->ptl */ | |
170 | }; | |
7aa555bf | 171 | #endif |
49e22585 | 172 | }; |
fc9bb8c7 CL |
173 | |
174 | /* Remainder is not double word aligned */ | |
5b99cd0e | 175 | union { |
5b99cd0e HC |
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. | |
182 | */ | |
57c1ffce | 183 | #if USE_SPLIT_PTE_PTLOCKS |
597d795a | 184 | #if ALLOC_SPLIT_PTLOCKS |
539edb58 PZ |
185 | spinlock_t *ptl; |
186 | #else | |
187 | spinlock_t ptl; | |
188 | #endif | |
5b99cd0e | 189 | #endif |
1b4f59e3 | 190 | struct kmem_cache *slab_cache; /* SL[AU]B: Pointer to slab */ |
81819f0f | 191 | }; |
fc9bb8c7 | 192 | |
1306a85a JW |
193 | #ifdef CONFIG_MEMCG |
194 | struct mem_cgroup *mem_cgroup; | |
195 | #endif | |
196 | ||
5b99cd0e HC |
197 | /* |
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 ... ;) | |
203 | * | |
204 | * Architectures with slow multiplication can define | |
205 | * WANT_PAGE_VIRTUAL in asm/page.h | |
206 | */ | |
207 | #if defined(WANT_PAGE_VIRTUAL) | |
208 | void *virtual; /* Kernel virtual address (NULL if | |
209 | not kmapped, ie. highmem) */ | |
210 | #endif /* WANT_PAGE_VIRTUAL */ | |
dfec072e | 211 | |
90572890 PZ |
212 | #ifdef LAST_CPUPID_NOT_IN_PAGE_FLAGS |
213 | int _last_cpupid; | |
57e0a030 | 214 | #endif |
fc9bb8c7 CL |
215 | } |
216 | /* | |
43570fd2 HC |
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. | |
fc9bb8c7 | 219 | */ |
43570fd2 HC |
220 | #ifdef CONFIG_HAVE_ALIGNED_STRUCT_PAGE |
221 | __aligned(2 * sizeof(unsigned long)) | |
fc9bb8c7 CL |
222 | #endif |
223 | ; | |
5b99cd0e | 224 | |
b63ae8ca AD |
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) | |
227 | ||
228 | struct page_frag_cache { | |
229 | void * va; | |
230 | #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE) | |
231 | __u16 offset; | |
232 | __u16 size; | |
233 | #else | |
234 | __u32 offset; | |
235 | #endif | |
236 | /* we maintain a pagecount bias, so that we dont dirty cache line | |
0139aa7b | 237 | * containing page->_refcount every time we allocate a fragment. |
b63ae8ca AD |
238 | */ |
239 | unsigned int pagecnt_bias; | |
240 | bool pfmemalloc; | |
241 | }; | |
242 | ||
64b990d2 | 243 | typedef unsigned long vm_flags_t; |
ca16d140 | 244 | |
95bb9fd7 YS |
245 | static inline atomic_t *compound_mapcount_ptr(struct page *page) |
246 | { | |
247 | return &page[1].compound_mapcount; | |
248 | } | |
249 | ||
8feae131 DH |
250 | /* |
251 | * A region containing a mapping of a non-memory backed file under NOMMU | |
252 | * conditions. These are held in a global tree and are pinned by the VMAs that | |
253 | * map parts of them. | |
254 | */ | |
255 | struct vm_region { | |
256 | struct rb_node vm_rb; /* link in global region tree */ | |
ca16d140 | 257 | vm_flags_t vm_flags; /* VMA vm_flags */ |
8feae131 DH |
258 | unsigned long vm_start; /* start address of region */ |
259 | unsigned long vm_end; /* region initialised to here */ | |
dd8632a1 | 260 | unsigned long vm_top; /* region allocated to here */ |
8feae131 DH |
261 | unsigned long vm_pgoff; /* the offset in vm_file corresponding to vm_start */ |
262 | struct file *vm_file; /* the backing file or NULL */ | |
c088e31d | 263 | struct file *vm_prfile; /* the virtual backing file or NULL */ |
8feae131 | 264 | |
1e2ae599 | 265 | int vm_usage; /* region usage count (access under nommu_region_sem) */ |
cfe79c00 MF |
266 | bool vm_icache_flushed : 1; /* true if the icache has been flushed for |
267 | * this region */ | |
8feae131 DH |
268 | }; |
269 | ||
745f234b AA |
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; | |
274 | }; | |
275 | #else /* CONFIG_USERFAULTFD */ | |
276 | #define NULL_VM_UFFD_CTX ((struct vm_userfaultfd_ctx) {}) | |
277 | struct vm_userfaultfd_ctx {}; | |
278 | #endif /* CONFIG_USERFAULTFD */ | |
279 | ||
c92ff1bd MS |
280 | /* |
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). | |
285 | */ | |
286 | struct vm_area_struct { | |
e4c6bfd2 RR |
287 | /* The first cache line has the info for VMA tree walking. */ |
288 | ||
c92ff1bd MS |
289 | unsigned long vm_start; /* Our start address within vm_mm. */ |
290 | unsigned long vm_end; /* The first byte after our end address | |
291 | within vm_mm. */ | |
292 | ||
293 | /* linked list of VM areas per task, sorted by address */ | |
297c5eee | 294 | struct vm_area_struct *vm_next, *vm_prev; |
c92ff1bd | 295 | |
c92ff1bd MS |
296 | struct rb_node vm_rb; |
297 | ||
d3737187 ML |
298 | /* |
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. | |
303 | */ | |
304 | unsigned long rb_subtree_gap; | |
305 | ||
e4c6bfd2 RR |
306 | /* Second cache line starts here. */ |
307 | ||
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. */ | |
311 | ||
c92ff1bd MS |
312 | /* |
313 | * For areas with an address space and backing store, | |
27ba0644 | 314 | * linkage into the address_space->i_mmap interval tree. |
c92ff1bd | 315 | */ |
ac51b934 KS |
316 | struct { |
317 | struct rb_node rb; | |
318 | unsigned long rb_subtree_last; | |
c92ff1bd MS |
319 | } shared; |
320 | ||
321 | /* | |
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. | |
326 | */ | |
5beb4930 RR |
327 | struct list_head anon_vma_chain; /* Serialized by mmap_sem & |
328 | * page_table_lock */ | |
c92ff1bd MS |
329 | struct anon_vma *anon_vma; /* Serialized by page_table_lock */ |
330 | ||
331 | /* Function pointers to deal with this struct. */ | |
f0f37e2f | 332 | const struct vm_operations_struct *vm_ops; |
c92ff1bd MS |
333 | |
334 | /* Information about our backing store: */ | |
335 | unsigned long vm_pgoff; /* Offset (within vm_file) in PAGE_SIZE | |
ea1754a0 | 336 | units */ |
c92ff1bd | 337 | struct file * vm_file; /* File we map to (can be NULL). */ |
c088e31d | 338 | struct file *vm_prfile; /* shadow of vm_file */ |
c92ff1bd | 339 | void * vm_private_data; /* was vm_pte (shared mem) */ |
c92ff1bd | 340 | |
ec560175 | 341 | atomic_long_t swap_readahead_info; |
c92ff1bd | 342 | #ifndef CONFIG_MMU |
8feae131 | 343 | struct vm_region *vm_region; /* NOMMU mapping region */ |
c92ff1bd MS |
344 | #endif |
345 | #ifdef CONFIG_NUMA | |
346 | struct mempolicy *vm_policy; /* NUMA policy for the VMA */ | |
347 | #endif | |
745f234b | 348 | struct vm_userfaultfd_ctx vm_userfaultfd_ctx; |
3859a271 | 349 | } __randomize_layout; |
c92ff1bd | 350 | |
b564daf8 ON |
351 | struct core_thread { |
352 | struct task_struct *task; | |
353 | struct core_thread *next; | |
354 | }; | |
355 | ||
32ecb1f2 | 356 | struct core_state { |
c5f1cc8c | 357 | atomic_t nr_threads; |
b564daf8 | 358 | struct core_thread dumper; |
32ecb1f2 ON |
359 | struct completion startup; |
360 | }; | |
361 | ||
db446a08 | 362 | struct kioctx_table; |
c92ff1bd | 363 | struct mm_struct { |
615d6e87 | 364 | struct vm_area_struct *mmap; /* list of VMAs */ |
c92ff1bd | 365 | struct rb_root mm_rb; |
656f77fa | 366 | u64 vmacache_seqnum; /* per-thread vmacache */ |
efc1a3b1 | 367 | #ifdef CONFIG_MMU |
c92ff1bd MS |
368 | unsigned long (*get_unmapped_area) (struct file *filp, |
369 | unsigned long addr, unsigned long len, | |
370 | unsigned long pgoff, unsigned long flags); | |
efc1a3b1 | 371 | #endif |
c92ff1bd | 372 | unsigned long mmap_base; /* base of mmap area */ |
41aacc1e | 373 | unsigned long mmap_legacy_base; /* base of mmap area in bottom-up allocations */ |
1b028f78 DS |
374 | #ifdef CONFIG_HAVE_ARCH_COMPAT_MMAP_BASES |
375 | /* Base adresses for compatible mmap() */ | |
376 | unsigned long mmap_compat_base; | |
377 | unsigned long mmap_compat_legacy_base; | |
378 | #endif | |
c92ff1bd | 379 | unsigned long task_size; /* size of task vm space */ |
d3737187 | 380 | unsigned long highest_vm_end; /* highest vma end address */ |
c92ff1bd | 381 | pgd_t * pgd; |
b279ddc3 VN |
382 | |
383 | /** | |
384 | * @mm_users: The number of users including userspace. | |
385 | * | |
386 | * Use mmget()/mmget_not_zero()/mmput() to modify. When this drops | |
387 | * to 0 (i.e. when the task exits and there are no other temporary | |
388 | * reference holders), we also release a reference on @mm_count | |
389 | * (which may then free the &struct mm_struct if @mm_count also | |
390 | * drops to 0). | |
391 | */ | |
392 | atomic_t mm_users; | |
393 | ||
394 | /** | |
395 | * @mm_count: The number of references to &struct mm_struct | |
396 | * (@mm_users count as 1). | |
397 | * | |
398 | * Use mmgrab()/mmdrop() to modify. When this drops to 0, the | |
399 | * &struct mm_struct is freed. | |
400 | */ | |
401 | atomic_t mm_count; | |
402 | ||
c4812909 | 403 | #ifdef CONFIG_MMU |
af5b0f6a | 404 | atomic_long_t pgtables_bytes; /* PTE page table pages */ |
5a3fbef3 | 405 | #endif |
c92ff1bd | 406 | int map_count; /* number of VMAs */ |
481b4bb5 | 407 | |
c92ff1bd | 408 | spinlock_t page_table_lock; /* Protects page tables and some counters */ |
481b4bb5 | 409 | struct rw_semaphore mmap_sem; |
c92ff1bd MS |
410 | |
411 | struct list_head mmlist; /* List of maybe swapped mm's. These are globally strung | |
412 | * together off init_mm.mmlist, and are protected | |
413 | * by mmlist_lock | |
414 | */ | |
415 | ||
c92ff1bd MS |
416 | |
417 | unsigned long hiwater_rss; /* High-watermark of RSS usage */ | |
418 | unsigned long hiwater_vm; /* High-water virtual memory usage */ | |
419 | ||
e10d59f2 CL |
420 | unsigned long total_vm; /* Total pages mapped */ |
421 | unsigned long locked_vm; /* Pages that have PG_mlocked set */ | |
422 | unsigned long pinned_vm; /* Refcount permanently increased */ | |
30bdbb78 KK |
423 | unsigned long data_vm; /* VM_WRITE & ~VM_SHARED & ~VM_STACK */ |
424 | unsigned long exec_vm; /* VM_EXEC & ~VM_WRITE & ~VM_STACK */ | |
425 | unsigned long stack_vm; /* VM_STACK */ | |
e10d59f2 | 426 | unsigned long def_flags; |
c92ff1bd MS |
427 | unsigned long start_code, end_code, start_data, end_data; |
428 | unsigned long start_brk, brk, start_stack; | |
429 | unsigned long arg_start, arg_end, env_start, env_end; | |
430 | ||
431 | unsigned long saved_auxv[AT_VECTOR_SIZE]; /* for /proc/PID/auxv */ | |
432 | ||
d559db08 KH |
433 | /* |
434 | * Special counters, in some configurations protected by the | |
435 | * page_table_lock, in other configurations by being atomic. | |
436 | */ | |
437 | struct mm_rss_stat rss_stat; | |
438 | ||
801460d0 HS |
439 | struct linux_binfmt *binfmt; |
440 | ||
6345d24d LT |
441 | cpumask_var_t cpu_vm_mask_var; |
442 | ||
c92ff1bd MS |
443 | /* Architecture-specific MM context */ |
444 | mm_context_t context; | |
445 | ||
c92ff1bd MS |
446 | unsigned long flags; /* Must use atomic bitops to access the bits */ |
447 | ||
a94e2d40 | 448 | struct core_state *core_state; /* coredumping support */ |
a961e409 MD |
449 | #ifdef CONFIG_MEMBARRIER |
450 | atomic_t membarrier_state; | |
451 | #endif | |
858f0993 | 452 | #ifdef CONFIG_AIO |
db446a08 BL |
453 | spinlock_t ioctx_lock; |
454 | struct kioctx_table __rcu *ioctx_table; | |
858f0993 | 455 | #endif |
f98bafa0 | 456 | #ifdef CONFIG_MEMCG |
4cd1a8fc KM |
457 | /* |
458 | * "owner" points to a task that is regarded as the canonical | |
459 | * user/owner of this mm. All of the following must be true in | |
460 | * order for it to be changed: | |
461 | * | |
462 | * current == mm->owner | |
463 | * current->mm != mm | |
464 | * new_owner->mm == mm | |
465 | * new_owner->alloc_lock is held | |
466 | */ | |
4d2deb40 | 467 | struct task_struct __rcu *owner; |
78fb7466 | 468 | #endif |
bfedb589 | 469 | struct user_namespace *user_ns; |
925d1c40 | 470 | |
925d1c40 | 471 | /* store ref to file /proc/<pid>/exe symlink points to */ |
90f31d0e | 472 | struct file __rcu *exe_file; |
cddb8a5c AA |
473 | #ifdef CONFIG_MMU_NOTIFIER |
474 | struct mmu_notifier_mm *mmu_notifier_mm; | |
e7a00c45 | 475 | #endif |
e009bb30 | 476 | #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS |
e7a00c45 | 477 | pgtable_t pmd_huge_pte; /* protected by page_table_lock */ |
cddb8a5c | 478 | #endif |
6345d24d LT |
479 | #ifdef CONFIG_CPUMASK_OFFSTACK |
480 | struct cpumask cpumask_allocation; | |
cbee9f88 PZ |
481 | #endif |
482 | #ifdef CONFIG_NUMA_BALANCING | |
483 | /* | |
34f0315a MG |
484 | * numa_next_scan is the next time that the PTEs will be marked |
485 | * pte_numa. NUMA hinting faults will gather statistics and migrate | |
486 | * pages to new nodes if necessary. | |
cbee9f88 PZ |
487 | */ |
488 | unsigned long numa_next_scan; | |
489 | ||
6e5fb223 PZ |
490 | /* Restart point for scanning and setting pte_numa */ |
491 | unsigned long numa_scan_offset; | |
492 | ||
cbee9f88 PZ |
493 | /* numa_scan_seq prevents two threads setting pte_numa */ |
494 | int numa_scan_seq; | |
20841405 | 495 | #endif |
20841405 RR |
496 | /* |
497 | * An operation with batched TLB flushing is going on. Anything that | |
498 | * can move process memory needs to flush the TLB when moving a | |
499 | * PROT_NONE or PROT_NUMA mapped page. | |
500 | */ | |
16af97dc | 501 | atomic_t tlb_flush_pending; |
3ea27719 MG |
502 | #ifdef CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH |
503 | /* See flush_tlb_batched_pending() */ | |
504 | bool tlb_flush_batched; | |
6345d24d | 505 | #endif |
d4b3b638 | 506 | struct uprobes_state uprobes_state; |
5d317b2b NH |
507 | #ifdef CONFIG_HUGETLB_PAGE |
508 | atomic_long_t hugetlb_usage; | |
509 | #endif | |
ec8d7c14 | 510 | struct work_struct async_put_work; |
133ff0ea JG |
511 | |
512 | #if IS_ENABLED(CONFIG_HMM) | |
513 | /* HMM needs to track a few things per mm */ | |
514 | struct hmm *hmm; | |
515 | #endif | |
3859a271 | 516 | } __randomize_layout; |
c92ff1bd | 517 | |
abe722a1 IM |
518 | extern struct mm_struct init_mm; |
519 | ||
6345d24d LT |
520 | static inline void mm_init_cpumask(struct mm_struct *mm) |
521 | { | |
522 | #ifdef CONFIG_CPUMASK_OFFSTACK | |
523 | mm->cpu_vm_mask_var = &mm->cpumask_allocation; | |
524 | #endif | |
41f727fd | 525 | cpumask_clear(mm->cpu_vm_mask_var); |
6345d24d LT |
526 | } |
527 | ||
45e575ab | 528 | /* Future-safe accessor for struct mm_struct's cpu_vm_mask. */ |
de03c72c KM |
529 | static inline cpumask_t *mm_cpumask(struct mm_struct *mm) |
530 | { | |
531 | return mm->cpu_vm_mask_var; | |
532 | } | |
45e575ab | 533 | |
56236a59 MK |
534 | struct mmu_gather; |
535 | extern void tlb_gather_mmu(struct mmu_gather *tlb, struct mm_struct *mm, | |
536 | unsigned long start, unsigned long end); | |
537 | extern void tlb_finish_mmu(struct mmu_gather *tlb, | |
538 | unsigned long start, unsigned long end); | |
539 | ||
16af97dc | 540 | static inline void init_tlb_flush_pending(struct mm_struct *mm) |
20841405 | 541 | { |
16af97dc | 542 | atomic_set(&mm->tlb_flush_pending, 0); |
20841405 | 543 | } |
16af97dc NA |
544 | |
545 | static inline void inc_tlb_flush_pending(struct mm_struct *mm) | |
20841405 | 546 | { |
16af97dc | 547 | atomic_inc(&mm->tlb_flush_pending); |
af2c1401 | 548 | /* |
8b1b436d PZ |
549 | * The only time this value is relevant is when there are indeed pages |
550 | * to flush. And we'll only flush pages after changing them, which | |
551 | * requires the PTL. | |
552 | * | |
553 | * So the ordering here is: | |
554 | * | |
040cca3a | 555 | * atomic_inc(&mm->tlb_flush_pending); |
8b1b436d PZ |
556 | * spin_lock(&ptl); |
557 | * ... | |
558 | * set_pte_at(); | |
559 | * spin_unlock(&ptl); | |
560 | * | |
561 | * spin_lock(&ptl) | |
562 | * mm_tlb_flush_pending(); | |
563 | * .... | |
564 | * spin_unlock(&ptl); | |
565 | * | |
566 | * flush_tlb_range(); | |
040cca3a | 567 | * atomic_dec(&mm->tlb_flush_pending); |
8b1b436d | 568 | * |
0e709703 PZ |
569 | * Where the increment if constrained by the PTL unlock, it thus |
570 | * ensures that the increment is visible if the PTE modification is | |
571 | * visible. After all, if there is no PTE modification, nobody cares | |
572 | * about TLB flushes either. | |
573 | * | |
574 | * This very much relies on users (mm_tlb_flush_pending() and | |
575 | * mm_tlb_flush_nested()) only caring about _specific_ PTEs (and | |
576 | * therefore specific PTLs), because with SPLIT_PTE_PTLOCKS and RCpc | |
577 | * locks (PPC) the unlock of one doesn't order against the lock of | |
578 | * another PTL. | |
579 | * | |
580 | * The decrement is ordered by the flush_tlb_range(), such that | |
581 | * mm_tlb_flush_pending() will not return false unless all flushes have | |
582 | * completed. | |
af2c1401 | 583 | */ |
20841405 | 584 | } |
16af97dc | 585 | |
16af97dc | 586 | static inline void dec_tlb_flush_pending(struct mm_struct *mm) |
20841405 | 587 | { |
0a2c4048 | 588 | /* |
0e709703 PZ |
589 | * See inc_tlb_flush_pending(). |
590 | * | |
591 | * This cannot be smp_mb__before_atomic() because smp_mb() simply does | |
592 | * not order against TLB invalidate completion, which is what we need. | |
593 | * | |
594 | * Therefore we must rely on tlb_flush_*() to guarantee order. | |
0a2c4048 | 595 | */ |
16af97dc | 596 | atomic_dec(&mm->tlb_flush_pending); |
20841405 | 597 | } |
20841405 | 598 | |
0e709703 PZ |
599 | static inline bool mm_tlb_flush_pending(struct mm_struct *mm) |
600 | { | |
601 | /* | |
602 | * Must be called after having acquired the PTL; orders against that | |
603 | * PTLs release and therefore ensures that if we observe the modified | |
604 | * PTE we must also observe the increment from inc_tlb_flush_pending(). | |
605 | * | |
606 | * That is, it only guarantees to return true if there is a flush | |
607 | * pending for _this_ PTL. | |
608 | */ | |
609 | return atomic_read(&mm->tlb_flush_pending); | |
610 | } | |
611 | ||
612 | static inline bool mm_tlb_flush_nested(struct mm_struct *mm) | |
613 | { | |
614 | /* | |
615 | * Similar to mm_tlb_flush_pending(), we must have acquired the PTL | |
616 | * for which there is a TLB flush pending in order to guarantee | |
617 | * we've seen both that PTE modification and the increment. | |
618 | * | |
619 | * (no requirement on actually still holding the PTL, that is irrelevant) | |
620 | */ | |
621 | return atomic_read(&mm->tlb_flush_pending) > 1; | |
622 | } | |
623 | ||
f872f540 AL |
624 | struct vm_fault; |
625 | ||
626 | struct vm_special_mapping { | |
627 | const char *name; /* The name, e.g. "[vdso]". */ | |
628 | ||
629 | /* | |
630 | * If .fault is not provided, this points to a | |
631 | * NULL-terminated array of pages that back the special mapping. | |
632 | * | |
633 | * This must not be NULL unless .fault is provided. | |
634 | */ | |
a62c34bd | 635 | struct page **pages; |
f872f540 AL |
636 | |
637 | /* | |
638 | * If non-NULL, then this is called to resolve page faults | |
639 | * on the special mapping. If used, .pages is not checked. | |
640 | */ | |
641 | int (*fault)(const struct vm_special_mapping *sm, | |
642 | struct vm_area_struct *vma, | |
643 | struct vm_fault *vmf); | |
b059a453 DS |
644 | |
645 | int (*mremap)(const struct vm_special_mapping *sm, | |
646 | struct vm_area_struct *new_vma); | |
a62c34bd AL |
647 | }; |
648 | ||
d17d8f9d DH |
649 | enum tlb_flush_reason { |
650 | TLB_FLUSH_ON_TASK_SWITCH, | |
651 | TLB_REMOTE_SHOOTDOWN, | |
652 | TLB_LOCAL_SHOOTDOWN, | |
653 | TLB_LOCAL_MM_SHOOTDOWN, | |
5b74283a | 654 | TLB_REMOTE_SEND_IPI, |
d17d8f9d DH |
655 | NR_TLB_FLUSH_REASONS, |
656 | }; | |
657 | ||
bd6dace7 TH |
658 | /* |
659 | * A swap entry has to fit into a "unsigned long", as the entry is hidden | |
660 | * in the "index" field of the swapper address space. | |
661 | */ | |
662 | typedef struct { | |
663 | unsigned long val; | |
664 | } swp_entry_t; | |
665 | ||
5b99cd0e | 666 | #endif /* _LINUX_MM_TYPES_H */ |