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