Line data Source code
1 : /* SPDX-License-Identifier: GPL-2.0 */
2 : #ifndef _LINUX_MM_TYPES_H
3 : #define _LINUX_MM_TYPES_H
4 :
5 : #include <linux/mm_types_task.h>
6 :
7 : #include <linux/auxvec.h>
8 : #include <linux/kref.h>
9 : #include <linux/list.h>
10 : #include <linux/spinlock.h>
11 : #include <linux/rbtree.h>
12 : #include <linux/rwsem.h>
13 : #include <linux/completion.h>
14 : #include <linux/cpumask.h>
15 : #include <linux/uprobes.h>
16 : #include <linux/rcupdate.h>
17 : #include <linux/page-flags-layout.h>
18 : #include <linux/workqueue.h>
19 : #include <linux/seqlock.h>
20 :
21 : #include <asm/mmu.h>
22 :
23 : #ifndef AT_VECTOR_SIZE_ARCH
24 : #define AT_VECTOR_SIZE_ARCH 0
25 : #endif
26 : #define AT_VECTOR_SIZE (2*(AT_VECTOR_SIZE_ARCH + AT_VECTOR_SIZE_BASE + 1))
27 :
28 : #define INIT_PASID 0
29 :
30 : struct address_space;
31 : struct mem_cgroup;
32 :
33 : /*
34 : * Each physical page in the system has a struct page associated with
35 : * it to keep track of whatever it is we are using the page for at the
36 : * moment. Note that we have no way to track which tasks are using
37 : * a page, though if it is a pagecache page, rmap structures can tell us
38 : * who is mapping it.
39 : *
40 : * If you allocate the page using alloc_pages(), you can use some of the
41 : * space in struct page for your own purposes. The five words in the main
42 : * union are available, except for bit 0 of the first word which must be
43 : * kept clear. Many users use this word to store a pointer to an object
44 : * which is guaranteed to be aligned. If you use the same storage as
45 : * page->mapping, you must restore it to NULL before freeing the page.
46 : *
47 : * If your page will not be mapped to userspace, you can also use the four
48 : * bytes in the mapcount union, but you must call page_mapcount_reset()
49 : * before freeing it.
50 : *
51 : * If you want to use the refcount field, it must be used in such a way
52 : * that other CPUs temporarily incrementing and then decrementing the
53 : * refcount does not cause problems. On receiving the page from
54 : * alloc_pages(), the refcount will be positive.
55 : *
56 : * If you allocate pages of order > 0, you can use some of the fields
57 : * in each subpage, but you may need to restore some of their values
58 : * afterwards.
59 : *
60 : * SLUB uses cmpxchg_double() to atomically update its freelist and counters.
61 : * That requires that freelist & counters in struct slab be adjacent and
62 : * double-word aligned. Because struct slab currently just reinterprets the
63 : * bits of struct page, we align all struct pages to double-word boundaries,
64 : * and ensure that 'freelist' is aligned within struct slab.
65 : */
66 : #ifdef CONFIG_HAVE_ALIGNED_STRUCT_PAGE
67 : #define _struct_page_alignment __aligned(2 * sizeof(unsigned long))
68 : #else
69 : #define _struct_page_alignment
70 : #endif
71 :
72 : struct page {
73 : unsigned long flags; /* Atomic flags, some possibly
74 : * updated asynchronously */
75 : /*
76 : * Five words (20/40 bytes) are available in this union.
77 : * WARNING: bit 0 of the first word is used for PageTail(). That
78 : * means the other users of this union MUST NOT use the bit to
79 : * avoid collision and false-positive PageTail().
80 : */
81 : union {
82 : struct { /* Page cache and anonymous pages */
83 : /**
84 : * @lru: Pageout list, eg. active_list protected by
85 : * lruvec->lru_lock. Sometimes used as a generic list
86 : * by the page owner.
87 : */
88 : union {
89 : struct list_head lru;
90 : /* Or, for the Unevictable "LRU list" slot */
91 : struct {
92 : /* Always even, to negate PageTail */
93 : void *__filler;
94 : /* Count page's or folio's mlocks */
95 : unsigned int mlock_count;
96 : };
97 : };
98 : /* See page-flags.h for PAGE_MAPPING_FLAGS */
99 : struct address_space *mapping;
100 : pgoff_t index; /* Our offset within mapping. */
101 : /**
102 : * @private: Mapping-private opaque data.
103 : * Usually used for buffer_heads if PagePrivate.
104 : * Used for swp_entry_t if PageSwapCache.
105 : * Indicates order in the buddy system if PageBuddy.
106 : */
107 : unsigned long private;
108 : };
109 : struct { /* page_pool used by netstack */
110 : /**
111 : * @pp_magic: magic value to avoid recycling non
112 : * page_pool allocated pages.
113 : */
114 : unsigned long pp_magic;
115 : struct page_pool *pp;
116 : unsigned long _pp_mapping_pad;
117 : unsigned long dma_addr;
118 : union {
119 : /**
120 : * dma_addr_upper: might require a 64-bit
121 : * value on 32-bit architectures.
122 : */
123 : unsigned long dma_addr_upper;
124 : /**
125 : * For frag page support, not supported in
126 : * 32-bit architectures with 64-bit DMA.
127 : */
128 : atomic_long_t pp_frag_count;
129 : };
130 : };
131 : struct { /* Tail pages of compound page */
132 : unsigned long compound_head; /* Bit zero is set */
133 :
134 : /* First tail page only */
135 : unsigned char compound_dtor;
136 : unsigned char compound_order;
137 : atomic_t compound_mapcount;
138 : atomic_t compound_pincount;
139 : #ifdef CONFIG_64BIT
140 : unsigned int compound_nr; /* 1 << compound_order */
141 : #endif
142 : };
143 : struct { /* Second tail page of compound page */
144 : unsigned long _compound_pad_1; /* compound_head */
145 : unsigned long _compound_pad_2;
146 : /* For both global and memcg */
147 : struct list_head deferred_list;
148 : };
149 : struct { /* Page table pages */
150 : unsigned long _pt_pad_1; /* compound_head */
151 : pgtable_t pmd_huge_pte; /* protected by page->ptl */
152 : unsigned long _pt_pad_2; /* mapping */
153 : union {
154 : struct mm_struct *pt_mm; /* x86 pgds only */
155 : atomic_t pt_frag_refcount; /* powerpc */
156 : };
157 : #if ALLOC_SPLIT_PTLOCKS
158 : spinlock_t *ptl;
159 : #else
160 : spinlock_t ptl;
161 : #endif
162 : };
163 : struct { /* ZONE_DEVICE pages */
164 : /** @pgmap: Points to the hosting device page map. */
165 : struct dev_pagemap *pgmap;
166 : void *zone_device_data;
167 : /*
168 : * ZONE_DEVICE private pages are counted as being
169 : * mapped so the next 3 words hold the mapping, index,
170 : * and private fields from the source anonymous or
171 : * page cache page while the page is migrated to device
172 : * private memory.
173 : * ZONE_DEVICE MEMORY_DEVICE_FS_DAX pages also
174 : * use the mapping, index, and private fields when
175 : * pmem backed DAX files are mapped.
176 : */
177 : };
178 :
179 : /** @rcu_head: You can use this to free a page by RCU. */
180 : struct rcu_head rcu_head;
181 : };
182 :
183 : union { /* This union is 4 bytes in size. */
184 : /*
185 : * If the page can be mapped to userspace, encodes the number
186 : * of times this page is referenced by a page table.
187 : */
188 : atomic_t _mapcount;
189 :
190 : /*
191 : * If the page is neither PageSlab nor mappable to userspace,
192 : * the value stored here may help determine what this page
193 : * is used for. See page-flags.h for a list of page types
194 : * which are currently stored here.
195 : */
196 : unsigned int page_type;
197 : };
198 :
199 : /* Usage count. *DO NOT USE DIRECTLY*. See page_ref.h */
200 : atomic_t _refcount;
201 :
202 : #ifdef CONFIG_MEMCG
203 : unsigned long memcg_data;
204 : #endif
205 :
206 : /*
207 : * On machines where all RAM is mapped into kernel address space,
208 : * we can simply calculate the virtual address. On machines with
209 : * highmem some memory is mapped into kernel virtual memory
210 : * dynamically, so we need a place to store that address.
211 : * Note that this field could be 16 bits on x86 ... ;)
212 : *
213 : * Architectures with slow multiplication can define
214 : * WANT_PAGE_VIRTUAL in asm/page.h
215 : */
216 : #if defined(WANT_PAGE_VIRTUAL)
217 : void *virtual; /* Kernel virtual address (NULL if
218 : not kmapped, ie. highmem) */
219 : #endif /* WANT_PAGE_VIRTUAL */
220 :
221 : #ifdef LAST_CPUPID_NOT_IN_PAGE_FLAGS
222 : int _last_cpupid;
223 : #endif
224 : } _struct_page_alignment;
225 :
226 : /**
227 : * struct folio - Represents a contiguous set of bytes.
228 : * @flags: Identical to the page flags.
229 : * @lru: Least Recently Used list; tracks how recently this folio was used.
230 : * @mapping: The file this page belongs to, or refers to the anon_vma for
231 : * anonymous memory.
232 : * @index: Offset within the file, in units of pages. For anonymous memory,
233 : * this is the index from the beginning of the mmap.
234 : * @private: Filesystem per-folio data (see folio_attach_private()).
235 : * Used for swp_entry_t if folio_test_swapcache().
236 : * @_mapcount: Do not access this member directly. Use folio_mapcount() to
237 : * find out how many times this folio is mapped by userspace.
238 : * @_refcount: Do not access this member directly. Use folio_ref_count()
239 : * to find how many references there are to this folio.
240 : * @memcg_data: Memory Control Group data.
241 : *
242 : * A folio is a physically, virtually and logically contiguous set
243 : * of bytes. It is a power-of-two in size, and it is aligned to that
244 : * same power-of-two. It is at least as large as %PAGE_SIZE. If it is
245 : * in the page cache, it is at a file offset which is a multiple of that
246 : * power-of-two. It may be mapped into userspace at an address which is
247 : * at an arbitrary page offset, but its kernel virtual address is aligned
248 : * to its size.
249 : */
250 : struct folio {
251 : /* private: don't document the anon union */
252 : union {
253 : struct {
254 : /* public: */
255 : unsigned long flags;
256 : union {
257 : struct list_head lru;
258 : struct {
259 : void *__filler;
260 : unsigned int mlock_count;
261 : };
262 : };
263 : struct address_space *mapping;
264 : pgoff_t index;
265 : void *private;
266 : atomic_t _mapcount;
267 : atomic_t _refcount;
268 : #ifdef CONFIG_MEMCG
269 : unsigned long memcg_data;
270 : #endif
271 : /* private: the union with struct page is transitional */
272 : };
273 : struct page page;
274 : };
275 : };
276 :
277 : static_assert(sizeof(struct page) == sizeof(struct folio));
278 : #define FOLIO_MATCH(pg, fl) \
279 : static_assert(offsetof(struct page, pg) == offsetof(struct folio, fl))
280 : FOLIO_MATCH(flags, flags);
281 : FOLIO_MATCH(lru, lru);
282 : FOLIO_MATCH(mapping, mapping);
283 : FOLIO_MATCH(compound_head, lru);
284 : FOLIO_MATCH(index, index);
285 : FOLIO_MATCH(private, private);
286 : FOLIO_MATCH(_mapcount, _mapcount);
287 : FOLIO_MATCH(_refcount, _refcount);
288 : #ifdef CONFIG_MEMCG
289 : FOLIO_MATCH(memcg_data, memcg_data);
290 : #endif
291 : #undef FOLIO_MATCH
292 :
293 : static inline atomic_t *folio_mapcount_ptr(struct folio *folio)
294 : {
295 0 : struct page *tail = &folio->page + 1;
296 : return &tail->compound_mapcount;
297 : }
298 :
299 : static inline atomic_t *compound_mapcount_ptr(struct page *page)
300 : {
301 0 : return &page[1].compound_mapcount;
302 : }
303 :
304 : static inline atomic_t *compound_pincount_ptr(struct page *page)
305 : {
306 : return &page[1].compound_pincount;
307 : }
308 :
309 : /*
310 : * Used for sizing the vmemmap region on some architectures
311 : */
312 : #define STRUCT_PAGE_MAX_SHIFT (order_base_2(sizeof(struct page)))
313 :
314 : #define PAGE_FRAG_CACHE_MAX_SIZE __ALIGN_MASK(32768, ~PAGE_MASK)
315 : #define PAGE_FRAG_CACHE_MAX_ORDER get_order(PAGE_FRAG_CACHE_MAX_SIZE)
316 :
317 : /*
318 : * page_private can be used on tail pages. However, PagePrivate is only
319 : * checked by the VM on the head page. So page_private on the tail pages
320 : * should be used for data that's ancillary to the head page (eg attaching
321 : * buffer heads to tail pages after attaching buffer heads to the head page)
322 : */
323 : #define page_private(page) ((page)->private)
324 :
325 : static inline void set_page_private(struct page *page, unsigned long private)
326 : {
327 2189 : page->private = private;
328 : }
329 :
330 : static inline void *folio_get_private(struct folio *folio)
331 : {
332 : return folio->private;
333 : }
334 :
335 : struct page_frag_cache {
336 : void * va;
337 : #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
338 : __u16 offset;
339 : __u16 size;
340 : #else
341 : __u32 offset;
342 : #endif
343 : /* we maintain a pagecount bias, so that we dont dirty cache line
344 : * containing page->_refcount every time we allocate a fragment.
345 : */
346 : unsigned int pagecnt_bias;
347 : bool pfmemalloc;
348 : };
349 :
350 : typedef unsigned long vm_flags_t;
351 :
352 : /*
353 : * A region containing a mapping of a non-memory backed file under NOMMU
354 : * conditions. These are held in a global tree and are pinned by the VMAs that
355 : * map parts of them.
356 : */
357 : struct vm_region {
358 : struct rb_node vm_rb; /* link in global region tree */
359 : vm_flags_t vm_flags; /* VMA vm_flags */
360 : unsigned long vm_start; /* start address of region */
361 : unsigned long vm_end; /* region initialised to here */
362 : unsigned long vm_top; /* region allocated to here */
363 : unsigned long vm_pgoff; /* the offset in vm_file corresponding to vm_start */
364 : struct file *vm_file; /* the backing file or NULL */
365 :
366 : int vm_usage; /* region usage count (access under nommu_region_sem) */
367 : bool vm_icache_flushed : 1; /* true if the icache has been flushed for
368 : * this region */
369 : };
370 :
371 : #ifdef CONFIG_USERFAULTFD
372 : #define NULL_VM_UFFD_CTX ((struct vm_userfaultfd_ctx) { NULL, })
373 : struct vm_userfaultfd_ctx {
374 : struct userfaultfd_ctx *ctx;
375 : };
376 : #else /* CONFIG_USERFAULTFD */
377 : #define NULL_VM_UFFD_CTX ((struct vm_userfaultfd_ctx) {})
378 : struct vm_userfaultfd_ctx {};
379 : #endif /* CONFIG_USERFAULTFD */
380 :
381 : struct anon_vma_name {
382 : struct kref kref;
383 : /* The name needs to be at the end because it is dynamically sized. */
384 : char name[];
385 : };
386 :
387 : /*
388 : * This struct describes a virtual memory area. There is one of these
389 : * per VM-area/task. A VM area is any part of the process virtual memory
390 : * space that has a special rule for the page-fault handlers (ie a shared
391 : * library, the executable area etc).
392 : */
393 : struct vm_area_struct {
394 : /* The first cache line has the info for VMA tree walking. */
395 :
396 : unsigned long vm_start; /* Our start address within vm_mm. */
397 : unsigned long vm_end; /* The first byte after our end address
398 : within vm_mm. */
399 :
400 : /* linked list of VM areas per task, sorted by address */
401 : struct vm_area_struct *vm_next, *vm_prev;
402 :
403 : struct rb_node vm_rb;
404 :
405 : /*
406 : * Largest free memory gap in bytes to the left of this VMA.
407 : * Either between this VMA and vma->vm_prev, or between one of the
408 : * VMAs below us in the VMA rbtree and its ->vm_prev. This helps
409 : * get_unmapped_area find a free area of the right size.
410 : */
411 : unsigned long rb_subtree_gap;
412 :
413 : /* Second cache line starts here. */
414 :
415 : struct mm_struct *vm_mm; /* The address space we belong to. */
416 :
417 : /*
418 : * Access permissions of this VMA.
419 : * See vmf_insert_mixed_prot() for discussion.
420 : */
421 : pgprot_t vm_page_prot;
422 : unsigned long vm_flags; /* Flags, see mm.h. */
423 :
424 : /*
425 : * For areas with an address space and backing store,
426 : * linkage into the address_space->i_mmap interval tree.
427 : *
428 : * For private anonymous mappings, a pointer to a null terminated string
429 : * containing the name given to the vma, or NULL if unnamed.
430 : */
431 :
432 : union {
433 : struct {
434 : struct rb_node rb;
435 : unsigned long rb_subtree_last;
436 : } shared;
437 : /*
438 : * Serialized by mmap_sem. Never use directly because it is
439 : * valid only when vm_file is NULL. Use anon_vma_name instead.
440 : */
441 : struct anon_vma_name *anon_name;
442 : };
443 :
444 : /*
445 : * A file's MAP_PRIVATE vma can be in both i_mmap tree and anon_vma
446 : * list, after a COW of one of the file pages. A MAP_SHARED vma
447 : * can only be in the i_mmap tree. An anonymous MAP_PRIVATE, stack
448 : * or brk vma (with NULL file) can only be in an anon_vma list.
449 : */
450 : struct list_head anon_vma_chain; /* Serialized by mmap_lock &
451 : * page_table_lock */
452 : struct anon_vma *anon_vma; /* Serialized by page_table_lock */
453 :
454 : /* Function pointers to deal with this struct. */
455 : const struct vm_operations_struct *vm_ops;
456 :
457 : /* Information about our backing store: */
458 : unsigned long vm_pgoff; /* Offset (within vm_file) in PAGE_SIZE
459 : units */
460 : struct file * vm_file; /* File we map to (can be NULL). */
461 : void * vm_private_data; /* was vm_pte (shared mem) */
462 :
463 : #ifdef CONFIG_SWAP
464 : atomic_long_t swap_readahead_info;
465 : #endif
466 : #ifndef CONFIG_MMU
467 : struct vm_region *vm_region; /* NOMMU mapping region */
468 : #endif
469 : #ifdef CONFIG_NUMA
470 : struct mempolicy *vm_policy; /* NUMA policy for the VMA */
471 : #endif
472 : struct vm_userfaultfd_ctx vm_userfaultfd_ctx;
473 : } __randomize_layout;
474 :
475 : struct kioctx_table;
476 : struct mm_struct {
477 : struct {
478 : struct vm_area_struct *mmap; /* list of VMAs */
479 : struct rb_root mm_rb;
480 : u64 vmacache_seqnum; /* per-thread vmacache */
481 : #ifdef CONFIG_MMU
482 : unsigned long (*get_unmapped_area) (struct file *filp,
483 : unsigned long addr, unsigned long len,
484 : unsigned long pgoff, unsigned long flags);
485 : #endif
486 : unsigned long mmap_base; /* base of mmap area */
487 : unsigned long mmap_legacy_base; /* base of mmap area in bottom-up allocations */
488 : #ifdef CONFIG_HAVE_ARCH_COMPAT_MMAP_BASES
489 : /* Base addresses for compatible mmap() */
490 : unsigned long mmap_compat_base;
491 : unsigned long mmap_compat_legacy_base;
492 : #endif
493 : unsigned long task_size; /* size of task vm space */
494 : unsigned long highest_vm_end; /* highest vma end address */
495 : pgd_t * pgd;
496 :
497 : #ifdef CONFIG_MEMBARRIER
498 : /**
499 : * @membarrier_state: Flags controlling membarrier behavior.
500 : *
501 : * This field is close to @pgd to hopefully fit in the same
502 : * cache-line, which needs to be touched by switch_mm().
503 : */
504 : atomic_t membarrier_state;
505 : #endif
506 :
507 : /**
508 : * @mm_users: The number of users including userspace.
509 : *
510 : * Use mmget()/mmget_not_zero()/mmput() to modify. When this
511 : * drops to 0 (i.e. when the task exits and there are no other
512 : * temporary reference holders), we also release a reference on
513 : * @mm_count (which may then free the &struct mm_struct if
514 : * @mm_count also drops to 0).
515 : */
516 : atomic_t mm_users;
517 :
518 : /**
519 : * @mm_count: The number of references to &struct mm_struct
520 : * (@mm_users count as 1).
521 : *
522 : * Use mmgrab()/mmdrop() to modify. When this drops to 0, the
523 : * &struct mm_struct is freed.
524 : */
525 : atomic_t mm_count;
526 :
527 : #ifdef CONFIG_MMU
528 : atomic_long_t pgtables_bytes; /* PTE page table pages */
529 : #endif
530 : int map_count; /* number of VMAs */
531 :
532 : spinlock_t page_table_lock; /* Protects page tables and some
533 : * counters
534 : */
535 : /*
536 : * With some kernel config, the current mmap_lock's offset
537 : * inside 'mm_struct' is at 0x120, which is very optimal, as
538 : * its two hot fields 'count' and 'owner' sit in 2 different
539 : * cachelines, and when mmap_lock is highly contended, both
540 : * of the 2 fields will be accessed frequently, current layout
541 : * will help to reduce cache bouncing.
542 : *
543 : * So please be careful with adding new fields before
544 : * mmap_lock, which can easily push the 2 fields into one
545 : * cacheline.
546 : */
547 : struct rw_semaphore mmap_lock;
548 :
549 : struct list_head mmlist; /* List of maybe swapped mm's. These
550 : * are globally strung together off
551 : * init_mm.mmlist, and are protected
552 : * by mmlist_lock
553 : */
554 :
555 :
556 : unsigned long hiwater_rss; /* High-watermark of RSS usage */
557 : unsigned long hiwater_vm; /* High-water virtual memory usage */
558 :
559 : unsigned long total_vm; /* Total pages mapped */
560 : unsigned long locked_vm; /* Pages that have PG_mlocked set */
561 : atomic64_t pinned_vm; /* Refcount permanently increased */
562 : unsigned long data_vm; /* VM_WRITE & ~VM_SHARED & ~VM_STACK */
563 : unsigned long exec_vm; /* VM_EXEC & ~VM_WRITE & ~VM_STACK */
564 : unsigned long stack_vm; /* VM_STACK */
565 : unsigned long def_flags;
566 :
567 : /**
568 : * @write_protect_seq: Locked when any thread is write
569 : * protecting pages mapped by this mm to enforce a later COW,
570 : * for instance during page table copying for fork().
571 : */
572 : seqcount_t write_protect_seq;
573 :
574 : spinlock_t arg_lock; /* protect the below fields */
575 :
576 : unsigned long start_code, end_code, start_data, end_data;
577 : unsigned long start_brk, brk, start_stack;
578 : unsigned long arg_start, arg_end, env_start, env_end;
579 :
580 : unsigned long saved_auxv[AT_VECTOR_SIZE]; /* for /proc/PID/auxv */
581 :
582 : /*
583 : * Special counters, in some configurations protected by the
584 : * page_table_lock, in other configurations by being atomic.
585 : */
586 : struct mm_rss_stat rss_stat;
587 :
588 : struct linux_binfmt *binfmt;
589 :
590 : /* Architecture-specific MM context */
591 : mm_context_t context;
592 :
593 : unsigned long flags; /* Must use atomic bitops to access */
594 :
595 : #ifdef CONFIG_AIO
596 : spinlock_t ioctx_lock;
597 : struct kioctx_table __rcu *ioctx_table;
598 : #endif
599 : #ifdef CONFIG_MEMCG
600 : /*
601 : * "owner" points to a task that is regarded as the canonical
602 : * user/owner of this mm. All of the following must be true in
603 : * order for it to be changed:
604 : *
605 : * current == mm->owner
606 : * current->mm != mm
607 : * new_owner->mm == mm
608 : * new_owner->alloc_lock is held
609 : */
610 : struct task_struct __rcu *owner;
611 : #endif
612 : struct user_namespace *user_ns;
613 :
614 : /* store ref to file /proc/<pid>/exe symlink points to */
615 : struct file __rcu *exe_file;
616 : #ifdef CONFIG_MMU_NOTIFIER
617 : struct mmu_notifier_subscriptions *notifier_subscriptions;
618 : #endif
619 : #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
620 : pgtable_t pmd_huge_pte; /* protected by page_table_lock */
621 : #endif
622 : #ifdef CONFIG_NUMA_BALANCING
623 : /*
624 : * numa_next_scan is the next time that the PTEs will be marked
625 : * pte_numa. NUMA hinting faults will gather statistics and
626 : * migrate pages to new nodes if necessary.
627 : */
628 : unsigned long numa_next_scan;
629 :
630 : /* Restart point for scanning and setting pte_numa */
631 : unsigned long numa_scan_offset;
632 :
633 : /* numa_scan_seq prevents two threads setting pte_numa */
634 : int numa_scan_seq;
635 : #endif
636 : /*
637 : * An operation with batched TLB flushing is going on. Anything
638 : * that can move process memory needs to flush the TLB when
639 : * moving a PROT_NONE or PROT_NUMA mapped page.
640 : */
641 : atomic_t tlb_flush_pending;
642 : #ifdef CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH
643 : /* See flush_tlb_batched_pending() */
644 : atomic_t tlb_flush_batched;
645 : #endif
646 : struct uprobes_state uprobes_state;
647 : #ifdef CONFIG_PREEMPT_RT
648 : struct rcu_head delayed_drop;
649 : #endif
650 : #ifdef CONFIG_HUGETLB_PAGE
651 : atomic_long_t hugetlb_usage;
652 : #endif
653 : struct work_struct async_put_work;
654 :
655 : #ifdef CONFIG_IOMMU_SVA
656 : u32 pasid;
657 : #endif
658 : } __randomize_layout;
659 :
660 : /*
661 : * The mm_cpumask needs to be at the end of mm_struct, because it
662 : * is dynamically sized based on nr_cpu_ids.
663 : */
664 : unsigned long cpu_bitmap[];
665 : };
666 :
667 : extern struct mm_struct init_mm;
668 :
669 : /* Pointer magic because the dynamic array size confuses some compilers. */
670 : static inline void mm_init_cpumask(struct mm_struct *mm)
671 : {
672 0 : unsigned long cpu_bitmap = (unsigned long)mm;
673 :
674 0 : cpu_bitmap += offsetof(struct mm_struct, cpu_bitmap);
675 0 : cpumask_clear((struct cpumask *)cpu_bitmap);
676 : }
677 :
678 : /* Future-safe accessor for struct mm_struct's cpu_vm_mask. */
679 : static inline cpumask_t *mm_cpumask(struct mm_struct *mm)
680 : {
681 : return (struct cpumask *)&mm->cpu_bitmap;
682 : }
683 :
684 : struct mmu_gather;
685 : extern void tlb_gather_mmu(struct mmu_gather *tlb, struct mm_struct *mm);
686 : extern void tlb_gather_mmu_fullmm(struct mmu_gather *tlb, struct mm_struct *mm);
687 : extern void tlb_finish_mmu(struct mmu_gather *tlb);
688 :
689 : struct vm_fault;
690 :
691 : /**
692 : * typedef vm_fault_t - Return type for page fault handlers.
693 : *
694 : * Page fault handlers return a bitmask of %VM_FAULT values.
695 : */
696 : typedef __bitwise unsigned int vm_fault_t;
697 :
698 : /**
699 : * enum vm_fault_reason - Page fault handlers return a bitmask of
700 : * these values to tell the core VM what happened when handling the
701 : * fault. Used to decide whether a process gets delivered SIGBUS or
702 : * just gets major/minor fault counters bumped up.
703 : *
704 : * @VM_FAULT_OOM: Out Of Memory
705 : * @VM_FAULT_SIGBUS: Bad access
706 : * @VM_FAULT_MAJOR: Page read from storage
707 : * @VM_FAULT_WRITE: Special case for get_user_pages
708 : * @VM_FAULT_HWPOISON: Hit poisoned small page
709 : * @VM_FAULT_HWPOISON_LARGE: Hit poisoned large page. Index encoded
710 : * in upper bits
711 : * @VM_FAULT_SIGSEGV: segmentation fault
712 : * @VM_FAULT_NOPAGE: ->fault installed the pte, not return page
713 : * @VM_FAULT_LOCKED: ->fault locked the returned page
714 : * @VM_FAULT_RETRY: ->fault blocked, must retry
715 : * @VM_FAULT_FALLBACK: huge page fault failed, fall back to small
716 : * @VM_FAULT_DONE_COW: ->fault has fully handled COW
717 : * @VM_FAULT_NEEDDSYNC: ->fault did not modify page tables and needs
718 : * fsync() to complete (for synchronous page faults
719 : * in DAX)
720 : * @VM_FAULT_HINDEX_MASK: mask HINDEX value
721 : *
722 : */
723 : enum vm_fault_reason {
724 : VM_FAULT_OOM = (__force vm_fault_t)0x000001,
725 : VM_FAULT_SIGBUS = (__force vm_fault_t)0x000002,
726 : VM_FAULT_MAJOR = (__force vm_fault_t)0x000004,
727 : VM_FAULT_WRITE = (__force vm_fault_t)0x000008,
728 : VM_FAULT_HWPOISON = (__force vm_fault_t)0x000010,
729 : VM_FAULT_HWPOISON_LARGE = (__force vm_fault_t)0x000020,
730 : VM_FAULT_SIGSEGV = (__force vm_fault_t)0x000040,
731 : VM_FAULT_NOPAGE = (__force vm_fault_t)0x000100,
732 : VM_FAULT_LOCKED = (__force vm_fault_t)0x000200,
733 : VM_FAULT_RETRY = (__force vm_fault_t)0x000400,
734 : VM_FAULT_FALLBACK = (__force vm_fault_t)0x000800,
735 : VM_FAULT_DONE_COW = (__force vm_fault_t)0x001000,
736 : VM_FAULT_NEEDDSYNC = (__force vm_fault_t)0x002000,
737 : VM_FAULT_HINDEX_MASK = (__force vm_fault_t)0x0f0000,
738 : };
739 :
740 : /* Encode hstate index for a hwpoisoned large page */
741 : #define VM_FAULT_SET_HINDEX(x) ((__force vm_fault_t)((x) << 16))
742 : #define VM_FAULT_GET_HINDEX(x) (((__force unsigned int)(x) >> 16) & 0xf)
743 :
744 : #define VM_FAULT_ERROR (VM_FAULT_OOM | VM_FAULT_SIGBUS | \
745 : VM_FAULT_SIGSEGV | VM_FAULT_HWPOISON | \
746 : VM_FAULT_HWPOISON_LARGE | VM_FAULT_FALLBACK)
747 :
748 : #define VM_FAULT_RESULT_TRACE \
749 : { VM_FAULT_OOM, "OOM" }, \
750 : { VM_FAULT_SIGBUS, "SIGBUS" }, \
751 : { VM_FAULT_MAJOR, "MAJOR" }, \
752 : { VM_FAULT_WRITE, "WRITE" }, \
753 : { VM_FAULT_HWPOISON, "HWPOISON" }, \
754 : { VM_FAULT_HWPOISON_LARGE, "HWPOISON_LARGE" }, \
755 : { VM_FAULT_SIGSEGV, "SIGSEGV" }, \
756 : { VM_FAULT_NOPAGE, "NOPAGE" }, \
757 : { VM_FAULT_LOCKED, "LOCKED" }, \
758 : { VM_FAULT_RETRY, "RETRY" }, \
759 : { VM_FAULT_FALLBACK, "FALLBACK" }, \
760 : { VM_FAULT_DONE_COW, "DONE_COW" }, \
761 : { VM_FAULT_NEEDDSYNC, "NEEDDSYNC" }
762 :
763 : struct vm_special_mapping {
764 : const char *name; /* The name, e.g. "[vdso]". */
765 :
766 : /*
767 : * If .fault is not provided, this points to a
768 : * NULL-terminated array of pages that back the special mapping.
769 : *
770 : * This must not be NULL unless .fault is provided.
771 : */
772 : struct page **pages;
773 :
774 : /*
775 : * If non-NULL, then this is called to resolve page faults
776 : * on the special mapping. If used, .pages is not checked.
777 : */
778 : vm_fault_t (*fault)(const struct vm_special_mapping *sm,
779 : struct vm_area_struct *vma,
780 : struct vm_fault *vmf);
781 :
782 : int (*mremap)(const struct vm_special_mapping *sm,
783 : struct vm_area_struct *new_vma);
784 : };
785 :
786 : enum tlb_flush_reason {
787 : TLB_FLUSH_ON_TASK_SWITCH,
788 : TLB_REMOTE_SHOOTDOWN,
789 : TLB_LOCAL_SHOOTDOWN,
790 : TLB_LOCAL_MM_SHOOTDOWN,
791 : TLB_REMOTE_SEND_IPI,
792 : NR_TLB_FLUSH_REASONS,
793 : };
794 :
795 : /*
796 : * A swap entry has to fit into a "unsigned long", as the entry is hidden
797 : * in the "index" field of the swapper address space.
798 : */
799 : typedef struct {
800 : unsigned long val;
801 : } swp_entry_t;
802 :
803 : /**
804 : * enum fault_flag - Fault flag definitions.
805 : * @FAULT_FLAG_WRITE: Fault was a write fault.
806 : * @FAULT_FLAG_MKWRITE: Fault was mkwrite of existing PTE.
807 : * @FAULT_FLAG_ALLOW_RETRY: Allow to retry the fault if blocked.
808 : * @FAULT_FLAG_RETRY_NOWAIT: Don't drop mmap_lock and wait when retrying.
809 : * @FAULT_FLAG_KILLABLE: The fault task is in SIGKILL killable region.
810 : * @FAULT_FLAG_TRIED: The fault has been tried once.
811 : * @FAULT_FLAG_USER: The fault originated in userspace.
812 : * @FAULT_FLAG_REMOTE: The fault is not for current task/mm.
813 : * @FAULT_FLAG_INSTRUCTION: The fault was during an instruction fetch.
814 : * @FAULT_FLAG_INTERRUPTIBLE: The fault can be interrupted by non-fatal signals.
815 : *
816 : * About @FAULT_FLAG_ALLOW_RETRY and @FAULT_FLAG_TRIED: we can specify
817 : * whether we would allow page faults to retry by specifying these two
818 : * fault flags correctly. Currently there can be three legal combinations:
819 : *
820 : * (a) ALLOW_RETRY and !TRIED: this means the page fault allows retry, and
821 : * this is the first try
822 : *
823 : * (b) ALLOW_RETRY and TRIED: this means the page fault allows retry, and
824 : * we've already tried at least once
825 : *
826 : * (c) !ALLOW_RETRY and !TRIED: this means the page fault does not allow retry
827 : *
828 : * The unlisted combination (!ALLOW_RETRY && TRIED) is illegal and should never
829 : * be used. Note that page faults can be allowed to retry for multiple times,
830 : * in which case we'll have an initial fault with flags (a) then later on
831 : * continuous faults with flags (b). We should always try to detect pending
832 : * signals before a retry to make sure the continuous page faults can still be
833 : * interrupted if necessary.
834 : */
835 : enum fault_flag {
836 : FAULT_FLAG_WRITE = 1 << 0,
837 : FAULT_FLAG_MKWRITE = 1 << 1,
838 : FAULT_FLAG_ALLOW_RETRY = 1 << 2,
839 : FAULT_FLAG_RETRY_NOWAIT = 1 << 3,
840 : FAULT_FLAG_KILLABLE = 1 << 4,
841 : FAULT_FLAG_TRIED = 1 << 5,
842 : FAULT_FLAG_USER = 1 << 6,
843 : FAULT_FLAG_REMOTE = 1 << 7,
844 : FAULT_FLAG_INSTRUCTION = 1 << 8,
845 : FAULT_FLAG_INTERRUPTIBLE = 1 << 9,
846 : };
847 :
848 : #endif /* _LINUX_MM_TYPES_H */
|
