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