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 #include <linux/sched.h>
7
8 #include <linux/auxvec.h>
9 #include <linux/kref.h>
10 #include <linux/list.h>
11 #include <linux/spinlock.h>
12 #include <linux/rbtree.h>
13 #include <linux/rwsem.h>
14 #include <linux/completion.h>
15 #include <linux/cpumask.h>
16 #include <linux/uprobes.h>
17 #include <linux/page-flags-layout.h>
18 #include <linux/workqueue.h>
19 #include <linux/seqlock.h>
20 #include <linux/nodemask.h>
21 #include <linux/mmdebug.h>
22 #include <linux/android_kabi.h>
23
24 #include <asm/mmu.h>
25
26 #ifndef AT_VECTOR_SIZE_ARCH
27 #define AT_VECTOR_SIZE_ARCH 0
28 #endif
29 #define AT_VECTOR_SIZE (2*(AT_VECTOR_SIZE_ARCH + AT_VECTOR_SIZE_BASE + 1))
30
31 #define INIT_PASID 0
32
33 struct address_space;
34 struct mem_cgroup;
35
36 /*
37 * Each physical page in the system has a struct page associated with
38 * it to keep track of whatever it is we are using the page for at the
39 * moment. Note that we have no way to track which tasks are using
40 * a page, though if it is a pagecache page, rmap structures can tell us
41 * who is mapping it.
42 *
43 * If you allocate the page using alloc_pages(), you can use some of the
44 * space in struct page for your own purposes. The five words in the main
45 * union are available, except for bit 0 of the first word which must be
46 * kept clear. Many users use this word to store a pointer to an object
47 * which is guaranteed to be aligned. If you use the same storage as
48 * page->mapping, you must restore it to NULL before freeing the page.
49 *
50 * If your page will not be mapped to userspace, you can also use the four
51 * bytes in the mapcount union, but you must call page_mapcount_reset()
52 * before freeing it.
53 *
54 * If you want to use the refcount field, it must be used in such a way
55 * that other CPUs temporarily incrementing and then decrementing the
56 * refcount does not cause problems. On receiving the page from
57 * alloc_pages(), the refcount will be positive.
58 *
59 * If you allocate pages of order > 0, you can use some of the fields
60 * in each subpage, but you may need to restore some of their values
61 * afterwards.
62 *
63 * SLUB uses cmpxchg_double() to atomically update its freelist and
64 * counters. That requires that freelist & counters be adjacent and
65 * double-word aligned. We align all struct pages to double-word
66 * boundaries, and ensure that 'freelist' is aligned within the
67 * struct.
68 */
69 #ifdef CONFIG_HAVE_ALIGNED_STRUCT_PAGE
70 #define _struct_page_alignment __aligned(2 * sizeof(unsigned long))
71 #else
72 #define _struct_page_alignment
73 #endif
74
75 struct page {
76 unsigned long flags; /* Atomic flags, some possibly
77 * updated asynchronously */
78 /*
79 * Five words (20/40 bytes) are available in this union.
80 * WARNING: bit 0 of the first word is used for PageTail(). That
81 * means the other users of this union MUST NOT use the bit to
82 * avoid collision and false-positive PageTail().
83 */
84 union {
85 struct { /* Page cache and anonymous pages */
86 /**
87 * @lru: Pageout list, eg. active_list protected by
88 * lruvec->lru_lock. Sometimes used as a generic list
89 * by the page owner.
90 */
91 struct list_head lru;
92
93 /* See page-flags.h for PAGE_MAPPING_FLAGS */
94 struct address_space *mapping;
95 pgoff_t index; /* Our offset within mapping. */
96 /**
97 * @private: Mapping-private opaque data.
98 * Usually used for buffer_heads if PagePrivate.
99 * Used for swp_entry_t if PageSwapCache.
100 * Indicates order in the buddy system if PageBuddy.
101 */
102 unsigned long private;
103 };
104 struct { /* page_pool used by netstack */
105 /**
106 * @pp_magic: magic value to avoid recycling non
107 * page_pool allocated pages.
108 */
109 unsigned long pp_magic;
110 struct page_pool *pp;
111 unsigned long _pp_mapping_pad;
112 unsigned long dma_addr;
113 union {
114 /**
115 * dma_addr_upper: might require a 64-bit
116 * value on 32-bit architectures.
117 */
118 unsigned long dma_addr_upper;
119 /**
120 * For frag page support, not supported in
121 * 32-bit architectures with 64-bit DMA.
122 */
123 atomic_long_t pp_frag_count;
124 };
125 };
126 struct { /* slab, slob and slub */
127 union {
128 struct list_head slab_list;
129 struct { /* Partial pages */
130 struct page *next;
131 #ifdef CONFIG_64BIT
132 int pages; /* Nr of pages left */
133 int pobjects; /* Approximate count */
134 #else
135 short int pages;
136 short int pobjects;
137 #endif
138 };
139 };
140 struct kmem_cache *slab_cache; /* not slob */
141 /* Double-word boundary */
142 void *freelist; /* first free object */
143 union {
144 void *s_mem; /* slab: first object */
145 unsigned long counters; /* SLUB */
146 struct { /* SLUB */
147 unsigned inuse:16;
148 unsigned objects:15;
149 unsigned frozen:1;
150 };
151 };
152 };
153 struct { /* Tail pages of compound page */
154 unsigned long compound_head; /* Bit zero is set */
155
156 /* First tail page only */
157 unsigned char compound_dtor;
158 unsigned char compound_order;
159 atomic_t compound_mapcount;
160 unsigned int compound_nr; /* 1 << compound_order */
161 };
162 struct { /* Second tail page of compound page */
163 unsigned long _compound_pad_1; /* compound_head */
164 atomic_t hpage_pinned_refcount;
165 /* For both global and memcg */
166 struct list_head deferred_list;
167 };
168 struct { /* Page table pages */
169 unsigned long _pt_pad_1; /* compound_head */
170 pgtable_t pmd_huge_pte; /* protected by page->ptl */
171 unsigned long _pt_pad_2; /* mapping */
172 union {
173 struct mm_struct *pt_mm; /* x86 pgds only */
174 atomic_t pt_frag_refcount; /* powerpc */
175 };
176 #if ALLOC_SPLIT_PTLOCKS
177 spinlock_t *ptl;
178 #else
179 spinlock_t ptl;
180 #endif
181 };
182 struct { /* ZONE_DEVICE pages */
183 /** @pgmap: Points to the hosting device page map. */
184 struct dev_pagemap *pgmap;
185 void *zone_device_data;
186 /*
187 * ZONE_DEVICE private pages are counted as being
188 * mapped so the next 3 words hold the mapping, index,
189 * and private fields from the source anonymous or
190 * page cache page while the page is migrated to device
191 * private memory.
192 * ZONE_DEVICE MEMORY_DEVICE_FS_DAX pages also
193 * use the mapping, index, and private fields when
194 * pmem backed DAX files are mapped.
195 */
196 };
197
198 /** @rcu_head: You can use this to free a page by RCU. */
199 struct rcu_head rcu_head;
200 };
201
202 union { /* This union is 4 bytes in size. */
203 /*
204 * If the page can be mapped to userspace, encodes the number
205 * of times this page is referenced by a page table.
206 */
207 atomic_t _mapcount;
208
209 /*
210 * If the page is neither PageSlab nor mappable to userspace,
211 * the value stored here may help determine what this page
212 * is used for. See page-flags.h for a list of page types
213 * which are currently stored here.
214 */
215 unsigned int page_type;
216
217 unsigned int active; /* SLAB */
218 int units; /* SLOB */
219 };
220
221 /* Usage count. *DO NOT USE DIRECTLY*. See page_ref.h */
222 atomic_t _refcount;
223
224 #ifdef CONFIG_MEMCG
225 unsigned long memcg_data;
226 #endif
227
228 /*
229 * On machines where all RAM is mapped into kernel address space,
230 * we can simply calculate the virtual address. On machines with
231 * highmem some memory is mapped into kernel virtual memory
232 * dynamically, so we need a place to store that address.
233 * Note that this field could be 16 bits on x86 ... ;)
234 *
235 * Architectures with slow multiplication can define
236 * WANT_PAGE_VIRTUAL in asm/page.h
237 */
238 #if defined(WANT_PAGE_VIRTUAL)
239 void *virtual; /* Kernel virtual address (NULL if
240 not kmapped, ie. highmem) */
241 #endif /* WANT_PAGE_VIRTUAL */
242
243 #ifdef LAST_CPUPID_NOT_IN_PAGE_FLAGS
244 int _last_cpupid;
245 #endif
246 } _struct_page_alignment;
247
compound_mapcount_ptr(struct page * page)248 static inline atomic_t *compound_mapcount_ptr(struct page *page)
249 {
250 return &page[1].compound_mapcount;
251 }
252
compound_pincount_ptr(struct page * page)253 static inline atomic_t *compound_pincount_ptr(struct page *page)
254 {
255 return &page[2].hpage_pinned_refcount;
256 }
257
258 /*
259 * Used for sizing the vmemmap region on some architectures
260 */
261 #define STRUCT_PAGE_MAX_SHIFT (order_base_2(sizeof(struct page)))
262
263 #define PAGE_FRAG_CACHE_MAX_SIZE __ALIGN_MASK(32768, ~PAGE_MASK)
264 #define PAGE_FRAG_CACHE_MAX_ORDER get_order(PAGE_FRAG_CACHE_MAX_SIZE)
265
266 #define page_private(page) ((page)->private)
267
set_page_private(struct page * page,unsigned long private)268 static inline void set_page_private(struct page *page, unsigned long private)
269 {
270 page->private = private;
271 }
272
273 struct page_frag_cache {
274 void * va;
275 #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
276 __u16 offset;
277 __u16 size;
278 #else
279 __u32 offset;
280 #endif
281 /* we maintain a pagecount bias, so that we dont dirty cache line
282 * containing page->_refcount every time we allocate a fragment.
283 */
284 unsigned int pagecnt_bias;
285 bool pfmemalloc;
286 };
287
288 typedef unsigned long vm_flags_t;
289
290 /*
291 * A region containing a mapping of a non-memory backed file under NOMMU
292 * conditions. These are held in a global tree and are pinned by the VMAs that
293 * map parts of them.
294 */
295 struct vm_region {
296 struct rb_node vm_rb; /* link in global region tree */
297 vm_flags_t vm_flags; /* VMA vm_flags */
298 unsigned long vm_start; /* start address of region */
299 unsigned long vm_end; /* region initialised to here */
300 unsigned long vm_top; /* region allocated to here */
301 unsigned long vm_pgoff; /* the offset in vm_file corresponding to vm_start */
302 struct file *vm_file; /* the backing file or NULL */
303
304 int vm_usage; /* region usage count (access under nommu_region_sem) */
305 bool vm_icache_flushed : 1; /* true if the icache has been flushed for
306 * this region */
307 };
308
309 #ifdef CONFIG_USERFAULTFD
310 #define NULL_VM_UFFD_CTX ((struct vm_userfaultfd_ctx) { NULL, })
311 struct vm_userfaultfd_ctx {
312 struct userfaultfd_ctx __rcu *ctx;
313 };
314 #else /* CONFIG_USERFAULTFD */
315 #define NULL_VM_UFFD_CTX ((struct vm_userfaultfd_ctx) {})
316 struct vm_userfaultfd_ctx {};
317 #endif /* CONFIG_USERFAULTFD */
318
319 struct anon_vma_name {
320 struct kref kref;
321 /* The name needs to be at the end because it is dynamically sized. */
322 char name[];
323 };
324
325 /*
326 * This struct describes a virtual memory area. There is one of these
327 * per VM-area/task. A VM area is any part of the process virtual memory
328 * space that has a special rule for the page-fault handlers (ie a shared
329 * library, the executable area etc).
330 *
331 * Note that speculative page faults make an on-stack copy of the VMA,
332 * so the structure size matters.
333 * (TODO - it would be preferable to copy only the required vma attributes
334 * rather than the entire vma).
335 */
336 struct vm_area_struct {
337 /* The first cache line has the info for VMA tree walking. */
338
339 union {
340 struct {
341 /* VMA covers [vm_start; vm_end) addresses within mm */
342 unsigned long vm_start, vm_end;
343
344 /* linked list of VMAs per task, sorted by address */
345 struct vm_area_struct *vm_next, *vm_prev;
346 };
347 #ifdef CONFIG_SPECULATIVE_PAGE_FAULT
348 struct rcu_head vm_rcu; /* Used for deferred freeing. */
349 #endif
350 };
351
352 struct rb_node vm_rb;
353
354 /*
355 * Largest free memory gap in bytes to the left of this VMA.
356 * Either between this VMA and vma->vm_prev, or between one of the
357 * VMAs below us in the VMA rbtree and its ->vm_prev. This helps
358 * get_unmapped_area find a free area of the right size.
359 */
360 unsigned long rb_subtree_gap;
361
362 /* Second cache line starts here. */
363
364 struct mm_struct *vm_mm; /* The address space we belong to. */
365
366 /*
367 * Access permissions of this VMA.
368 * See vmf_insert_mixed_prot() for discussion.
369 */
370 pgprot_t vm_page_prot;
371 unsigned long vm_flags; /* Flags, see mm.h. */
372
373 /*
374 * For areas with an address space and backing store,
375 * linkage into the address_space->i_mmap interval tree.
376 *
377 * For private anonymous mappings, a pointer to a null terminated string
378 * containing the name given to the vma, or NULL if unnamed.
379 */
380
381 union {
382 struct {
383 struct rb_node rb;
384 unsigned long rb_subtree_last;
385 } shared;
386 /*
387 * Serialized by mmap_sem. Never use directly because it is
388 * valid only when vm_file is NULL. Use anon_vma_name instead.
389 */
390 struct anon_vma_name *anon_name;
391 };
392
393 /*
394 * A file's MAP_PRIVATE vma can be in both i_mmap tree and anon_vma
395 * list, after a COW of one of the file pages. A MAP_SHARED vma
396 * can only be in the i_mmap tree. An anonymous MAP_PRIVATE, stack
397 * or brk vma (with NULL file) can only be in an anon_vma list.
398 */
399 struct list_head anon_vma_chain; /* Serialized by mmap_lock &
400 * page_table_lock */
401 struct anon_vma *anon_vma; /* Serialized by page_table_lock */
402
403 /* Function pointers to deal with this struct. */
404 const struct vm_operations_struct *vm_ops;
405
406 /* Information about our backing store: */
407 unsigned long vm_pgoff; /* Offset (within vm_file) in PAGE_SIZE
408 units */
409 struct file * vm_file; /* File we map to (can be NULL). */
410 void * vm_private_data; /* was vm_pte (shared mem) */
411
412 #ifdef CONFIG_SWAP
413 atomic_long_t swap_readahead_info;
414 #endif
415 #ifndef CONFIG_MMU
416 struct vm_region *vm_region; /* NOMMU mapping region */
417 #endif
418 #ifdef CONFIG_NUMA
419 struct mempolicy *vm_policy; /* NUMA policy for the VMA */
420 #endif
421 struct vm_userfaultfd_ctx vm_userfaultfd_ctx;
422 #ifdef CONFIG_SPECULATIVE_PAGE_FAULT
423 /*
424 * The name does not reflect the usage and is not renamed to keep
425 * the ABI intact.
426 * This is used to refcount VMA in get_vma/put_vma.
427 */
428 atomic_t file_ref_count;
429 #endif
430
431 ANDROID_KABI_RESERVE(1);
432 ANDROID_KABI_RESERVE(2);
433 ANDROID_KABI_RESERVE(3);
434 ANDROID_KABI_RESERVE(4);
435 } __randomize_layout;
436
437 struct core_thread {
438 struct task_struct *task;
439 struct core_thread *next;
440 };
441
442 struct core_state {
443 atomic_t nr_threads;
444 struct core_thread dumper;
445 struct completion startup;
446 };
447
448 struct kioctx_table;
449 struct percpu_rw_semaphore;
450 struct mm_struct {
451 struct {
452 struct vm_area_struct *mmap; /* list of VMAs */
453 struct rb_root mm_rb;
454 u64 vmacache_seqnum; /* per-thread vmacache */
455 #ifdef CONFIG_MMU
456 unsigned long (*get_unmapped_area) (struct file *filp,
457 unsigned long addr, unsigned long len,
458 unsigned long pgoff, unsigned long flags);
459 #endif
460 unsigned long mmap_base; /* base of mmap area */
461 unsigned long mmap_legacy_base; /* base of mmap area in bottom-up allocations */
462 #ifdef CONFIG_HAVE_ARCH_COMPAT_MMAP_BASES
463 /* Base addresses for compatible mmap() */
464 unsigned long mmap_compat_base;
465 unsigned long mmap_compat_legacy_base;
466 #endif
467 unsigned long task_size; /* size of task vm space */
468 unsigned long highest_vm_end; /* highest vma end address */
469 pgd_t * pgd;
470
471 #ifdef CONFIG_MEMBARRIER
472 /**
473 * @membarrier_state: Flags controlling membarrier behavior.
474 *
475 * This field is close to @pgd to hopefully fit in the same
476 * cache-line, which needs to be touched by switch_mm().
477 */
478 atomic_t membarrier_state;
479 #endif
480
481 /**
482 * @mm_users: The number of users including userspace.
483 *
484 * Use mmget()/mmget_not_zero()/mmput() to modify. When this
485 * drops to 0 (i.e. when the task exits and there are no other
486 * temporary reference holders), we also release a reference on
487 * @mm_count (which may then free the &struct mm_struct if
488 * @mm_count also drops to 0).
489 */
490 atomic_t mm_users;
491
492 /**
493 * @mm_count: The number of references to &struct mm_struct
494 * (@mm_users count as 1).
495 *
496 * Use mmgrab()/mmdrop() to modify. When this drops to 0, the
497 * &struct mm_struct is freed.
498 */
499 atomic_t mm_count;
500
501 #ifdef CONFIG_MMU
502 atomic_long_t pgtables_bytes; /* PTE page table pages */
503 #endif
504 int map_count; /* number of VMAs */
505
506 spinlock_t page_table_lock; /* Protects page tables and some
507 * counters
508 */
509 /*
510 * With some kernel config, the current mmap_lock's offset
511 * inside 'mm_struct' is at 0x120, which is very optimal, as
512 * its two hot fields 'count' and 'owner' sit in 2 different
513 * cachelines, and when mmap_lock is highly contended, both
514 * of the 2 fields will be accessed frequently, current layout
515 * will help to reduce cache bouncing.
516 *
517 * So please be careful with adding new fields before
518 * mmap_lock, which can easily push the 2 fields into one
519 * cacheline.
520 */
521 struct rw_semaphore mmap_lock;
522 #ifdef CONFIG_SPECULATIVE_PAGE_FAULT
523 unsigned long mmap_seq;
524 #endif
525
526
527 struct list_head mmlist; /* List of maybe swapped mm's. These
528 * are globally strung together off
529 * init_mm.mmlist, and are protected
530 * by mmlist_lock
531 */
532
533
534 unsigned long hiwater_rss; /* High-watermark of RSS usage */
535 unsigned long hiwater_vm; /* High-water virtual memory usage */
536
537 unsigned long total_vm; /* Total pages mapped */
538 unsigned long locked_vm; /* Pages that have PG_mlocked set */
539 atomic64_t pinned_vm; /* Refcount permanently increased */
540 unsigned long data_vm; /* VM_WRITE & ~VM_SHARED & ~VM_STACK */
541 unsigned long exec_vm; /* VM_EXEC & ~VM_WRITE & ~VM_STACK */
542 unsigned long stack_vm; /* VM_STACK */
543 unsigned long def_flags;
544
545 /**
546 * @write_protect_seq: Locked when any thread is write
547 * protecting pages mapped by this mm to enforce a later COW,
548 * for instance during page table copying for fork().
549 */
550 seqcount_t write_protect_seq;
551
552 spinlock_t arg_lock; /* protect the below fields */
553
554 unsigned long start_code, end_code, start_data, end_data;
555 unsigned long start_brk, brk, start_stack;
556 unsigned long arg_start, arg_end, env_start, env_end;
557
558 unsigned long saved_auxv[AT_VECTOR_SIZE]; /* for /proc/PID/auxv */
559
560 /*
561 * Special counters, in some configurations protected by the
562 * page_table_lock, in other configurations by being atomic.
563 */
564 struct mm_rss_stat rss_stat;
565
566 struct linux_binfmt *binfmt;
567
568 /* Architecture-specific MM context */
569 mm_context_t context;
570
571 unsigned long flags; /* Must use atomic bitops to access */
572
573 struct core_state *core_state; /* coredumping support */
574
575 #ifdef CONFIG_AIO
576 spinlock_t ioctx_lock;
577 struct kioctx_table __rcu *ioctx_table;
578 #endif
579 #ifdef CONFIG_MEMCG
580 /*
581 * "owner" points to a task that is regarded as the canonical
582 * user/owner of this mm. All of the following must be true in
583 * order for it to be changed:
584 *
585 * current == mm->owner
586 * current->mm != mm
587 * new_owner->mm == mm
588 * new_owner->alloc_lock is held
589 */
590 struct task_struct __rcu *owner;
591 #endif
592 struct user_namespace *user_ns;
593
594 /* store ref to file /proc/<pid>/exe symlink points to */
595 struct file __rcu *exe_file;
596 #ifdef CONFIG_MMU_NOTIFIER
597 struct mmu_notifier_subscriptions *notifier_subscriptions;
598 #ifdef CONFIG_SPECULATIVE_PAGE_FAULT
599 struct percpu_rw_semaphore *mmu_notifier_lock;
600 #endif /* CONFIG_SPECULATIVE_PAGE_FAULT */
601 #endif /* CONFIG_MMU_NOTIFIER */
602 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
603 pgtable_t pmd_huge_pte; /* protected by page_table_lock */
604 #endif
605 #ifdef CONFIG_NUMA_BALANCING
606 /*
607 * numa_next_scan is the next time that the PTEs will be marked
608 * pte_numa. NUMA hinting faults will gather statistics and
609 * migrate pages to new nodes if necessary.
610 */
611 unsigned long numa_next_scan;
612
613 /* Restart point for scanning and setting pte_numa */
614 unsigned long numa_scan_offset;
615
616 /* numa_scan_seq prevents two threads setting pte_numa */
617 int numa_scan_seq;
618 #endif
619 /*
620 * An operation with batched TLB flushing is going on. Anything
621 * that can move process memory needs to flush the TLB when
622 * moving a PROT_NONE or PROT_NUMA mapped page.
623 */
624 atomic_t tlb_flush_pending;
625 #ifdef CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH
626 /* See flush_tlb_batched_pending() */
627 bool tlb_flush_batched;
628 #endif
629 struct uprobes_state uprobes_state;
630 #ifdef CONFIG_HUGETLB_PAGE
631 atomic_long_t hugetlb_usage;
632 #endif
633 struct work_struct async_put_work;
634
635 #ifdef CONFIG_IOMMU_SUPPORT
636 u32 pasid;
637 #endif
638 #ifdef CONFIG_LRU_GEN
639 struct {
640 /* this mm_struct is on lru_gen_mm_list */
641 struct list_head list;
642 #ifdef CONFIG_MEMCG
643 /* points to the memcg of "owner" above */
644 struct mem_cgroup *memcg;
645 #endif
646 /*
647 * Set when switching to this mm_struct, as a hint of
648 * whether it has been used since the last time per-node
649 * page table walkers cleared the corresponding bits.
650 */
651 nodemask_t nodes;
652 } lru_gen;
653 #endif /* CONFIG_LRU_GEN */
654
655 ANDROID_KABI_RESERVE(1);
656 } __randomize_layout;
657
658 /*
659 * The mm_cpumask needs to be at the end of mm_struct, because it
660 * is dynamically sized based on nr_cpu_ids.
661 */
662 unsigned long cpu_bitmap[];
663 };
664
665 extern struct mm_struct init_mm;
666
667 /* Pointer magic because the dynamic array size confuses some compilers. */
mm_init_cpumask(struct mm_struct * mm)668 static inline void mm_init_cpumask(struct mm_struct *mm)
669 {
670 unsigned long cpu_bitmap = (unsigned long)mm;
671
672 cpu_bitmap += offsetof(struct mm_struct, cpu_bitmap);
673 cpumask_clear((struct cpumask *)cpu_bitmap);
674 }
675
676 /* Future-safe accessor for struct mm_struct's cpu_vm_mask. */
mm_cpumask(struct mm_struct * mm)677 static inline cpumask_t *mm_cpumask(struct mm_struct *mm)
678 {
679 return (struct cpumask *)&mm->cpu_bitmap;
680 }
681
682 #ifdef CONFIG_LRU_GEN
683
684 struct lru_gen_mm_list {
685 /* mm_struct list for page table walkers */
686 struct list_head fifo;
687 /* protects the list above */
688 spinlock_t lock;
689 };
690
691 void lru_gen_add_mm(struct mm_struct *mm);
692 void lru_gen_del_mm(struct mm_struct *mm);
693 #ifdef CONFIG_MEMCG
694 void lru_gen_migrate_mm(struct mm_struct *mm);
695 #endif
696
lru_gen_init_mm(struct mm_struct * mm)697 static inline void lru_gen_init_mm(struct mm_struct *mm)
698 {
699 INIT_LIST_HEAD(&mm->lru_gen.list);
700 nodes_clear(mm->lru_gen.nodes);
701 #ifdef CONFIG_MEMCG
702 mm->lru_gen.memcg = NULL;
703 #endif
704 }
705
lru_gen_use_mm(struct mm_struct * mm)706 static inline void lru_gen_use_mm(struct mm_struct *mm)
707 {
708 /*
709 * When the bitmap is set, page reclaim knows this mm_struct has been
710 * used since the last time it cleared the bitmap. So it might be worth
711 * walking the page tables of this mm_struct to clear the accessed bit.
712 */
713 nodes_setall(mm->lru_gen.nodes);
714 }
715
716 #else /* !CONFIG_LRU_GEN */
717
lru_gen_add_mm(struct mm_struct * mm)718 static inline void lru_gen_add_mm(struct mm_struct *mm)
719 {
720 }
721
lru_gen_del_mm(struct mm_struct * mm)722 static inline void lru_gen_del_mm(struct mm_struct *mm)
723 {
724 }
725
726 #ifdef CONFIG_MEMCG
lru_gen_migrate_mm(struct mm_struct * mm)727 static inline void lru_gen_migrate_mm(struct mm_struct *mm)
728 {
729 }
730 #endif
731
lru_gen_init_mm(struct mm_struct * mm)732 static inline void lru_gen_init_mm(struct mm_struct *mm)
733 {
734 }
735
lru_gen_use_mm(struct mm_struct * mm)736 static inline void lru_gen_use_mm(struct mm_struct *mm)
737 {
738 }
739
740 #endif /* CONFIG_LRU_GEN */
741
742 struct mmu_gather;
743 extern void tlb_gather_mmu(struct mmu_gather *tlb, struct mm_struct *mm);
744 extern void tlb_gather_mmu_fullmm(struct mmu_gather *tlb, struct mm_struct *mm);
745 extern void tlb_finish_mmu(struct mmu_gather *tlb);
746
init_tlb_flush_pending(struct mm_struct * mm)747 static inline void init_tlb_flush_pending(struct mm_struct *mm)
748 {
749 atomic_set(&mm->tlb_flush_pending, 0);
750 }
751
inc_tlb_flush_pending(struct mm_struct * mm)752 static inline void inc_tlb_flush_pending(struct mm_struct *mm)
753 {
754 atomic_inc(&mm->tlb_flush_pending);
755 /*
756 * The only time this value is relevant is when there are indeed pages
757 * to flush. And we'll only flush pages after changing them, which
758 * requires the PTL.
759 *
760 * So the ordering here is:
761 *
762 * atomic_inc(&mm->tlb_flush_pending);
763 * spin_lock(&ptl);
764 * ...
765 * set_pte_at();
766 * spin_unlock(&ptl);
767 *
768 * spin_lock(&ptl)
769 * mm_tlb_flush_pending();
770 * ....
771 * spin_unlock(&ptl);
772 *
773 * flush_tlb_range();
774 * atomic_dec(&mm->tlb_flush_pending);
775 *
776 * Where the increment if constrained by the PTL unlock, it thus
777 * ensures that the increment is visible if the PTE modification is
778 * visible. After all, if there is no PTE modification, nobody cares
779 * about TLB flushes either.
780 *
781 * This very much relies on users (mm_tlb_flush_pending() and
782 * mm_tlb_flush_nested()) only caring about _specific_ PTEs (and
783 * therefore specific PTLs), because with SPLIT_PTE_PTLOCKS and RCpc
784 * locks (PPC) the unlock of one doesn't order against the lock of
785 * another PTL.
786 *
787 * The decrement is ordered by the flush_tlb_range(), such that
788 * mm_tlb_flush_pending() will not return false unless all flushes have
789 * completed.
790 */
791 }
792
dec_tlb_flush_pending(struct mm_struct * mm)793 static inline void dec_tlb_flush_pending(struct mm_struct *mm)
794 {
795 /*
796 * See inc_tlb_flush_pending().
797 *
798 * This cannot be smp_mb__before_atomic() because smp_mb() simply does
799 * not order against TLB invalidate completion, which is what we need.
800 *
801 * Therefore we must rely on tlb_flush_*() to guarantee order.
802 */
803 atomic_dec(&mm->tlb_flush_pending);
804 }
805
mm_tlb_flush_pending(struct mm_struct * mm)806 static inline bool mm_tlb_flush_pending(struct mm_struct *mm)
807 {
808 /*
809 * Must be called after having acquired the PTL; orders against that
810 * PTLs release and therefore ensures that if we observe the modified
811 * PTE we must also observe the increment from inc_tlb_flush_pending().
812 *
813 * That is, it only guarantees to return true if there is a flush
814 * pending for _this_ PTL.
815 */
816 return atomic_read(&mm->tlb_flush_pending);
817 }
818
mm_tlb_flush_nested(struct mm_struct * mm)819 static inline bool mm_tlb_flush_nested(struct mm_struct *mm)
820 {
821 /*
822 * Similar to mm_tlb_flush_pending(), we must have acquired the PTL
823 * for which there is a TLB flush pending in order to guarantee
824 * we've seen both that PTE modification and the increment.
825 *
826 * (no requirement on actually still holding the PTL, that is irrelevant)
827 */
828 return atomic_read(&mm->tlb_flush_pending) > 1;
829 }
830
831 struct vm_fault;
832
833 /**
834 * typedef vm_fault_t - Return type for page fault handlers.
835 *
836 * Page fault handlers return a bitmask of %VM_FAULT values.
837 */
838 typedef __bitwise unsigned int vm_fault_t;
839
840 /**
841 * enum vm_fault_reason - Page fault handlers return a bitmask of
842 * these values to tell the core VM what happened when handling the
843 * fault. Used to decide whether a process gets delivered SIGBUS or
844 * just gets major/minor fault counters bumped up.
845 *
846 * @VM_FAULT_OOM: Out Of Memory
847 * @VM_FAULT_SIGBUS: Bad access
848 * @VM_FAULT_MAJOR: Page read from storage
849 * @VM_FAULT_WRITE: Special case for get_user_pages
850 * @VM_FAULT_HWPOISON: Hit poisoned small page
851 * @VM_FAULT_HWPOISON_LARGE: Hit poisoned large page. Index encoded
852 * in upper bits
853 * @VM_FAULT_SIGSEGV: segmentation fault
854 * @VM_FAULT_NOPAGE: ->fault installed the pte, not return page
855 * @VM_FAULT_LOCKED: ->fault locked the returned page
856 * @VM_FAULT_RETRY: ->fault blocked, must retry
857 * @VM_FAULT_FALLBACK: huge page fault failed, fall back to small
858 * @VM_FAULT_DONE_COW: ->fault has fully handled COW
859 * @VM_FAULT_NEEDDSYNC: ->fault did not modify page tables and needs
860 * fsync() to complete (for synchronous page faults
861 * in DAX)
862 * @VM_FAULT_HINDEX_MASK: mask HINDEX value
863 *
864 */
865 enum vm_fault_reason {
866 VM_FAULT_OOM = (__force vm_fault_t)0x000001,
867 VM_FAULT_SIGBUS = (__force vm_fault_t)0x000002,
868 VM_FAULT_MAJOR = (__force vm_fault_t)0x000004,
869 VM_FAULT_WRITE = (__force vm_fault_t)0x000008,
870 VM_FAULT_HWPOISON = (__force vm_fault_t)0x000010,
871 VM_FAULT_HWPOISON_LARGE = (__force vm_fault_t)0x000020,
872 VM_FAULT_SIGSEGV = (__force vm_fault_t)0x000040,
873 VM_FAULT_NOPAGE = (__force vm_fault_t)0x000100,
874 VM_FAULT_LOCKED = (__force vm_fault_t)0x000200,
875 VM_FAULT_RETRY = (__force vm_fault_t)0x000400,
876 VM_FAULT_FALLBACK = (__force vm_fault_t)0x000800,
877 VM_FAULT_DONE_COW = (__force vm_fault_t)0x001000,
878 VM_FAULT_NEEDDSYNC = (__force vm_fault_t)0x002000,
879 VM_FAULT_HINDEX_MASK = (__force vm_fault_t)0x0f0000,
880 };
881
882 /* Encode hstate index for a hwpoisoned large page */
883 #define VM_FAULT_SET_HINDEX(x) ((__force vm_fault_t)((x) << 16))
884 #define VM_FAULT_GET_HINDEX(x) (((__force unsigned int)(x) >> 16) & 0xf)
885
886 #define VM_FAULT_ERROR (VM_FAULT_OOM | VM_FAULT_SIGBUS | \
887 VM_FAULT_SIGSEGV | VM_FAULT_HWPOISON | \
888 VM_FAULT_HWPOISON_LARGE | VM_FAULT_FALLBACK)
889
890 #define VM_FAULT_RESULT_TRACE \
891 { VM_FAULT_OOM, "OOM" }, \
892 { VM_FAULT_SIGBUS, "SIGBUS" }, \
893 { VM_FAULT_MAJOR, "MAJOR" }, \
894 { VM_FAULT_WRITE, "WRITE" }, \
895 { VM_FAULT_HWPOISON, "HWPOISON" }, \
896 { VM_FAULT_HWPOISON_LARGE, "HWPOISON_LARGE" }, \
897 { VM_FAULT_SIGSEGV, "SIGSEGV" }, \
898 { VM_FAULT_NOPAGE, "NOPAGE" }, \
899 { VM_FAULT_LOCKED, "LOCKED" }, \
900 { VM_FAULT_RETRY, "RETRY" }, \
901 { VM_FAULT_FALLBACK, "FALLBACK" }, \
902 { VM_FAULT_DONE_COW, "DONE_COW" }, \
903 { VM_FAULT_NEEDDSYNC, "NEEDDSYNC" }
904
905 struct vm_special_mapping {
906 const char *name; /* The name, e.g. "[vdso]". */
907
908 /*
909 * If .fault is not provided, this points to a
910 * NULL-terminated array of pages that back the special mapping.
911 *
912 * This must not be NULL unless .fault is provided.
913 */
914 struct page **pages;
915
916 /*
917 * If non-NULL, then this is called to resolve page faults
918 * on the special mapping. If used, .pages is not checked.
919 */
920 vm_fault_t (*fault)(const struct vm_special_mapping *sm,
921 struct vm_area_struct *vma,
922 struct vm_fault *vmf);
923
924 int (*mremap)(const struct vm_special_mapping *sm,
925 struct vm_area_struct *new_vma);
926 };
927
928 enum tlb_flush_reason {
929 TLB_FLUSH_ON_TASK_SWITCH,
930 TLB_REMOTE_SHOOTDOWN,
931 TLB_LOCAL_SHOOTDOWN,
932 TLB_LOCAL_MM_SHOOTDOWN,
933 TLB_REMOTE_SEND_IPI,
934 NR_TLB_FLUSH_REASONS,
935 };
936
937 /*
938 * A swap entry has to fit into a "unsigned long", as the entry is hidden
939 * in the "index" field of the swapper address space.
940 */
941 typedef struct {
942 unsigned long val;
943 } swp_entry_t;
944
945 #endif /* _LINUX_MM_TYPES_H */
946