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