]>
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 | |
51 | * it points to anon_vma object: | |
52 | * see PAGE_MAPPING_ANON below. | |
53 | */ | |
54 | void *s_mem; /* slab first object */ | |
53f9263b | 55 | atomic_t compound_mapcount; /* first tail page */ |
9a982250 | 56 | /* page_deferred_list().next -- second tail page */ |
8456a648 JK |
57 | }; |
58 | ||
fc9bb8c7 | 59 | /* Second double word */ |
99691add VD |
60 | union { |
61 | pgoff_t index; /* Our offset within mapping. */ | |
62 | void *freelist; /* sl[aou]b first free object */ | |
63 | /* page_deferred_list().prev -- second tail page */ | |
64 | }; | |
013e8963 | 65 | |
99691add | 66 | union { |
abca7c49 PS |
67 | #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \ |
68 | defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE) | |
99691add VD |
69 | /* Used for cmpxchg_double in slub */ |
70 | unsigned long counters; | |
abca7c49 | 71 | #else |
99691add VD |
72 | /* |
73 | * Keep _refcount separate from slub cmpxchg_double data. | |
74 | * As the rest of the double word is protected by slab_lock | |
75 | * but _refcount is not. | |
76 | */ | |
77 | unsigned counters; | |
abca7c49 | 78 | #endif |
99691add | 79 | struct { |
013e8963 | 80 | |
99691add | 81 | union { |
0139aa7b | 82 | /* |
99691add VD |
83 | * Count of ptes mapped in mms, to show when |
84 | * page is mapped & limit reverse map searches. | |
632c0a1a VD |
85 | * |
86 | * Extra information about page type may be | |
87 | * stored here for pages that are never mapped, | |
88 | * in which case the value MUST BE <= -2. | |
89 | * See page-flags.h for more details. | |
0139aa7b | 90 | */ |
99691add VD |
91 | atomic_t _mapcount; |
92 | ||
93 | unsigned int active; /* SLAB */ | |
94 | struct { /* SLUB */ | |
95 | unsigned inuse:16; | |
96 | unsigned objects:15; | |
97 | unsigned frozen:1; | |
98 | }; | |
99 | int units; /* SLOB */ | |
fc9bb8c7 | 100 | }; |
99691add VD |
101 | /* |
102 | * Usage count, *USE WRAPPER FUNCTION* when manual | |
103 | * accounting. See page_ref.h | |
104 | */ | |
105 | atomic_t _refcount; | |
39b26464 | 106 | }; |
81819f0f | 107 | }; |
fc9bb8c7 | 108 | |
1d798ca3 KS |
109 | /* |
110 | * Third double word block | |
111 | * | |
112 | * WARNING: bit 0 of the first word encode PageTail(). That means | |
113 | * the rest users of the storage space MUST NOT use the bit to | |
114 | * avoid collision and false-positive PageTail(). | |
115 | */ | |
49e22585 CL |
116 | union { |
117 | struct list_head lru; /* Pageout list, eg. active_list | |
a52633d8 | 118 | * protected by zone_lru_lock ! |
34bf6ef9 DH |
119 | * Can be used as a generic list |
120 | * by the page owner. | |
fc9bb8c7 | 121 | */ |
5c2c2587 DW |
122 | struct dev_pagemap *pgmap; /* ZONE_DEVICE pages are never on an |
123 | * lru or handled by a slab | |
124 | * allocator, this points to the | |
125 | * hosting device page map. | |
126 | */ | |
49e22585 CL |
127 | struct { /* slub per cpu partial pages */ |
128 | struct page *next; /* Next partial slab */ | |
129 | #ifdef CONFIG_64BIT | |
130 | int pages; /* Nr of partial slabs left */ | |
131 | int pobjects; /* Approximate # of objects */ | |
132 | #else | |
133 | short int pages; | |
134 | short int pobjects; | |
135 | #endif | |
136 | }; | |
b8c24c4a | 137 | |
68126702 JK |
138 | struct rcu_head rcu_head; /* Used by SLAB |
139 | * when destroying via RCU | |
140 | */ | |
1d798ca3 | 141 | /* Tail pages of compound page */ |
e4b294c2 | 142 | struct { |
1d798ca3 KS |
143 | unsigned long compound_head; /* If bit zero is set */ |
144 | ||
145 | /* First tail page only */ | |
1965c8b7 KS |
146 | #ifdef CONFIG_64BIT |
147 | /* | |
148 | * On 64 bit system we have enough space in struct page | |
149 | * to encode compound_dtor and compound_order with | |
150 | * unsigned int. It can help compiler generate better or | |
151 | * smaller code on some archtectures. | |
152 | */ | |
153 | unsigned int compound_dtor; | |
154 | unsigned int compound_order; | |
155 | #else | |
f1e61557 KS |
156 | unsigned short int compound_dtor; |
157 | unsigned short int compound_order; | |
1965c8b7 | 158 | #endif |
e4b294c2 KS |
159 | }; |
160 | ||
7aa555bf | 161 | #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && USE_SPLIT_PMD_PTLOCKS |
1d798ca3 KS |
162 | struct { |
163 | unsigned long __pad; /* do not overlay pmd_huge_pte | |
164 | * with compound_head to avoid | |
165 | * possible bit 0 collision. | |
166 | */ | |
167 | pgtable_t pmd_huge_pte; /* protected by page->ptl */ | |
168 | }; | |
7aa555bf | 169 | #endif |
49e22585 | 170 | }; |
fc9bb8c7 CL |
171 | |
172 | /* Remainder is not double word aligned */ | |
5b99cd0e | 173 | union { |
5b99cd0e HC |
174 | unsigned long private; /* Mapping-private opaque data: |
175 | * usually used for buffer_heads | |
176 | * if PagePrivate set; used for | |
177 | * swp_entry_t if PageSwapCache; | |
178 | * indicates order in the buddy | |
179 | * system if PG_buddy is set. | |
180 | */ | |
57c1ffce | 181 | #if USE_SPLIT_PTE_PTLOCKS |
597d795a | 182 | #if ALLOC_SPLIT_PTLOCKS |
539edb58 PZ |
183 | spinlock_t *ptl; |
184 | #else | |
185 | spinlock_t ptl; | |
186 | #endif | |
5b99cd0e | 187 | #endif |
1b4f59e3 | 188 | struct kmem_cache *slab_cache; /* SL[AU]B: Pointer to slab */ |
81819f0f | 189 | }; |
fc9bb8c7 | 190 | |
1306a85a JW |
191 | #ifdef CONFIG_MEMCG |
192 | struct mem_cgroup *mem_cgroup; | |
193 | #endif | |
194 | ||
5b99cd0e HC |
195 | /* |
196 | * On machines where all RAM is mapped into kernel address space, | |
197 | * we can simply calculate the virtual address. On machines with | |
198 | * highmem some memory is mapped into kernel virtual memory | |
199 | * dynamically, so we need a place to store that address. | |
200 | * Note that this field could be 16 bits on x86 ... ;) | |
201 | * | |
202 | * Architectures with slow multiplication can define | |
203 | * WANT_PAGE_VIRTUAL in asm/page.h | |
204 | */ | |
205 | #if defined(WANT_PAGE_VIRTUAL) | |
206 | void *virtual; /* Kernel virtual address (NULL if | |
207 | not kmapped, ie. highmem) */ | |
208 | #endif /* WANT_PAGE_VIRTUAL */ | |
dfec072e | 209 | |
90572890 PZ |
210 | #ifdef LAST_CPUPID_NOT_IN_PAGE_FLAGS |
211 | int _last_cpupid; | |
57e0a030 | 212 | #endif |
fc9bb8c7 CL |
213 | } |
214 | /* | |
43570fd2 HC |
215 | * The struct page can be forced to be double word aligned so that atomic ops |
216 | * on double words work. The SLUB allocator can make use of such a feature. | |
fc9bb8c7 | 217 | */ |
43570fd2 HC |
218 | #ifdef CONFIG_HAVE_ALIGNED_STRUCT_PAGE |
219 | __aligned(2 * sizeof(unsigned long)) | |
fc9bb8c7 CL |
220 | #endif |
221 | ; | |
5b99cd0e | 222 | |
b63ae8ca AD |
223 | #define PAGE_FRAG_CACHE_MAX_SIZE __ALIGN_MASK(32768, ~PAGE_MASK) |
224 | #define PAGE_FRAG_CACHE_MAX_ORDER get_order(PAGE_FRAG_CACHE_MAX_SIZE) | |
225 | ||
226 | struct page_frag_cache { | |
227 | void * va; | |
228 | #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE) | |
229 | __u16 offset; | |
230 | __u16 size; | |
231 | #else | |
232 | __u32 offset; | |
233 | #endif | |
234 | /* we maintain a pagecount bias, so that we dont dirty cache line | |
0139aa7b | 235 | * containing page->_refcount every time we allocate a fragment. |
b63ae8ca AD |
236 | */ |
237 | unsigned int pagecnt_bias; | |
238 | bool pfmemalloc; | |
239 | }; | |
240 | ||
64b990d2 | 241 | typedef unsigned long vm_flags_t; |
ca16d140 | 242 | |
8feae131 DH |
243 | /* |
244 | * A region containing a mapping of a non-memory backed file under NOMMU | |
245 | * conditions. These are held in a global tree and are pinned by the VMAs that | |
246 | * map parts of them. | |
247 | */ | |
248 | struct vm_region { | |
249 | struct rb_node vm_rb; /* link in global region tree */ | |
ca16d140 | 250 | vm_flags_t vm_flags; /* VMA vm_flags */ |
8feae131 DH |
251 | unsigned long vm_start; /* start address of region */ |
252 | unsigned long vm_end; /* region initialised to here */ | |
dd8632a1 | 253 | unsigned long vm_top; /* region allocated to here */ |
8feae131 DH |
254 | unsigned long vm_pgoff; /* the offset in vm_file corresponding to vm_start */ |
255 | struct file *vm_file; /* the backing file or NULL */ | |
256 | ||
1e2ae599 | 257 | int vm_usage; /* region usage count (access under nommu_region_sem) */ |
cfe79c00 MF |
258 | bool vm_icache_flushed : 1; /* true if the icache has been flushed for |
259 | * this region */ | |
8feae131 DH |
260 | }; |
261 | ||
745f234b AA |
262 | #ifdef CONFIG_USERFAULTFD |
263 | #define NULL_VM_UFFD_CTX ((struct vm_userfaultfd_ctx) { NULL, }) | |
264 | struct vm_userfaultfd_ctx { | |
265 | struct userfaultfd_ctx *ctx; | |
266 | }; | |
267 | #else /* CONFIG_USERFAULTFD */ | |
268 | #define NULL_VM_UFFD_CTX ((struct vm_userfaultfd_ctx) {}) | |
269 | struct vm_userfaultfd_ctx {}; | |
270 | #endif /* CONFIG_USERFAULTFD */ | |
271 | ||
c92ff1bd MS |
272 | /* |
273 | * This struct defines a memory VMM memory area. There is one of these | |
274 | * per VM-area/task. A VM area is any part of the process virtual memory | |
275 | * space that has a special rule for the page-fault handlers (ie a shared | |
276 | * library, the executable area etc). | |
277 | */ | |
278 | struct vm_area_struct { | |
e4c6bfd2 RR |
279 | /* The first cache line has the info for VMA tree walking. */ |
280 | ||
c92ff1bd MS |
281 | unsigned long vm_start; /* Our start address within vm_mm. */ |
282 | unsigned long vm_end; /* The first byte after our end address | |
283 | within vm_mm. */ | |
284 | ||
285 | /* linked list of VM areas per task, sorted by address */ | |
297c5eee | 286 | struct vm_area_struct *vm_next, *vm_prev; |
c92ff1bd | 287 | |
c92ff1bd MS |
288 | struct rb_node vm_rb; |
289 | ||
d3737187 ML |
290 | /* |
291 | * Largest free memory gap in bytes to the left of this VMA. | |
292 | * Either between this VMA and vma->vm_prev, or between one of the | |
293 | * VMAs below us in the VMA rbtree and its ->vm_prev. This helps | |
294 | * get_unmapped_area find a free area of the right size. | |
295 | */ | |
296 | unsigned long rb_subtree_gap; | |
297 | ||
e4c6bfd2 RR |
298 | /* Second cache line starts here. */ |
299 | ||
300 | struct mm_struct *vm_mm; /* The address space we belong to. */ | |
301 | pgprot_t vm_page_prot; /* Access permissions of this VMA. */ | |
302 | unsigned long vm_flags; /* Flags, see mm.h. */ | |
303 | ||
c92ff1bd MS |
304 | /* |
305 | * For areas with an address space and backing store, | |
27ba0644 | 306 | * linkage into the address_space->i_mmap interval tree. |
c92ff1bd | 307 | */ |
ac51b934 KS |
308 | struct { |
309 | struct rb_node rb; | |
310 | unsigned long rb_subtree_last; | |
c92ff1bd MS |
311 | } shared; |
312 | ||
313 | /* | |
314 | * A file's MAP_PRIVATE vma can be in both i_mmap tree and anon_vma | |
315 | * list, after a COW of one of the file pages. A MAP_SHARED vma | |
316 | * can only be in the i_mmap tree. An anonymous MAP_PRIVATE, stack | |
317 | * or brk vma (with NULL file) can only be in an anon_vma list. | |
318 | */ | |
5beb4930 RR |
319 | struct list_head anon_vma_chain; /* Serialized by mmap_sem & |
320 | * page_table_lock */ | |
c92ff1bd MS |
321 | struct anon_vma *anon_vma; /* Serialized by page_table_lock */ |
322 | ||
323 | /* Function pointers to deal with this struct. */ | |
f0f37e2f | 324 | const struct vm_operations_struct *vm_ops; |
c92ff1bd MS |
325 | |
326 | /* Information about our backing store: */ | |
327 | unsigned long vm_pgoff; /* Offset (within vm_file) in PAGE_SIZE | |
ea1754a0 | 328 | units */ |
c92ff1bd MS |
329 | struct file * vm_file; /* File we map to (can be NULL). */ |
330 | void * vm_private_data; /* was vm_pte (shared mem) */ | |
c92ff1bd | 331 | |
ec560175 | 332 | atomic_long_t swap_readahead_info; |
c92ff1bd | 333 | #ifndef CONFIG_MMU |
8feae131 | 334 | struct vm_region *vm_region; /* NOMMU mapping region */ |
c92ff1bd MS |
335 | #endif |
336 | #ifdef CONFIG_NUMA | |
337 | struct mempolicy *vm_policy; /* NUMA policy for the VMA */ | |
338 | #endif | |
745f234b | 339 | struct vm_userfaultfd_ctx vm_userfaultfd_ctx; |
3859a271 | 340 | } __randomize_layout; |
c92ff1bd | 341 | |
b564daf8 ON |
342 | struct core_thread { |
343 | struct task_struct *task; | |
344 | struct core_thread *next; | |
345 | }; | |
346 | ||
32ecb1f2 | 347 | struct core_state { |
c5f1cc8c | 348 | atomic_t nr_threads; |
b564daf8 | 349 | struct core_thread dumper; |
32ecb1f2 ON |
350 | struct completion startup; |
351 | }; | |
352 | ||
db446a08 | 353 | struct kioctx_table; |
c92ff1bd | 354 | struct mm_struct { |
615d6e87 | 355 | struct vm_area_struct *mmap; /* list of VMAs */ |
c92ff1bd | 356 | struct rb_root mm_rb; |
06274364 | 357 | u64 vmacache_seqnum; /* per-thread vmacache */ |
efc1a3b1 | 358 | #ifdef CONFIG_MMU |
c92ff1bd MS |
359 | unsigned long (*get_unmapped_area) (struct file *filp, |
360 | unsigned long addr, unsigned long len, | |
361 | unsigned long pgoff, unsigned long flags); | |
efc1a3b1 | 362 | #endif |
c92ff1bd | 363 | unsigned long mmap_base; /* base of mmap area */ |
41aacc1e | 364 | unsigned long mmap_legacy_base; /* base of mmap area in bottom-up allocations */ |
1b028f78 DS |
365 | #ifdef CONFIG_HAVE_ARCH_COMPAT_MMAP_BASES |
366 | /* Base adresses for compatible mmap() */ | |
367 | unsigned long mmap_compat_base; | |
368 | unsigned long mmap_compat_legacy_base; | |
369 | #endif | |
c92ff1bd | 370 | unsigned long task_size; /* size of task vm space */ |
d3737187 | 371 | unsigned long highest_vm_end; /* highest vma end address */ |
c92ff1bd | 372 | pgd_t * pgd; |
b279ddc3 VN |
373 | |
374 | /** | |
375 | * @mm_users: The number of users including userspace. | |
376 | * | |
377 | * Use mmget()/mmget_not_zero()/mmput() to modify. When this drops | |
378 | * to 0 (i.e. when the task exits and there are no other temporary | |
379 | * reference holders), we also release a reference on @mm_count | |
380 | * (which may then free the &struct mm_struct if @mm_count also | |
381 | * drops to 0). | |
382 | */ | |
383 | atomic_t mm_users; | |
384 | ||
385 | /** | |
386 | * @mm_count: The number of references to &struct mm_struct | |
387 | * (@mm_users count as 1). | |
388 | * | |
389 | * Use mmgrab()/mmdrop() to modify. When this drops to 0, the | |
390 | * &struct mm_struct is freed. | |
391 | */ | |
392 | atomic_t mm_count; | |
393 | ||
dc6c9a35 | 394 | atomic_long_t nr_ptes; /* PTE page table pages */ |
5a3fbef3 | 395 | #if CONFIG_PGTABLE_LEVELS > 2 |
dc6c9a35 | 396 | atomic_long_t nr_pmds; /* PMD page table pages */ |
5a3fbef3 | 397 | #endif |
c92ff1bd | 398 | int map_count; /* number of VMAs */ |
481b4bb5 | 399 | |
c92ff1bd | 400 | spinlock_t page_table_lock; /* Protects page tables and some counters */ |
481b4bb5 | 401 | struct rw_semaphore mmap_sem; |
c92ff1bd MS |
402 | |
403 | struct list_head mmlist; /* List of maybe swapped mm's. These are globally strung | |
404 | * together off init_mm.mmlist, and are protected | |
405 | * by mmlist_lock | |
406 | */ | |
407 | ||
c92ff1bd MS |
408 | |
409 | unsigned long hiwater_rss; /* High-watermark of RSS usage */ | |
410 | unsigned long hiwater_vm; /* High-water virtual memory usage */ | |
411 | ||
e10d59f2 CL |
412 | unsigned long total_vm; /* Total pages mapped */ |
413 | unsigned long locked_vm; /* Pages that have PG_mlocked set */ | |
414 | unsigned long pinned_vm; /* Refcount permanently increased */ | |
30bdbb78 KK |
415 | unsigned long data_vm; /* VM_WRITE & ~VM_SHARED & ~VM_STACK */ |
416 | unsigned long exec_vm; /* VM_EXEC & ~VM_WRITE & ~VM_STACK */ | |
417 | unsigned long stack_vm; /* VM_STACK */ | |
e10d59f2 | 418 | unsigned long def_flags; |
c92ff1bd MS |
419 | unsigned long start_code, end_code, start_data, end_data; |
420 | unsigned long start_brk, brk, start_stack; | |
421 | unsigned long arg_start, arg_end, env_start, env_end; | |
422 | ||
423 | unsigned long saved_auxv[AT_VECTOR_SIZE]; /* for /proc/PID/auxv */ | |
424 | ||
d559db08 KH |
425 | /* |
426 | * Special counters, in some configurations protected by the | |
427 | * page_table_lock, in other configurations by being atomic. | |
428 | */ | |
429 | struct mm_rss_stat rss_stat; | |
430 | ||
801460d0 HS |
431 | struct linux_binfmt *binfmt; |
432 | ||
6345d24d LT |
433 | cpumask_var_t cpu_vm_mask_var; |
434 | ||
c92ff1bd MS |
435 | /* Architecture-specific MM context */ |
436 | mm_context_t context; | |
437 | ||
c92ff1bd MS |
438 | unsigned long flags; /* Must use atomic bitops to access the bits */ |
439 | ||
a94e2d40 | 440 | struct core_state *core_state; /* coredumping support */ |
a961e409 MD |
441 | #ifdef CONFIG_MEMBARRIER |
442 | atomic_t membarrier_state; | |
443 | #endif | |
858f0993 | 444 | #ifdef CONFIG_AIO |
db446a08 BL |
445 | spinlock_t ioctx_lock; |
446 | struct kioctx_table __rcu *ioctx_table; | |
858f0993 | 447 | #endif |
f98bafa0 | 448 | #ifdef CONFIG_MEMCG |
4cd1a8fc KM |
449 | /* |
450 | * "owner" points to a task that is regarded as the canonical | |
451 | * user/owner of this mm. All of the following must be true in | |
452 | * order for it to be changed: | |
453 | * | |
454 | * current == mm->owner | |
455 | * current->mm != mm | |
456 | * new_owner->mm == mm | |
457 | * new_owner->alloc_lock is held | |
458 | */ | |
4d2deb40 | 459 | struct task_struct __rcu *owner; |
78fb7466 | 460 | #endif |
bfedb589 | 461 | struct user_namespace *user_ns; |
925d1c40 | 462 | |
925d1c40 | 463 | /* store ref to file /proc/<pid>/exe symlink points to */ |
90f31d0e | 464 | struct file __rcu *exe_file; |
cddb8a5c AA |
465 | #ifdef CONFIG_MMU_NOTIFIER |
466 | struct mmu_notifier_mm *mmu_notifier_mm; | |
e7a00c45 | 467 | #endif |
e009bb30 | 468 | #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS |
e7a00c45 | 469 | pgtable_t pmd_huge_pte; /* protected by page_table_lock */ |
cddb8a5c | 470 | #endif |
6345d24d LT |
471 | #ifdef CONFIG_CPUMASK_OFFSTACK |
472 | struct cpumask cpumask_allocation; | |
cbee9f88 PZ |
473 | #endif |
474 | #ifdef CONFIG_NUMA_BALANCING | |
475 | /* | |
34f0315a MG |
476 | * numa_next_scan is the next time that the PTEs will be marked |
477 | * pte_numa. NUMA hinting faults will gather statistics and migrate | |
478 | * pages to new nodes if necessary. | |
cbee9f88 PZ |
479 | */ |
480 | unsigned long numa_next_scan; | |
481 | ||
6e5fb223 PZ |
482 | /* Restart point for scanning and setting pte_numa */ |
483 | unsigned long numa_scan_offset; | |
484 | ||
cbee9f88 PZ |
485 | /* numa_scan_seq prevents two threads setting pte_numa */ |
486 | int numa_scan_seq; | |
20841405 | 487 | #endif |
20841405 RR |
488 | /* |
489 | * An operation with batched TLB flushing is going on. Anything that | |
490 | * can move process memory needs to flush the TLB when moving a | |
491 | * PROT_NONE or PROT_NUMA mapped page. | |
492 | */ | |
16af97dc | 493 | atomic_t tlb_flush_pending; |
3ea27719 MG |
494 | #ifdef CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH |
495 | /* See flush_tlb_batched_pending() */ | |
496 | bool tlb_flush_batched; | |
6345d24d | 497 | #endif |
d4b3b638 | 498 | struct uprobes_state uprobes_state; |
5d317b2b NH |
499 | #ifdef CONFIG_HUGETLB_PAGE |
500 | atomic_long_t hugetlb_usage; | |
501 | #endif | |
ec8d7c14 | 502 | struct work_struct async_put_work; |
133ff0ea JG |
503 | |
504 | #if IS_ENABLED(CONFIG_HMM) | |
505 | /* HMM needs to track a few things per mm */ | |
506 | struct hmm *hmm; | |
507 | #endif | |
3859a271 | 508 | } __randomize_layout; |
c92ff1bd | 509 | |
abe722a1 IM |
510 | extern struct mm_struct init_mm; |
511 | ||
6345d24d LT |
512 | static inline void mm_init_cpumask(struct mm_struct *mm) |
513 | { | |
514 | #ifdef CONFIG_CPUMASK_OFFSTACK | |
515 | mm->cpu_vm_mask_var = &mm->cpumask_allocation; | |
516 | #endif | |
41f727fd | 517 | cpumask_clear(mm->cpu_vm_mask_var); |
6345d24d LT |
518 | } |
519 | ||
45e575ab | 520 | /* Future-safe accessor for struct mm_struct's cpu_vm_mask. */ |
de03c72c KM |
521 | static inline cpumask_t *mm_cpumask(struct mm_struct *mm) |
522 | { | |
523 | return mm->cpu_vm_mask_var; | |
524 | } | |
45e575ab | 525 | |
56236a59 MK |
526 | struct mmu_gather; |
527 | extern void tlb_gather_mmu(struct mmu_gather *tlb, struct mm_struct *mm, | |
528 | unsigned long start, unsigned long end); | |
529 | extern void tlb_finish_mmu(struct mmu_gather *tlb, | |
530 | unsigned long start, unsigned long end); | |
531 | ||
16af97dc | 532 | static inline void init_tlb_flush_pending(struct mm_struct *mm) |
20841405 | 533 | { |
16af97dc | 534 | atomic_set(&mm->tlb_flush_pending, 0); |
20841405 | 535 | } |
16af97dc NA |
536 | |
537 | static inline void inc_tlb_flush_pending(struct mm_struct *mm) | |
20841405 | 538 | { |
16af97dc | 539 | atomic_inc(&mm->tlb_flush_pending); |
af2c1401 | 540 | /* |
8b1b436d PZ |
541 | * The only time this value is relevant is when there are indeed pages |
542 | * to flush. And we'll only flush pages after changing them, which | |
543 | * requires the PTL. | |
544 | * | |
545 | * So the ordering here is: | |
546 | * | |
040cca3a | 547 | * atomic_inc(&mm->tlb_flush_pending); |
8b1b436d PZ |
548 | * spin_lock(&ptl); |
549 | * ... | |
550 | * set_pte_at(); | |
551 | * spin_unlock(&ptl); | |
552 | * | |
553 | * spin_lock(&ptl) | |
554 | * mm_tlb_flush_pending(); | |
555 | * .... | |
556 | * spin_unlock(&ptl); | |
557 | * | |
558 | * flush_tlb_range(); | |
040cca3a | 559 | * atomic_dec(&mm->tlb_flush_pending); |
8b1b436d | 560 | * |
0e709703 PZ |
561 | * Where the increment if constrained by the PTL unlock, it thus |
562 | * ensures that the increment is visible if the PTE modification is | |
563 | * visible. After all, if there is no PTE modification, nobody cares | |
564 | * about TLB flushes either. | |
565 | * | |
566 | * This very much relies on users (mm_tlb_flush_pending() and | |
567 | * mm_tlb_flush_nested()) only caring about _specific_ PTEs (and | |
568 | * therefore specific PTLs), because with SPLIT_PTE_PTLOCKS and RCpc | |
569 | * locks (PPC) the unlock of one doesn't order against the lock of | |
570 | * another PTL. | |
571 | * | |
572 | * The decrement is ordered by the flush_tlb_range(), such that | |
573 | * mm_tlb_flush_pending() will not return false unless all flushes have | |
574 | * completed. | |
af2c1401 | 575 | */ |
20841405 | 576 | } |
16af97dc | 577 | |
16af97dc | 578 | static inline void dec_tlb_flush_pending(struct mm_struct *mm) |
20841405 | 579 | { |
0a2c4048 | 580 | /* |
0e709703 PZ |
581 | * See inc_tlb_flush_pending(). |
582 | * | |
583 | * This cannot be smp_mb__before_atomic() because smp_mb() simply does | |
584 | * not order against TLB invalidate completion, which is what we need. | |
585 | * | |
586 | * Therefore we must rely on tlb_flush_*() to guarantee order. | |
0a2c4048 | 587 | */ |
16af97dc | 588 | atomic_dec(&mm->tlb_flush_pending); |
20841405 | 589 | } |
20841405 | 590 | |
0e709703 PZ |
591 | static inline bool mm_tlb_flush_pending(struct mm_struct *mm) |
592 | { | |
593 | /* | |
594 | * Must be called after having acquired the PTL; orders against that | |
595 | * PTLs release and therefore ensures that if we observe the modified | |
596 | * PTE we must also observe the increment from inc_tlb_flush_pending(). | |
597 | * | |
598 | * That is, it only guarantees to return true if there is a flush | |
599 | * pending for _this_ PTL. | |
600 | */ | |
601 | return atomic_read(&mm->tlb_flush_pending); | |
602 | } | |
603 | ||
604 | static inline bool mm_tlb_flush_nested(struct mm_struct *mm) | |
605 | { | |
606 | /* | |
607 | * Similar to mm_tlb_flush_pending(), we must have acquired the PTL | |
608 | * for which there is a TLB flush pending in order to guarantee | |
609 | * we've seen both that PTE modification and the increment. | |
610 | * | |
611 | * (no requirement on actually still holding the PTL, that is irrelevant) | |
612 | */ | |
613 | return atomic_read(&mm->tlb_flush_pending) > 1; | |
614 | } | |
615 | ||
f872f540 AL |
616 | struct vm_fault; |
617 | ||
618 | struct vm_special_mapping { | |
619 | const char *name; /* The name, e.g. "[vdso]". */ | |
620 | ||
621 | /* | |
622 | * If .fault is not provided, this points to a | |
623 | * NULL-terminated array of pages that back the special mapping. | |
624 | * | |
625 | * This must not be NULL unless .fault is provided. | |
626 | */ | |
a62c34bd | 627 | struct page **pages; |
f872f540 AL |
628 | |
629 | /* | |
630 | * If non-NULL, then this is called to resolve page faults | |
631 | * on the special mapping. If used, .pages is not checked. | |
632 | */ | |
633 | int (*fault)(const struct vm_special_mapping *sm, | |
634 | struct vm_area_struct *vma, | |
635 | struct vm_fault *vmf); | |
b059a453 DS |
636 | |
637 | int (*mremap)(const struct vm_special_mapping *sm, | |
638 | struct vm_area_struct *new_vma); | |
a62c34bd AL |
639 | }; |
640 | ||
d17d8f9d DH |
641 | enum tlb_flush_reason { |
642 | TLB_FLUSH_ON_TASK_SWITCH, | |
643 | TLB_REMOTE_SHOOTDOWN, | |
644 | TLB_LOCAL_SHOOTDOWN, | |
645 | TLB_LOCAL_MM_SHOOTDOWN, | |
5b74283a | 646 | TLB_REMOTE_SEND_IPI, |
d17d8f9d DH |
647 | NR_TLB_FLUSH_REASONS, |
648 | }; | |
649 | ||
bd6dace7 TH |
650 | /* |
651 | * A swap entry has to fit into a "unsigned long", as the entry is hidden | |
652 | * in the "index" field of the swapper address space. | |
653 | */ | |
654 | typedef struct { | |
655 | unsigned long val; | |
656 | } swp_entry_t; | |
657 | ||
5b99cd0e | 658 | #endif /* _LINUX_MM_TYPES_H */ |