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/page-flags-layout.h>
17 #include <linux/workqueue.h>
18 #include <linux/seqlock.h>
19 #include <linux/android_kabi.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
61 * counters. That requires that freelist & counters be adjacent and
62 * double-word aligned. We align all struct pages to double-word
63 * boundaries, and ensure that 'freelist' is aligned within the
64 * struct.
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 union {
84 /**
85 * @lru: Pageout list, eg. active_list protected by
86 * lruvec->lru_lock. Sometimes used as a generic list
87 * by the page owner.
88 */
89 struct list_head lru;
90
91 /* Or, free page */
92 struct list_head buddy_list;
93 struct list_head pcp_list;
94 };
95 /* See page-flags.h for PAGE_MAPPING_FLAGS */
96 struct address_space *mapping;
97 pgoff_t index; /* Our offset within mapping. */
98 /**
99 * @private: Mapping-private opaque data.
100 * Usually used for buffer_heads if PagePrivate.
101 * Used for swp_entry_t if PageSwapCache.
102 * Indicates order in the buddy system if PageBuddy.
103 */
104 unsigned long private;
105 };
106 struct { /* page_pool used by netstack */
107 /**
108 * @pp_magic: magic value to avoid recycling non
109 * page_pool allocated pages.
110 */
111 unsigned long pp_magic;
112 struct page_pool *pp;
113 unsigned long _pp_mapping_pad;
114 unsigned long dma_addr;
115 union {
116 /**
117 * dma_addr_upper: might require a 64-bit
118 * value on 32-bit architectures.
119 */
120 unsigned long dma_addr_upper;
121 /**
122 * For frag page support, not supported in
123 * 32-bit architectures with 64-bit DMA.
124 */
125 atomic_long_t pp_frag_count;
126 };
127 };
128 struct { /* slab, slob and slub */
129 union {
130 struct list_head slab_list;
131 struct { /* Partial pages */
132 struct page *next;
133 #ifdef CONFIG_64BIT
134 int pages; /* Nr of pages left */
135 int pobjects; /* Approximate count */
136 #else
137 short int pages;
138 short int pobjects;
139 #endif
140 };
141 };
142 struct kmem_cache *slab_cache; /* not slob */
143 /* Double-word boundary */
144 void *freelist; /* first free object */
145 union {
146 void *s_mem; /* slab: first object */
147 unsigned long counters; /* SLUB */
148 struct { /* SLUB */
149 unsigned inuse:16;
150 unsigned objects:15;
151 unsigned frozen:1;
152 };
153 };
154 };
155 struct { /* Tail pages of compound page */
156 unsigned long compound_head; /* Bit zero is set */
157
158 /* First tail page only */
159 unsigned char compound_dtor;
160 unsigned char compound_order;
161 atomic_t compound_mapcount;
162 unsigned int compound_nr; /* 1 << compound_order */
163 };
164 struct { /* Second tail page of compound page */
165 unsigned long _compound_pad_1; /* compound_head */
166 atomic_t hpage_pinned_refcount;
167 /* For both global and memcg */
168 struct list_head deferred_list;
169 };
170 struct { /* Page table pages */
171 unsigned long _pt_pad_1; /* compound_head */
172 pgtable_t pmd_huge_pte; /* protected by page->ptl */
173 unsigned long _pt_pad_2; /* mapping */
174 union {
175 struct mm_struct *pt_mm; /* x86 pgds only */
176 atomic_t pt_frag_refcount; /* powerpc */
177 };
178 #if ALLOC_SPLIT_PTLOCKS
179 spinlock_t *ptl;
180 #else
181 spinlock_t ptl;
182 #endif
183 };
184 struct { /* ZONE_DEVICE pages */
185 /** @pgmap: Points to the hosting device page map. */
186 struct dev_pagemap *pgmap;
187 void *zone_device_data;
188 /*
189 * ZONE_DEVICE private pages are counted as being
190 * mapped so the next 3 words hold the mapping, index,
191 * and private fields from the source anonymous or
192 * page cache page while the page is migrated to device
193 * private memory.
194 * ZONE_DEVICE MEMORY_DEVICE_FS_DAX pages also
195 * use the mapping, index, and private fields when
196 * pmem backed DAX files are mapped.
197 */
198 };
199
200 /** @rcu_head: You can use this to free a page by RCU. */
201 struct rcu_head rcu_head;
202 };
203
204 union { /* This union is 4 bytes in size. */
205 /*
206 * If the page can be mapped to userspace, encodes the number
207 * of times this page is referenced by a page table.
208 */
209 atomic_t _mapcount;
210
211 /*
212 * If the page is neither PageSlab nor mappable to userspace,
213 * the value stored here may help determine what this page
214 * is used for. See page-flags.h for a list of page types
215 * which are currently stored here.
216 */
217 unsigned int page_type;
218
219 unsigned int active; /* SLAB */
220 int units; /* SLOB */
221 };
222
223 /* Usage count. *DO NOT USE DIRECTLY*. See page_ref.h */
224 atomic_t _refcount;
225
226 #ifdef CONFIG_MEMCG
227 unsigned long memcg_data;
228 #endif
229
230 /*
231 * On machines where all RAM is mapped into kernel address space,
232 * we can simply calculate the virtual address. On machines with
233 * highmem some memory is mapped into kernel virtual memory
234 * dynamically, so we need a place to store that address.
235 * Note that this field could be 16 bits on x86 ... ;)
236 *
237 * Architectures with slow multiplication can define
238 * WANT_PAGE_VIRTUAL in asm/page.h
239 */
240 #if defined(WANT_PAGE_VIRTUAL)
241 void *virtual; /* Kernel virtual address (NULL if
242 not kmapped, ie. highmem) */
243 #endif /* WANT_PAGE_VIRTUAL */
244
245 #ifdef LAST_CPUPID_NOT_IN_PAGE_FLAGS
246 int _last_cpupid;
247 #endif
248 } _struct_page_alignment;
249
compound_mapcount_ptr(struct page * page)250 static inline atomic_t *compound_mapcount_ptr(struct page *page)
251 {
252 return &page[1].compound_mapcount;
253 }
254
compound_pincount_ptr(struct page * page)255 static inline atomic_t *compound_pincount_ptr(struct page *page)
256 {
257 return &page[2].hpage_pinned_refcount;
258 }
259
260 /*
261 * Used for sizing the vmemmap region on some architectures
262 */
263 #define STRUCT_PAGE_MAX_SHIFT (order_base_2(sizeof(struct page)))
264
265 #define PAGE_FRAG_CACHE_MAX_SIZE __ALIGN_MASK(32768, ~PAGE_MASK)
266 #define PAGE_FRAG_CACHE_MAX_ORDER get_order(PAGE_FRAG_CACHE_MAX_SIZE)
267
268 #define page_private(page) ((page)->private)
269
set_page_private(struct page * page,unsigned long private)270 static inline void set_page_private(struct page *page, unsigned long private)
271 {
272 page->private = private;
273 }
274
275 struct page_frag_cache {
276 void * va;
277 #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
278 __u16 offset;
279 __u16 size;
280 #else
281 __u32 offset;
282 #endif
283 /* we maintain a pagecount bias, so that we dont dirty cache line
284 * containing page->_refcount every time we allocate a fragment.
285 */
286 unsigned int pagecnt_bias;
287 bool pfmemalloc;
288 };
289
290 typedef unsigned long vm_flags_t;
291
292 /*
293 * A region containing a mapping of a non-memory backed file under NOMMU
294 * conditions. These are held in a global tree and are pinned by the VMAs that
295 * map parts of them.
296 */
297 struct vm_region {
298 struct rb_node vm_rb; /* link in global region tree */
299 vm_flags_t vm_flags; /* VMA vm_flags */
300 unsigned long vm_start; /* start address of region */
301 unsigned long vm_end; /* region initialised to here */
302 unsigned long vm_top; /* region allocated to here */
303 unsigned long vm_pgoff; /* the offset in vm_file corresponding to vm_start */
304 struct file *vm_file; /* the backing file or NULL */
305
306 int vm_usage; /* region usage count (access under nommu_region_sem) */
307 bool vm_icache_flushed : 1; /* true if the icache has been flushed for
308 * this region */
309 };
310
311 #ifdef CONFIG_USERFAULTFD
312 #define NULL_VM_UFFD_CTX ((struct vm_userfaultfd_ctx) { NULL, })
313 struct vm_userfaultfd_ctx {
314 struct userfaultfd_ctx __rcu *ctx;
315 };
316 #else /* CONFIG_USERFAULTFD */
317 #define NULL_VM_UFFD_CTX ((struct vm_userfaultfd_ctx) {})
318 struct vm_userfaultfd_ctx {};
319 #endif /* CONFIG_USERFAULTFD */
320
321 struct anon_vma_name {
322 struct kref kref;
323 /* The name needs to be at the end because it is dynamically sized. */
324 char name[];
325 };
326
327 /*
328 * This struct describes a virtual memory area. There is one of these
329 * per VM-area/task. A VM area is any part of the process virtual memory
330 * space that has a special rule for the page-fault handlers (ie a shared
331 * library, the executable area etc).
332 *
333 * Note that speculative page faults make an on-stack copy of the VMA,
334 * so the structure size matters.
335 * (TODO - it would be preferable to copy only the required vma attributes
336 * rather than the entire vma).
337 */
338 struct vm_area_struct {
339 /* The first cache line has the info for VMA tree walking. */
340
341 union {
342 struct {
343 /* VMA covers [vm_start; vm_end) addresses within mm */
344 unsigned long vm_start, vm_end;
345
346 /* linked list of VMAs per task, sorted by address */
347 struct vm_area_struct *vm_next, *vm_prev;
348 };
349 #ifdef CONFIG_SPECULATIVE_PAGE_FAULT
350 struct rcu_head vm_rcu; /* Used for deferred freeing. */
351 #endif
352 };
353
354 struct rb_node vm_rb;
355
356 /*
357 * Largest free memory gap in bytes to the left of this VMA.
358 * Either between this VMA and vma->vm_prev, or between one of the
359 * VMAs below us in the VMA rbtree and its ->vm_prev. This helps
360 * get_unmapped_area find a free area of the right size.
361 */
362 unsigned long rb_subtree_gap;
363
364 /* Second cache line starts here. */
365
366 struct mm_struct *vm_mm; /* The address space we belong to. */
367
368 /*
369 * Access permissions of this VMA.
370 * See vmf_insert_mixed_prot() for discussion.
371 */
372 pgprot_t vm_page_prot;
373 unsigned long vm_flags; /* Flags, see mm.h. */
374
375 /*
376 * For areas with an address space and backing store,
377 * linkage into the address_space->i_mmap interval tree.
378 *
379 * For private anonymous mappings, a pointer to a null terminated string
380 * containing the name given to the vma, or NULL if unnamed.
381 */
382
383 union {
384 struct {
385 struct rb_node rb;
386 unsigned long rb_subtree_last;
387 } shared;
388 /*
389 * Serialized by mmap_sem. Never use directly because it is
390 * valid only when vm_file is NULL. Use anon_vma_name instead.
391 */
392 struct anon_vma_name *anon_name;
393 };
394
395 /*
396 * A file's MAP_PRIVATE vma can be in both i_mmap tree and anon_vma
397 * list, after a COW of one of the file pages. A MAP_SHARED vma
398 * can only be in the i_mmap tree. An anonymous MAP_PRIVATE, stack
399 * or brk vma (with NULL file) can only be in an anon_vma list.
400 */
401 struct list_head anon_vma_chain; /* Serialized by mmap_lock &
402 * page_table_lock */
403 struct anon_vma *anon_vma; /* Serialized by page_table_lock */
404
405 /* Function pointers to deal with this struct. */
406 const struct vm_operations_struct *vm_ops;
407
408 /* Information about our backing store: */
409 unsigned long vm_pgoff; /* Offset (within vm_file) in PAGE_SIZE
410 units */
411 struct file * vm_file; /* File we map to (can be NULL). */
412 void * vm_private_data; /* was vm_pte (shared mem) */
413
414 #ifdef CONFIG_SWAP
415 atomic_long_t swap_readahead_info;
416 #endif
417 #ifndef CONFIG_MMU
418 struct vm_region *vm_region; /* NOMMU mapping region */
419 #endif
420 #ifdef CONFIG_NUMA
421 struct mempolicy *vm_policy; /* NUMA policy for the VMA */
422 #endif
423 struct vm_userfaultfd_ctx vm_userfaultfd_ctx;
424 #ifdef CONFIG_SPECULATIVE_PAGE_FAULT
425 /*
426 * The name does not reflect the usage and is not renamed to keep
427 * the ABI intact.
428 * This is used to refcount VMA in get_vma/put_vma.
429 */
430 atomic_t file_ref_count;
431 #endif
432
433 ANDROID_KABI_RESERVE(1);
434 ANDROID_KABI_RESERVE(2);
435 ANDROID_KABI_RESERVE(3);
436 ANDROID_KABI_RESERVE(4);
437 } __randomize_layout;
438
439 struct core_thread {
440 struct task_struct *task;
441 struct core_thread *next;
442 };
443
444 struct core_state {
445 atomic_t nr_threads;
446 struct core_thread dumper;
447 struct completion startup;
448 };
449
450 struct kioctx_table;
451 struct percpu_rw_semaphore;
452 struct mm_struct {
453 struct {
454 struct vm_area_struct *mmap; /* list of VMAs */
455 struct rb_root mm_rb;
456 u64 vmacache_seqnum; /* per-thread vmacache */
457 #ifdef CONFIG_MMU
458 unsigned long (*get_unmapped_area) (struct file *filp,
459 unsigned long addr, unsigned long len,
460 unsigned long pgoff, unsigned long flags);
461 #endif
462 unsigned long mmap_base; /* base of mmap area */
463 unsigned long mmap_legacy_base; /* base of mmap area in bottom-up allocations */
464 #ifdef CONFIG_HAVE_ARCH_COMPAT_MMAP_BASES
465 /* Base addresses for compatible mmap() */
466 unsigned long mmap_compat_base;
467 unsigned long mmap_compat_legacy_base;
468 #endif
469 unsigned long task_size; /* size of task vm space */
470 unsigned long highest_vm_end; /* highest vma end address */
471 pgd_t * pgd;
472
473 #ifdef CONFIG_MEMBARRIER
474 /**
475 * @membarrier_state: Flags controlling membarrier behavior.
476 *
477 * This field is close to @pgd to hopefully fit in the same
478 * cache-line, which needs to be touched by switch_mm().
479 */
480 atomic_t membarrier_state;
481 #endif
482
483 /**
484 * @mm_users: The number of users including userspace.
485 *
486 * Use mmget()/mmget_not_zero()/mmput() to modify. When this
487 * drops to 0 (i.e. when the task exits and there are no other
488 * temporary reference holders), we also release a reference on
489 * @mm_count (which may then free the &struct mm_struct if
490 * @mm_count also drops to 0).
491 */
492 atomic_t mm_users;
493
494 /**
495 * @mm_count: The number of references to &struct mm_struct
496 * (@mm_users count as 1).
497 *
498 * Use mmgrab()/mmdrop() to modify. When this drops to 0, the
499 * &struct mm_struct is freed.
500 */
501 atomic_t mm_count;
502
503 #ifdef CONFIG_MMU
504 atomic_long_t pgtables_bytes; /* PTE page table pages */
505 #endif
506 int map_count; /* number of VMAs */
507
508 spinlock_t page_table_lock; /* Protects page tables and some
509 * counters
510 */
511 /*
512 * With some kernel config, the current mmap_lock's offset
513 * inside 'mm_struct' is at 0x120, which is very optimal, as
514 * its two hot fields 'count' and 'owner' sit in 2 different
515 * cachelines, and when mmap_lock is highly contended, both
516 * of the 2 fields will be accessed frequently, current layout
517 * will help to reduce cache bouncing.
518 *
519 * So please be careful with adding new fields before
520 * mmap_lock, which can easily push the 2 fields into one
521 * cacheline.
522 */
523 struct rw_semaphore mmap_lock;
524 #ifdef CONFIG_SPECULATIVE_PAGE_FAULT
525 unsigned long mmap_seq;
526 #endif
527
528
529 struct list_head mmlist; /* List of maybe swapped mm's. These
530 * are globally strung together off
531 * init_mm.mmlist, and are protected
532 * by mmlist_lock
533 */
534
535
536 unsigned long hiwater_rss; /* High-watermark of RSS usage */
537 unsigned long hiwater_vm; /* High-water virtual memory usage */
538
539 unsigned long total_vm; /* Total pages mapped */
540 unsigned long locked_vm; /* Pages that have PG_mlocked set */
541 atomic64_t pinned_vm; /* Refcount permanently increased */
542 unsigned long data_vm; /* VM_WRITE & ~VM_SHARED & ~VM_STACK */
543 unsigned long exec_vm; /* VM_EXEC & ~VM_WRITE & ~VM_STACK */
544 unsigned long stack_vm; /* VM_STACK */
545 unsigned long def_flags;
546
547 /**
548 * @write_protect_seq: Locked when any thread is write
549 * protecting pages mapped by this mm to enforce a later COW,
550 * for instance during page table copying for fork().
551 */
552 seqcount_t write_protect_seq;
553
554 spinlock_t arg_lock; /* protect the below fields */
555
556 unsigned long start_code, end_code, start_data, end_data;
557 unsigned long start_brk, brk, start_stack;
558 unsigned long arg_start, arg_end, env_start, env_end;
559
560 unsigned long saved_auxv[AT_VECTOR_SIZE]; /* for /proc/PID/auxv */
561
562 /*
563 * Special counters, in some configurations protected by the
564 * page_table_lock, in other configurations by being atomic.
565 */
566 struct mm_rss_stat rss_stat;
567
568 struct linux_binfmt *binfmt;
569
570 /* Architecture-specific MM context */
571 mm_context_t context;
572
573 unsigned long flags; /* Must use atomic bitops to access */
574
575 struct core_state *core_state; /* coredumping support */
576
577 #ifdef CONFIG_AIO
578 spinlock_t ioctx_lock;
579 struct kioctx_table __rcu *ioctx_table;
580 #endif
581 #ifdef CONFIG_MEMCG
582 /*
583 * "owner" points to a task that is regarded as the canonical
584 * user/owner of this mm. All of the following must be true in
585 * order for it to be changed:
586 *
587 * current == mm->owner
588 * current->mm != mm
589 * new_owner->mm == mm
590 * new_owner->alloc_lock is held
591 */
592 struct task_struct __rcu *owner;
593 #endif
594 struct user_namespace *user_ns;
595
596 /* store ref to file /proc/<pid>/exe symlink points to */
597 struct file __rcu *exe_file;
598 #ifdef CONFIG_MMU_NOTIFIER
599 struct mmu_notifier_subscriptions *notifier_subscriptions;
600 #ifdef CONFIG_SPECULATIVE_PAGE_FAULT
601 struct percpu_rw_semaphore *mmu_notifier_lock;
602 #endif /* CONFIG_SPECULATIVE_PAGE_FAULT */
603 #endif /* CONFIG_MMU_NOTIFIER */
604 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
605 pgtable_t pmd_huge_pte; /* protected by page_table_lock */
606 #endif
607 #ifdef CONFIG_NUMA_BALANCING
608 /*
609 * numa_next_scan is the next time that the PTEs will be marked
610 * pte_numa. NUMA hinting faults will gather statistics and
611 * migrate pages to new nodes if necessary.
612 */
613 unsigned long numa_next_scan;
614
615 /* Restart point for scanning and setting pte_numa */
616 unsigned long numa_scan_offset;
617
618 /* numa_scan_seq prevents two threads setting pte_numa */
619 int numa_scan_seq;
620 #endif
621 /*
622 * An operation with batched TLB flushing is going on. Anything
623 * that can move process memory needs to flush the TLB when
624 * moving a PROT_NONE or PROT_NUMA mapped page.
625 */
626 atomic_t tlb_flush_pending;
627 #ifdef CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH
628 /* See flush_tlb_batched_pending() */
629 bool tlb_flush_batched;
630 #endif
631 struct uprobes_state uprobes_state;
632 #ifdef CONFIG_HUGETLB_PAGE
633 atomic_long_t hugetlb_usage;
634 #endif
635 struct work_struct async_put_work;
636
637 #ifdef CONFIG_IOMMU_SUPPORT
638 u32 pasid;
639 #endif
640 #ifdef CONFIG_LRU_GEN
641 struct {
642 /* this mm_struct is on lru_gen_mm_list */
643 struct list_head list;
644 /*
645 * Set when switching to this mm_struct, as a hint of
646 * whether it has been used since the last time per-node
647 * page table walkers cleared the corresponding bits.
648 */
649 unsigned long bitmap;
650 #ifdef CONFIG_MEMCG
651 /* points to the memcg of "owner" above */
652 struct mem_cgroup *memcg;
653 #endif
654 } lru_gen;
655 #endif /* CONFIG_LRU_GEN */
656
657 ANDROID_KABI_RESERVE(1);
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. */
mm_init_cpumask(struct mm_struct * mm)670 static inline void mm_init_cpumask(struct mm_struct *mm)
671 {
672 unsigned long cpu_bitmap = (unsigned long)mm;
673
674 cpu_bitmap += offsetof(struct mm_struct, cpu_bitmap);
675 cpumask_clear((struct cpumask *)cpu_bitmap);
676 }
677
678 /* Future-safe accessor for struct mm_struct's cpu_vm_mask. */
mm_cpumask(struct mm_struct * mm)679 static inline cpumask_t *mm_cpumask(struct mm_struct *mm)
680 {
681 return (struct cpumask *)&mm->cpu_bitmap;
682 }
683
684 #ifdef CONFIG_LRU_GEN
685
686 struct lru_gen_mm_list {
687 /* mm_struct list for page table walkers */
688 struct list_head fifo;
689 /* protects the list above */
690 spinlock_t lock;
691 };
692
693 void lru_gen_add_mm(struct mm_struct *mm);
694 void lru_gen_del_mm(struct mm_struct *mm);
695 #ifdef CONFIG_MEMCG
696 void lru_gen_migrate_mm(struct mm_struct *mm);
697 #endif
698
lru_gen_init_mm(struct mm_struct * mm)699 static inline void lru_gen_init_mm(struct mm_struct *mm)
700 {
701 INIT_LIST_HEAD(&mm->lru_gen.list);
702 mm->lru_gen.bitmap = 0;
703 #ifdef CONFIG_MEMCG
704 mm->lru_gen.memcg = NULL;
705 #endif
706 }
707
lru_gen_use_mm(struct mm_struct * mm)708 static inline void lru_gen_use_mm(struct mm_struct *mm)
709 {
710 /*
711 * When the bitmap is set, page reclaim knows this mm_struct has been
712 * used since the last time it cleared the bitmap. So it might be worth
713 * walking the page tables of this mm_struct to clear the accessed bit.
714 */
715 WRITE_ONCE(mm->lru_gen.bitmap, -1);
716 }
717
718 #else /* !CONFIG_LRU_GEN */
719
lru_gen_add_mm(struct mm_struct * mm)720 static inline void lru_gen_add_mm(struct mm_struct *mm)
721 {
722 }
723
lru_gen_del_mm(struct mm_struct * mm)724 static inline void lru_gen_del_mm(struct mm_struct *mm)
725 {
726 }
727
728 #ifdef CONFIG_MEMCG
lru_gen_migrate_mm(struct mm_struct * mm)729 static inline void lru_gen_migrate_mm(struct mm_struct *mm)
730 {
731 }
732 #endif
733
lru_gen_init_mm(struct mm_struct * mm)734 static inline void lru_gen_init_mm(struct mm_struct *mm)
735 {
736 }
737
lru_gen_use_mm(struct mm_struct * mm)738 static inline void lru_gen_use_mm(struct mm_struct *mm)
739 {
740 }
741
742 #endif /* CONFIG_LRU_GEN */
743
744 struct mmu_gather;
745 extern void tlb_gather_mmu(struct mmu_gather *tlb, struct mm_struct *mm);
746 extern void tlb_gather_mmu_fullmm(struct mmu_gather *tlb, struct mm_struct *mm);
747 extern void tlb_finish_mmu(struct mmu_gather *tlb);
748
init_tlb_flush_pending(struct mm_struct * mm)749 static inline void init_tlb_flush_pending(struct mm_struct *mm)
750 {
751 atomic_set(&mm->tlb_flush_pending, 0);
752 }
753
inc_tlb_flush_pending(struct mm_struct * mm)754 static inline void inc_tlb_flush_pending(struct mm_struct *mm)
755 {
756 atomic_inc(&mm->tlb_flush_pending);
757 /*
758 * The only time this value is relevant is when there are indeed pages
759 * to flush. And we'll only flush pages after changing them, which
760 * requires the PTL.
761 *
762 * So the ordering here is:
763 *
764 * atomic_inc(&mm->tlb_flush_pending);
765 * spin_lock(&ptl);
766 * ...
767 * set_pte_at();
768 * spin_unlock(&ptl);
769 *
770 * spin_lock(&ptl)
771 * mm_tlb_flush_pending();
772 * ....
773 * spin_unlock(&ptl);
774 *
775 * flush_tlb_range();
776 * atomic_dec(&mm->tlb_flush_pending);
777 *
778 * Where the increment if constrained by the PTL unlock, it thus
779 * ensures that the increment is visible if the PTE modification is
780 * visible. After all, if there is no PTE modification, nobody cares
781 * about TLB flushes either.
782 *
783 * This very much relies on users (mm_tlb_flush_pending() and
784 * mm_tlb_flush_nested()) only caring about _specific_ PTEs (and
785 * therefore specific PTLs), because with SPLIT_PTE_PTLOCKS and RCpc
786 * locks (PPC) the unlock of one doesn't order against the lock of
787 * another PTL.
788 *
789 * The decrement is ordered by the flush_tlb_range(), such that
790 * mm_tlb_flush_pending() will not return false unless all flushes have
791 * completed.
792 */
793 }
794
dec_tlb_flush_pending(struct mm_struct * mm)795 static inline void dec_tlb_flush_pending(struct mm_struct *mm)
796 {
797 /*
798 * See inc_tlb_flush_pending().
799 *
800 * This cannot be smp_mb__before_atomic() because smp_mb() simply does
801 * not order against TLB invalidate completion, which is what we need.
802 *
803 * Therefore we must rely on tlb_flush_*() to guarantee order.
804 */
805 atomic_dec(&mm->tlb_flush_pending);
806 }
807
mm_tlb_flush_pending(struct mm_struct * mm)808 static inline bool mm_tlb_flush_pending(struct mm_struct *mm)
809 {
810 /*
811 * Must be called after having acquired the PTL; orders against that
812 * PTLs release and therefore ensures that if we observe the modified
813 * PTE we must also observe the increment from inc_tlb_flush_pending().
814 *
815 * That is, it only guarantees to return true if there is a flush
816 * pending for _this_ PTL.
817 */
818 return atomic_read(&mm->tlb_flush_pending);
819 }
820
mm_tlb_flush_nested(struct mm_struct * mm)821 static inline bool mm_tlb_flush_nested(struct mm_struct *mm)
822 {
823 /*
824 * Similar to mm_tlb_flush_pending(), we must have acquired the PTL
825 * for which there is a TLB flush pending in order to guarantee
826 * we've seen both that PTE modification and the increment.
827 *
828 * (no requirement on actually still holding the PTL, that is irrelevant)
829 */
830 return atomic_read(&mm->tlb_flush_pending) > 1;
831 }
832
833 struct vm_fault;
834
835 /**
836 * typedef vm_fault_t - Return type for page fault handlers.
837 *
838 * Page fault handlers return a bitmask of %VM_FAULT values.
839 */
840 typedef __bitwise unsigned int vm_fault_t;
841
842 /**
843 * enum vm_fault_reason - Page fault handlers return a bitmask of
844 * these values to tell the core VM what happened when handling the
845 * fault. Used to decide whether a process gets delivered SIGBUS or
846 * just gets major/minor fault counters bumped up.
847 *
848 * @VM_FAULT_OOM: Out Of Memory
849 * @VM_FAULT_SIGBUS: Bad access
850 * @VM_FAULT_MAJOR: Page read from storage
851 * @VM_FAULT_WRITE: Special case for get_user_pages
852 * @VM_FAULT_HWPOISON: Hit poisoned small page
853 * @VM_FAULT_HWPOISON_LARGE: Hit poisoned large page. Index encoded
854 * in upper bits
855 * @VM_FAULT_SIGSEGV: segmentation fault
856 * @VM_FAULT_NOPAGE: ->fault installed the pte, not return page
857 * @VM_FAULT_LOCKED: ->fault locked the returned page
858 * @VM_FAULT_RETRY: ->fault blocked, must retry
859 * @VM_FAULT_FALLBACK: huge page fault failed, fall back to small
860 * @VM_FAULT_DONE_COW: ->fault has fully handled COW
861 * @VM_FAULT_NEEDDSYNC: ->fault did not modify page tables and needs
862 * fsync() to complete (for synchronous page faults
863 * in DAX)
864 * @VM_FAULT_HINDEX_MASK: mask HINDEX value
865 *
866 */
867 enum vm_fault_reason {
868 VM_FAULT_OOM = (__force vm_fault_t)0x000001,
869 VM_FAULT_SIGBUS = (__force vm_fault_t)0x000002,
870 VM_FAULT_MAJOR = (__force vm_fault_t)0x000004,
871 VM_FAULT_WRITE = (__force vm_fault_t)0x000008,
872 VM_FAULT_HWPOISON = (__force vm_fault_t)0x000010,
873 VM_FAULT_HWPOISON_LARGE = (__force vm_fault_t)0x000020,
874 VM_FAULT_SIGSEGV = (__force vm_fault_t)0x000040,
875 VM_FAULT_NOPAGE = (__force vm_fault_t)0x000100,
876 VM_FAULT_LOCKED = (__force vm_fault_t)0x000200,
877 VM_FAULT_RETRY = (__force vm_fault_t)0x000400,
878 VM_FAULT_FALLBACK = (__force vm_fault_t)0x000800,
879 VM_FAULT_DONE_COW = (__force vm_fault_t)0x001000,
880 VM_FAULT_NEEDDSYNC = (__force vm_fault_t)0x002000,
881 VM_FAULT_HINDEX_MASK = (__force vm_fault_t)0x0f0000,
882 };
883
884 /* Encode hstate index for a hwpoisoned large page */
885 #define VM_FAULT_SET_HINDEX(x) ((__force vm_fault_t)((x) << 16))
886 #define VM_FAULT_GET_HINDEX(x) (((__force unsigned int)(x) >> 16) & 0xf)
887
888 #define VM_FAULT_ERROR (VM_FAULT_OOM | VM_FAULT_SIGBUS | \
889 VM_FAULT_SIGSEGV | VM_FAULT_HWPOISON | \
890 VM_FAULT_HWPOISON_LARGE | VM_FAULT_FALLBACK)
891
892 #define VM_FAULT_RESULT_TRACE \
893 { VM_FAULT_OOM, "OOM" }, \
894 { VM_FAULT_SIGBUS, "SIGBUS" }, \
895 { VM_FAULT_MAJOR, "MAJOR" }, \
896 { VM_FAULT_WRITE, "WRITE" }, \
897 { VM_FAULT_HWPOISON, "HWPOISON" }, \
898 { VM_FAULT_HWPOISON_LARGE, "HWPOISON_LARGE" }, \
899 { VM_FAULT_SIGSEGV, "SIGSEGV" }, \
900 { VM_FAULT_NOPAGE, "NOPAGE" }, \
901 { VM_FAULT_LOCKED, "LOCKED" }, \
902 { VM_FAULT_RETRY, "RETRY" }, \
903 { VM_FAULT_FALLBACK, "FALLBACK" }, \
904 { VM_FAULT_DONE_COW, "DONE_COW" }, \
905 { VM_FAULT_NEEDDSYNC, "NEEDDSYNC" }
906
907 struct vm_special_mapping {
908 const char *name; /* The name, e.g. "[vdso]". */
909
910 /*
911 * If .fault is not provided, this points to a
912 * NULL-terminated array of pages that back the special mapping.
913 *
914 * This must not be NULL unless .fault is provided.
915 */
916 struct page **pages;
917
918 /*
919 * If non-NULL, then this is called to resolve page faults
920 * on the special mapping. If used, .pages is not checked.
921 */
922 vm_fault_t (*fault)(const struct vm_special_mapping *sm,
923 struct vm_area_struct *vma,
924 struct vm_fault *vmf);
925
926 int (*mremap)(const struct vm_special_mapping *sm,
927 struct vm_area_struct *new_vma);
928 };
929
930 enum tlb_flush_reason {
931 TLB_FLUSH_ON_TASK_SWITCH,
932 TLB_REMOTE_SHOOTDOWN,
933 TLB_LOCAL_SHOOTDOWN,
934 TLB_LOCAL_MM_SHOOTDOWN,
935 TLB_REMOTE_SEND_IPI,
936 NR_TLB_FLUSH_REASONS,
937 };
938
939 /*
940 * A swap entry has to fit into a "unsigned long", as the entry is hidden
941 * in the "index" field of the swapper address space.
942 */
943 typedef struct {
944 unsigned long val;
945 } swp_entry_t;
946
947 #endif /* _LINUX_MM_TYPES_H */
948