Branch data Line data Source code
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 : : };
138 : : struct { /* Second tail page of compound page */
139 : : unsigned long _compound_pad_1; /* compound_head */
140 : : unsigned long _compound_pad_2;
141 : : /* For both global and memcg */
142 : : struct list_head deferred_list;
143 : : };
144 : : struct { /* Page table pages */
145 : : unsigned long _pt_pad_1; /* compound_head */
146 : : pgtable_t pmd_huge_pte; /* protected by page->ptl */
147 : : unsigned long _pt_pad_2; /* mapping */
148 : : union {
149 : : struct mm_struct *pt_mm; /* x86 pgds only */
150 : : atomic_t pt_frag_refcount; /* powerpc */
151 : : };
152 : : #if ALLOC_SPLIT_PTLOCKS
153 : : spinlock_t *ptl;
154 : : #else
155 : : spinlock_t ptl;
156 : : #endif
157 : : };
158 : : struct { /* ZONE_DEVICE pages */
159 : : /** @pgmap: Points to the hosting device page map. */
160 : : struct dev_pagemap *pgmap;
161 : : void *zone_device_data;
162 : : /*
163 : : * ZONE_DEVICE private pages are counted as being
164 : : * mapped so the next 3 words hold the mapping, index,
165 : : * and private fields from the source anonymous or
166 : : * page cache page while the page is migrated to device
167 : : * private memory.
168 : : * ZONE_DEVICE MEMORY_DEVICE_FS_DAX pages also
169 : : * use the mapping, index, and private fields when
170 : : * pmem backed DAX files are mapped.
171 : : */
172 : : };
173 : :
174 : : /** @rcu_head: You can use this to free a page by RCU. */
175 : : struct rcu_head rcu_head;
176 : : };
177 : :
178 : : union { /* This union is 4 bytes in size. */
179 : : /*
180 : : * If the page can be mapped to userspace, encodes the number
181 : : * of times this page is referenced by a page table.
182 : : */
183 : : atomic_t _mapcount;
184 : :
185 : : /*
186 : : * If the page is neither PageSlab nor mappable to userspace,
187 : : * the value stored here may help determine what this page
188 : : * is used for. See page-flags.h for a list of page types
189 : : * which are currently stored here.
190 : : */
191 : : unsigned int page_type;
192 : :
193 : : unsigned int active; /* SLAB */
194 : : int units; /* SLOB */
195 : : };
196 : :
197 : : /* Usage count. *DO NOT USE DIRECTLY*. See page_ref.h */
198 : : atomic_t _refcount;
199 : :
200 : : #ifdef CONFIG_MEMCG
201 : : struct mem_cgroup *mem_cgroup;
202 : : #endif
203 : :
204 : : /*
205 : : * On machines where all RAM is mapped into kernel address space,
206 : : * we can simply calculate the virtual address. On machines with
207 : : * highmem some memory is mapped into kernel virtual memory
208 : : * dynamically, so we need a place to store that address.
209 : : * Note that this field could be 16 bits on x86 ... ;)
210 : : *
211 : : * Architectures with slow multiplication can define
212 : : * WANT_PAGE_VIRTUAL in asm/page.h
213 : : */
214 : : #if defined(WANT_PAGE_VIRTUAL)
215 : : void *virtual; /* Kernel virtual address (NULL if
216 : : not kmapped, ie. highmem) */
217 : : #endif /* WANT_PAGE_VIRTUAL */
218 : :
219 : : #ifdef LAST_CPUPID_NOT_IN_PAGE_FLAGS
220 : : int _last_cpupid;
221 : : #endif
222 : : } _struct_page_alignment;
223 : :
224 : : static inline atomic_t *compound_mapcount_ptr(struct page *page)
225 : : {
226 : 0 : return &page[1].compound_mapcount;
227 : : }
228 : :
229 : : /*
230 : : * Used for sizing the vmemmap region on some architectures
231 : : */
232 : : #define STRUCT_PAGE_MAX_SHIFT (order_base_2(sizeof(struct page)))
233 : :
234 : : #define PAGE_FRAG_CACHE_MAX_SIZE __ALIGN_MASK(32768, ~PAGE_MASK)
235 : : #define PAGE_FRAG_CACHE_MAX_ORDER get_order(PAGE_FRAG_CACHE_MAX_SIZE)
236 : :
237 : : #define page_private(page) ((page)->private)
238 : : #define set_page_private(page, v) ((page)->private = (v))
239 : :
240 : : struct page_frag_cache {
241 : : void * va;
242 : : #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
243 : : __u16 offset;
244 : : __u16 size;
245 : : #else
246 : : __u32 offset;
247 : : #endif
248 : : /* we maintain a pagecount bias, so that we dont dirty cache line
249 : : * containing page->_refcount every time we allocate a fragment.
250 : : */
251 : : unsigned int pagecnt_bias;
252 : : bool pfmemalloc;
253 : : };
254 : :
255 : : typedef unsigned long vm_flags_t;
256 : :
257 : : /*
258 : : * A region containing a mapping of a non-memory backed file under NOMMU
259 : : * conditions. These are held in a global tree and are pinned by the VMAs that
260 : : * map parts of them.
261 : : */
262 : : struct vm_region {
263 : : struct rb_node vm_rb; /* link in global region tree */
264 : : vm_flags_t vm_flags; /* VMA vm_flags */
265 : : unsigned long vm_start; /* start address of region */
266 : : unsigned long vm_end; /* region initialised to here */
267 : : unsigned long vm_top; /* region allocated to here */
268 : : unsigned long vm_pgoff; /* the offset in vm_file corresponding to vm_start */
269 : : struct file *vm_file; /* the backing file or NULL */
270 : :
271 : : int vm_usage; /* region usage count (access under nommu_region_sem) */
272 : : bool vm_icache_flushed : 1; /* true if the icache has been flushed for
273 : : * this region */
274 : : };
275 : :
276 : : #ifdef CONFIG_USERFAULTFD
277 : : #define NULL_VM_UFFD_CTX ((struct vm_userfaultfd_ctx) { NULL, })
278 : : struct vm_userfaultfd_ctx {
279 : : struct userfaultfd_ctx *ctx;
280 : : };
281 : : #else /* CONFIG_USERFAULTFD */
282 : : #define NULL_VM_UFFD_CTX ((struct vm_userfaultfd_ctx) {})
283 : : struct vm_userfaultfd_ctx {};
284 : : #endif /* CONFIG_USERFAULTFD */
285 : :
286 : : /*
287 : : * This struct defines a memory VMM memory area. There is one of these
288 : : * per VM-area/task. A VM area is any part of the process virtual memory
289 : : * space that has a special rule for the page-fault handlers (ie a shared
290 : : * library, the executable area etc).
291 : : */
292 : : struct vm_area_struct {
293 : : /* The first cache line has the info for VMA tree walking. */
294 : :
295 : : unsigned long vm_start; /* Our start address within vm_mm. */
296 : : unsigned long vm_end; /* The first byte after our end address
297 : : within vm_mm. */
298 : :
299 : : /* linked list of VM areas per task, sorted by address */
300 : : struct vm_area_struct *vm_next, *vm_prev;
301 : :
302 : : struct rb_node vm_rb;
303 : :
304 : : /*
305 : : * Largest free memory gap in bytes to the left of this VMA.
306 : : * Either between this VMA and vma->vm_prev, or between one of the
307 : : * VMAs below us in the VMA rbtree and its ->vm_prev. This helps
308 : : * get_unmapped_area find a free area of the right size.
309 : : */
310 : : unsigned long rb_subtree_gap;
311 : :
312 : : /* Second cache line starts here. */
313 : :
314 : : struct mm_struct *vm_mm; /* The address space we belong to. */
315 : : pgprot_t vm_page_prot; /* Access permissions of this VMA. */
316 : : unsigned long vm_flags; /* Flags, see mm.h. */
317 : :
318 : : /*
319 : : * For areas with an address space and backing store,
320 : : * linkage into the address_space->i_mmap interval tree.
321 : : */
322 : : struct {
323 : : struct rb_node rb;
324 : : unsigned long rb_subtree_last;
325 : : } shared;
326 : :
327 : : /*
328 : : * A file's MAP_PRIVATE vma can be in both i_mmap tree and anon_vma
329 : : * list, after a COW of one of the file pages. A MAP_SHARED vma
330 : : * can only be in the i_mmap tree. An anonymous MAP_PRIVATE, stack
331 : : * or brk vma (with NULL file) can only be in an anon_vma list.
332 : : */
333 : : struct list_head anon_vma_chain; /* Serialized by mmap_sem &
334 : : * page_table_lock */
335 : : struct anon_vma *anon_vma; /* Serialized by page_table_lock */
336 : :
337 : : /* Function pointers to deal with this struct. */
338 : : const struct vm_operations_struct *vm_ops;
339 : :
340 : : /* Information about our backing store: */
341 : : unsigned long vm_pgoff; /* Offset (within vm_file) in PAGE_SIZE
342 : : units */
343 : : struct file * vm_file; /* File we map to (can be NULL). */
344 : : void * vm_private_data; /* was vm_pte (shared mem) */
345 : :
346 : : #ifdef CONFIG_SWAP
347 : : atomic_long_t swap_readahead_info;
348 : : #endif
349 : : #ifndef CONFIG_MMU
350 : : struct vm_region *vm_region; /* NOMMU mapping region */
351 : : #endif
352 : : #ifdef CONFIG_NUMA
353 : : struct mempolicy *vm_policy; /* NUMA policy for the VMA */
354 : : #endif
355 : : struct vm_userfaultfd_ctx vm_userfaultfd_ctx;
356 : : } __randomize_layout;
357 : :
358 : : struct core_thread {
359 : : struct task_struct *task;
360 : : struct core_thread *next;
361 : : };
362 : :
363 : : struct core_state {
364 : : atomic_t nr_threads;
365 : : struct core_thread dumper;
366 : : struct completion startup;
367 : : };
368 : :
369 : : struct kioctx_table;
370 : : struct mm_struct {
371 : : struct {
372 : : struct vm_area_struct *mmap; /* list of VMAs */
373 : : struct rb_root mm_rb;
374 : : u64 vmacache_seqnum; /* per-thread vmacache */
375 : : #ifdef CONFIG_MMU
376 : : unsigned long (*get_unmapped_area) (struct file *filp,
377 : : unsigned long addr, unsigned long len,
378 : : unsigned long pgoff, unsigned long flags);
379 : : #endif
380 : : unsigned long mmap_base; /* base of mmap area */
381 : : unsigned long mmap_legacy_base; /* base of mmap area in bottom-up allocations */
382 : : #ifdef CONFIG_HAVE_ARCH_COMPAT_MMAP_BASES
383 : : /* Base adresses for compatible mmap() */
384 : : unsigned long mmap_compat_base;
385 : : unsigned long mmap_compat_legacy_base;
386 : : #endif
387 : : unsigned long task_size; /* size of task vm space */
388 : : unsigned long highest_vm_end; /* highest vma end address */
389 : : pgd_t * pgd;
390 : :
391 : : #ifdef CONFIG_MEMBARRIER
392 : : /**
393 : : * @membarrier_state: Flags controlling membarrier behavior.
394 : : *
395 : : * This field is close to @pgd to hopefully fit in the same
396 : : * cache-line, which needs to be touched by switch_mm().
397 : : */
398 : : atomic_t membarrier_state;
399 : : #endif
400 : :
401 : : /**
402 : : * @mm_users: The number of users including userspace.
403 : : *
404 : : * Use mmget()/mmget_not_zero()/mmput() to modify. When this
405 : : * drops to 0 (i.e. when the task exits and there are no other
406 : : * temporary reference holders), we also release a reference on
407 : : * @mm_count (which may then free the &struct mm_struct if
408 : : * @mm_count also drops to 0).
409 : : */
410 : : atomic_t mm_users;
411 : :
412 : : /**
413 : : * @mm_count: The number of references to &struct mm_struct
414 : : * (@mm_users count as 1).
415 : : *
416 : : * Use mmgrab()/mmdrop() to modify. When this drops to 0, the
417 : : * &struct mm_struct is freed.
418 : : */
419 : : atomic_t mm_count;
420 : :
421 : : #ifdef CONFIG_MMU
422 : : atomic_long_t pgtables_bytes; /* PTE page table pages */
423 : : #endif
424 : : int map_count; /* number of VMAs */
425 : :
426 : : spinlock_t page_table_lock; /* Protects page tables and some
427 : : * counters
428 : : */
429 : : struct rw_semaphore mmap_sem;
430 : :
431 : : struct list_head mmlist; /* List of maybe swapped mm's. These
432 : : * are globally strung together off
433 : : * init_mm.mmlist, and are protected
434 : : * by mmlist_lock
435 : : */
436 : :
437 : :
438 : : unsigned long hiwater_rss; /* High-watermark of RSS usage */
439 : : unsigned long hiwater_vm; /* High-water virtual memory usage */
440 : :
441 : : unsigned long total_vm; /* Total pages mapped */
442 : : unsigned long locked_vm; /* Pages that have PG_mlocked set */
443 : : atomic64_t pinned_vm; /* Refcount permanently increased */
444 : : unsigned long data_vm; /* VM_WRITE & ~VM_SHARED & ~VM_STACK */
445 : : unsigned long exec_vm; /* VM_EXEC & ~VM_WRITE & ~VM_STACK */
446 : : unsigned long stack_vm; /* VM_STACK */
447 : : unsigned long def_flags;
448 : :
449 : : spinlock_t arg_lock; /* protect the below fields */
450 : : unsigned long start_code, end_code, start_data, end_data;
451 : : unsigned long start_brk, brk, start_stack;
452 : : unsigned long arg_start, arg_end, env_start, env_end;
453 : :
454 : : unsigned long saved_auxv[AT_VECTOR_SIZE]; /* for /proc/PID/auxv */
455 : :
456 : : /*
457 : : * Special counters, in some configurations protected by the
458 : : * page_table_lock, in other configurations by being atomic.
459 : : */
460 : : struct mm_rss_stat rss_stat;
461 : :
462 : : struct linux_binfmt *binfmt;
463 : :
464 : : /* Architecture-specific MM context */
465 : : mm_context_t context;
466 : :
467 : : unsigned long flags; /* Must use atomic bitops to access */
468 : :
469 : : struct core_state *core_state; /* coredumping support */
470 : :
471 : : #ifdef CONFIG_AIO
472 : : spinlock_t ioctx_lock;
473 : : struct kioctx_table __rcu *ioctx_table;
474 : : #endif
475 : : #ifdef CONFIG_MEMCG
476 : : /*
477 : : * "owner" points to a task that is regarded as the canonical
478 : : * user/owner of this mm. All of the following must be true in
479 : : * order for it to be changed:
480 : : *
481 : : * current == mm->owner
482 : : * current->mm != mm
483 : : * new_owner->mm == mm
484 : : * new_owner->alloc_lock is held
485 : : */
486 : : struct task_struct __rcu *owner;
487 : : #endif
488 : : struct user_namespace *user_ns;
489 : :
490 : : /* store ref to file /proc/<pid>/exe symlink points to */
491 : : struct file __rcu *exe_file;
492 : : #ifdef CONFIG_MMU_NOTIFIER
493 : : struct mmu_notifier_mm *mmu_notifier_mm;
494 : : #endif
495 : : #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
496 : : pgtable_t pmd_huge_pte; /* protected by page_table_lock */
497 : : #endif
498 : : #ifdef CONFIG_NUMA_BALANCING
499 : : /*
500 : : * numa_next_scan is the next time that the PTEs will be marked
501 : : * pte_numa. NUMA hinting faults will gather statistics and
502 : : * migrate pages to new nodes if necessary.
503 : : */
504 : : unsigned long numa_next_scan;
505 : :
506 : : /* Restart point for scanning and setting pte_numa */
507 : : unsigned long numa_scan_offset;
508 : :
509 : : /* numa_scan_seq prevents two threads setting pte_numa */
510 : : int numa_scan_seq;
511 : : #endif
512 : : /*
513 : : * An operation with batched TLB flushing is going on. Anything
514 : : * that can move process memory needs to flush the TLB when
515 : : * moving a PROT_NONE or PROT_NUMA mapped page.
516 : : */
517 : : atomic_t tlb_flush_pending;
518 : : #ifdef CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH
519 : : /* See flush_tlb_batched_pending() */
520 : : bool tlb_flush_batched;
521 : : #endif
522 : : struct uprobes_state uprobes_state;
523 : : #ifdef CONFIG_HUGETLB_PAGE
524 : : atomic_long_t hugetlb_usage;
525 : : #endif
526 : : struct work_struct async_put_work;
527 : : } __randomize_layout;
528 : :
529 : : /*
530 : : * The mm_cpumask needs to be at the end of mm_struct, because it
531 : : * is dynamically sized based on nr_cpu_ids.
532 : : */
533 : : unsigned long cpu_bitmap[];
534 : : };
535 : :
536 : : extern struct mm_struct init_mm;
537 : :
538 : : /* Pointer magic because the dynamic array size confuses some compilers. */
539 : : static inline void mm_init_cpumask(struct mm_struct *mm)
540 : : {
541 : 3 : unsigned long cpu_bitmap = (unsigned long)mm;
542 : :
543 : 3 : cpu_bitmap += offsetof(struct mm_struct, cpu_bitmap);
544 : 3 : cpumask_clear((struct cpumask *)cpu_bitmap);
545 : : }
546 : :
547 : : /* Future-safe accessor for struct mm_struct's cpu_vm_mask. */
548 : : static inline cpumask_t *mm_cpumask(struct mm_struct *mm)
549 : : {
550 : 0 : return (struct cpumask *)&mm->cpu_bitmap;
551 : : }
552 : :
553 : : struct mmu_gather;
554 : : extern void tlb_gather_mmu(struct mmu_gather *tlb, struct mm_struct *mm,
555 : : unsigned long start, unsigned long end);
556 : : extern void tlb_finish_mmu(struct mmu_gather *tlb,
557 : : unsigned long start, unsigned long end);
558 : :
559 : : static inline void init_tlb_flush_pending(struct mm_struct *mm)
560 : : {
561 : : atomic_set(&mm->tlb_flush_pending, 0);
562 : : }
563 : :
564 : : static inline void inc_tlb_flush_pending(struct mm_struct *mm)
565 : : {
566 : 3 : atomic_inc(&mm->tlb_flush_pending);
567 : : /*
568 : : * The only time this value is relevant is when there are indeed pages
569 : : * to flush. And we'll only flush pages after changing them, which
570 : : * requires the PTL.
571 : : *
572 : : * So the ordering here is:
573 : : *
574 : : * atomic_inc(&mm->tlb_flush_pending);
575 : : * spin_lock(&ptl);
576 : : * ...
577 : : * set_pte_at();
578 : : * spin_unlock(&ptl);
579 : : *
580 : : * spin_lock(&ptl)
581 : : * mm_tlb_flush_pending();
582 : : * ....
583 : : * spin_unlock(&ptl);
584 : : *
585 : : * flush_tlb_range();
586 : : * atomic_dec(&mm->tlb_flush_pending);
587 : : *
588 : : * Where the increment if constrained by the PTL unlock, it thus
589 : : * ensures that the increment is visible if the PTE modification is
590 : : * visible. After all, if there is no PTE modification, nobody cares
591 : : * about TLB flushes either.
592 : : *
593 : : * This very much relies on users (mm_tlb_flush_pending() and
594 : : * mm_tlb_flush_nested()) only caring about _specific_ PTEs (and
595 : : * therefore specific PTLs), because with SPLIT_PTE_PTLOCKS and RCpc
596 : : * locks (PPC) the unlock of one doesn't order against the lock of
597 : : * another PTL.
598 : : *
599 : : * The decrement is ordered by the flush_tlb_range(), such that
600 : : * mm_tlb_flush_pending() will not return false unless all flushes have
601 : : * completed.
602 : : */
603 : : }
604 : :
605 : : static inline void dec_tlb_flush_pending(struct mm_struct *mm)
606 : : {
607 : : /*
608 : : * See inc_tlb_flush_pending().
609 : : *
610 : : * This cannot be smp_mb__before_atomic() because smp_mb() simply does
611 : : * not order against TLB invalidate completion, which is what we need.
612 : : *
613 : : * Therefore we must rely on tlb_flush_*() to guarantee order.
614 : : */
615 : 3 : atomic_dec(&mm->tlb_flush_pending);
616 : : }
617 : :
618 : : static inline bool mm_tlb_flush_pending(struct mm_struct *mm)
619 : : {
620 : : /*
621 : : * Must be called after having acquired the PTL; orders against that
622 : : * PTLs release and therefore ensures that if we observe the modified
623 : : * PTE we must also observe the increment from inc_tlb_flush_pending().
624 : : *
625 : : * That is, it only guarantees to return true if there is a flush
626 : : * pending for _this_ PTL.
627 : : */
628 : : return atomic_read(&mm->tlb_flush_pending);
629 : : }
630 : :
631 : : static inline bool mm_tlb_flush_nested(struct mm_struct *mm)
632 : : {
633 : : /*
634 : : * Similar to mm_tlb_flush_pending(), we must have acquired the PTL
635 : : * for which there is a TLB flush pending in order to guarantee
636 : : * we've seen both that PTE modification and the increment.
637 : : *
638 : : * (no requirement on actually still holding the PTL, that is irrelevant)
639 : : */
640 : 3 : return atomic_read(&mm->tlb_flush_pending) > 1;
641 : : }
642 : :
643 : : struct vm_fault;
644 : :
645 : : /**
646 : : * typedef vm_fault_t - Return type for page fault handlers.
647 : : *
648 : : * Page fault handlers return a bitmask of %VM_FAULT values.
649 : : */
650 : : typedef __bitwise unsigned int vm_fault_t;
651 : :
652 : : /**
653 : : * enum vm_fault_reason - Page fault handlers return a bitmask of
654 : : * these values to tell the core VM what happened when handling the
655 : : * fault. Used to decide whether a process gets delivered SIGBUS or
656 : : * just gets major/minor fault counters bumped up.
657 : : *
658 : : * @VM_FAULT_OOM: Out Of Memory
659 : : * @VM_FAULT_SIGBUS: Bad access
660 : : * @VM_FAULT_MAJOR: Page read from storage
661 : : * @VM_FAULT_WRITE: Special case for get_user_pages
662 : : * @VM_FAULT_HWPOISON: Hit poisoned small page
663 : : * @VM_FAULT_HWPOISON_LARGE: Hit poisoned large page. Index encoded
664 : : * in upper bits
665 : : * @VM_FAULT_SIGSEGV: segmentation fault
666 : : * @VM_FAULT_NOPAGE: ->fault installed the pte, not return page
667 : : * @VM_FAULT_LOCKED: ->fault locked the returned page
668 : : * @VM_FAULT_RETRY: ->fault blocked, must retry
669 : : * @VM_FAULT_FALLBACK: huge page fault failed, fall back to small
670 : : * @VM_FAULT_DONE_COW: ->fault has fully handled COW
671 : : * @VM_FAULT_NEEDDSYNC: ->fault did not modify page tables and needs
672 : : * fsync() to complete (for synchronous page faults
673 : : * in DAX)
674 : : * @VM_FAULT_HINDEX_MASK: mask HINDEX value
675 : : *
676 : : */
677 : : enum vm_fault_reason {
678 : : VM_FAULT_OOM = (__force vm_fault_t)0x000001,
679 : : VM_FAULT_SIGBUS = (__force vm_fault_t)0x000002,
680 : : VM_FAULT_MAJOR = (__force vm_fault_t)0x000004,
681 : : VM_FAULT_WRITE = (__force vm_fault_t)0x000008,
682 : : VM_FAULT_HWPOISON = (__force vm_fault_t)0x000010,
683 : : VM_FAULT_HWPOISON_LARGE = (__force vm_fault_t)0x000020,
684 : : VM_FAULT_SIGSEGV = (__force vm_fault_t)0x000040,
685 : : VM_FAULT_NOPAGE = (__force vm_fault_t)0x000100,
686 : : VM_FAULT_LOCKED = (__force vm_fault_t)0x000200,
687 : : VM_FAULT_RETRY = (__force vm_fault_t)0x000400,
688 : : VM_FAULT_FALLBACK = (__force vm_fault_t)0x000800,
689 : : VM_FAULT_DONE_COW = (__force vm_fault_t)0x001000,
690 : : VM_FAULT_NEEDDSYNC = (__force vm_fault_t)0x002000,
691 : : VM_FAULT_HINDEX_MASK = (__force vm_fault_t)0x0f0000,
692 : : };
693 : :
694 : : /* Encode hstate index for a hwpoisoned large page */
695 : : #define VM_FAULT_SET_HINDEX(x) ((__force vm_fault_t)((x) << 16))
696 : : #define VM_FAULT_GET_HINDEX(x) (((__force unsigned int)(x) >> 16) & 0xf)
697 : :
698 : : #define VM_FAULT_ERROR (VM_FAULT_OOM | VM_FAULT_SIGBUS | \
699 : : VM_FAULT_SIGSEGV | VM_FAULT_HWPOISON | \
700 : : VM_FAULT_HWPOISON_LARGE | VM_FAULT_FALLBACK)
701 : :
702 : : #define VM_FAULT_RESULT_TRACE \
703 : : { VM_FAULT_OOM, "OOM" }, \
704 : : { VM_FAULT_SIGBUS, "SIGBUS" }, \
705 : : { VM_FAULT_MAJOR, "MAJOR" }, \
706 : : { VM_FAULT_WRITE, "WRITE" }, \
707 : : { VM_FAULT_HWPOISON, "HWPOISON" }, \
708 : : { VM_FAULT_HWPOISON_LARGE, "HWPOISON_LARGE" }, \
709 : : { VM_FAULT_SIGSEGV, "SIGSEGV" }, \
710 : : { VM_FAULT_NOPAGE, "NOPAGE" }, \
711 : : { VM_FAULT_LOCKED, "LOCKED" }, \
712 : : { VM_FAULT_RETRY, "RETRY" }, \
713 : : { VM_FAULT_FALLBACK, "FALLBACK" }, \
714 : : { VM_FAULT_DONE_COW, "DONE_COW" }, \
715 : : { VM_FAULT_NEEDDSYNC, "NEEDDSYNC" }
716 : :
717 : : struct vm_special_mapping {
718 : : const char *name; /* The name, e.g. "[vdso]". */
719 : :
720 : : /*
721 : : * If .fault is not provided, this points to a
722 : : * NULL-terminated array of pages that back the special mapping.
723 : : *
724 : : * This must not be NULL unless .fault is provided.
725 : : */
726 : : struct page **pages;
727 : :
728 : : /*
729 : : * If non-NULL, then this is called to resolve page faults
730 : : * on the special mapping. If used, .pages is not checked.
731 : : */
732 : : vm_fault_t (*fault)(const struct vm_special_mapping *sm,
733 : : struct vm_area_struct *vma,
734 : : struct vm_fault *vmf);
735 : :
736 : : int (*mremap)(const struct vm_special_mapping *sm,
737 : : struct vm_area_struct *new_vma);
738 : : };
739 : :
740 : : enum tlb_flush_reason {
741 : : TLB_FLUSH_ON_TASK_SWITCH,
742 : : TLB_REMOTE_SHOOTDOWN,
743 : : TLB_LOCAL_SHOOTDOWN,
744 : : TLB_LOCAL_MM_SHOOTDOWN,
745 : : TLB_REMOTE_SEND_IPI,
746 : : NR_TLB_FLUSH_REASONS,
747 : : };
748 : :
749 : : /*
750 : : * A swap entry has to fit into a "unsigned long", as the entry is hidden
751 : : * in the "index" field of the swapper address space.
752 : : */
753 : : typedef struct {
754 : : unsigned long val;
755 : : } swp_entry_t;
756 : :
757 : : #endif /* _LINUX_MM_TYPES_H */
|