]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blob - include/linux/mm_types.h
Merge branch 'next-general' of git://git.kernel.org/pub/scm/linux/kernel/git/jmorris...
[mirror_ubuntu-bionic-kernel.git] / include / linux / mm_types.h
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 struct address_space;
26 struct mem_cgroup;
27 struct hmm;
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 * The objects in struct page are organized in double word blocks in
37 * order to allows us to use atomic double word operations on portions
38 * of struct page. That is currently only used by slub but the arrangement
39 * allows the use of atomic double word operations on the flags/mapping
40 * and lru list pointers also.
41 */
42 struct page {
43 /* First double word block */
44 unsigned long flags; /* Atomic flags, some possibly
45 * updated asynchronously */
46 union {
47 struct address_space *mapping; /* If low bit clear, points to
48 * inode address_space, or NULL.
49 * If page mapped as anonymous
50 * memory, low bit is set, and
51 * it points to anon_vma object:
52 * see PAGE_MAPPING_ANON below.
53 */
54 void *s_mem; /* slab first object */
55 atomic_t compound_mapcount; /* first tail page */
56 /* page_deferred_list().next -- second tail page */
57 };
58
59 /* Second double word */
60 union {
61 pgoff_t index; /* Our offset within mapping. */
62 void *freelist; /* sl[aou]b first free object */
63 /* page_deferred_list().prev -- second tail page */
64 };
65
66 union {
67 #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \
68 defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
69 /* Used for cmpxchg_double in slub */
70 unsigned long counters;
71 #else
72 /*
73 * Keep _refcount separate from slub cmpxchg_double data.
74 * As the rest of the double word is protected by slab_lock
75 * but _refcount is not.
76 */
77 unsigned counters;
78 #endif
79 struct {
80
81 union {
82 /*
83 * Count of ptes mapped in mms, to show when
84 * page is mapped & limit reverse map searches.
85 *
86 * Extra information about page type may be
87 * stored here for pages that are never mapped,
88 * in which case the value MUST BE <= -2.
89 * See page-flags.h for more details.
90 */
91 atomic_t _mapcount;
92
93 unsigned int active; /* SLAB */
94 struct { /* SLUB */
95 unsigned inuse:16;
96 unsigned objects:15;
97 unsigned frozen:1;
98 };
99 int units; /* SLOB */
100 };
101 /*
102 * Usage count, *USE WRAPPER FUNCTION* when manual
103 * accounting. See page_ref.h
104 */
105 atomic_t _refcount;
106 };
107 };
108
109 /*
110 * Third double word block
111 *
112 * WARNING: bit 0 of the first word encode PageTail(). That means
113 * the rest users of the storage space MUST NOT use the bit to
114 * avoid collision and false-positive PageTail().
115 */
116 union {
117 struct list_head lru; /* Pageout list, eg. active_list
118 * protected by zone_lru_lock !
119 * Can be used as a generic list
120 * by the page owner.
121 */
122 struct dev_pagemap *pgmap; /* ZONE_DEVICE pages are never on an
123 * lru or handled by a slab
124 * allocator, this points to the
125 * hosting device page map.
126 */
127 struct { /* slub per cpu partial pages */
128 struct page *next; /* Next partial slab */
129 #ifdef CONFIG_64BIT
130 int pages; /* Nr of partial slabs left */
131 int pobjects; /* Approximate # of objects */
132 #else
133 short int pages;
134 short int pobjects;
135 #endif
136 };
137
138 struct rcu_head rcu_head; /* Used by SLAB
139 * when destroying via RCU
140 */
141 /* Tail pages of compound page */
142 struct {
143 unsigned long compound_head; /* If bit zero is set */
144
145 /* First tail page only */
146 #ifdef CONFIG_64BIT
147 /*
148 * On 64 bit system we have enough space in struct page
149 * to encode compound_dtor and compound_order with
150 * unsigned int. It can help compiler generate better or
151 * smaller code on some archtectures.
152 */
153 unsigned int compound_dtor;
154 unsigned int compound_order;
155 #else
156 unsigned short int compound_dtor;
157 unsigned short int compound_order;
158 #endif
159 };
160
161 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && USE_SPLIT_PMD_PTLOCKS
162 struct {
163 unsigned long __pad; /* do not overlay pmd_huge_pte
164 * with compound_head to avoid
165 * possible bit 0 collision.
166 */
167 pgtable_t pmd_huge_pte; /* protected by page->ptl */
168 };
169 #endif
170 };
171
172 /* Remainder is not double word aligned */
173 union {
174 unsigned long private; /* Mapping-private opaque data:
175 * usually used for buffer_heads
176 * if PagePrivate set; used for
177 * swp_entry_t if PageSwapCache;
178 * indicates order in the buddy
179 * system if PG_buddy is set.
180 */
181 #if USE_SPLIT_PTE_PTLOCKS
182 #if ALLOC_SPLIT_PTLOCKS
183 spinlock_t *ptl;
184 #else
185 spinlock_t ptl;
186 #endif
187 #endif
188 struct kmem_cache *slab_cache; /* SL[AU]B: Pointer to slab */
189 };
190
191 #ifdef CONFIG_MEMCG
192 struct mem_cgroup *mem_cgroup;
193 #endif
194
195 /*
196 * On machines where all RAM is mapped into kernel address space,
197 * we can simply calculate the virtual address. On machines with
198 * highmem some memory is mapped into kernel virtual memory
199 * dynamically, so we need a place to store that address.
200 * Note that this field could be 16 bits on x86 ... ;)
201 *
202 * Architectures with slow multiplication can define
203 * WANT_PAGE_VIRTUAL in asm/page.h
204 */
205 #if defined(WANT_PAGE_VIRTUAL)
206 void *virtual; /* Kernel virtual address (NULL if
207 not kmapped, ie. highmem) */
208 #endif /* WANT_PAGE_VIRTUAL */
209
210 #ifdef CONFIG_KMEMCHECK
211 /*
212 * kmemcheck wants to track the status of each byte in a page; this
213 * is a pointer to such a status block. NULL if not tracked.
214 */
215 void *shadow;
216 #endif
217
218 #ifdef LAST_CPUPID_NOT_IN_PAGE_FLAGS
219 int _last_cpupid;
220 #endif
221 }
222 /*
223 * The struct page can be forced to be double word aligned so that atomic ops
224 * on double words work. The SLUB allocator can make use of such a feature.
225 */
226 #ifdef CONFIG_HAVE_ALIGNED_STRUCT_PAGE
227 __aligned(2 * sizeof(unsigned long))
228 #endif
229 ;
230
231 #define PAGE_FRAG_CACHE_MAX_SIZE __ALIGN_MASK(32768, ~PAGE_MASK)
232 #define PAGE_FRAG_CACHE_MAX_ORDER get_order(PAGE_FRAG_CACHE_MAX_SIZE)
233
234 struct page_frag_cache {
235 void * va;
236 #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
237 __u16 offset;
238 __u16 size;
239 #else
240 __u32 offset;
241 #endif
242 /* we maintain a pagecount bias, so that we dont dirty cache line
243 * containing page->_refcount every time we allocate a fragment.
244 */
245 unsigned int pagecnt_bias;
246 bool pfmemalloc;
247 };
248
249 typedef unsigned long vm_flags_t;
250
251 /*
252 * A region containing a mapping of a non-memory backed file under NOMMU
253 * conditions. These are held in a global tree and are pinned by the VMAs that
254 * map parts of them.
255 */
256 struct vm_region {
257 struct rb_node vm_rb; /* link in global region tree */
258 vm_flags_t vm_flags; /* VMA vm_flags */
259 unsigned long vm_start; /* start address of region */
260 unsigned long vm_end; /* region initialised to here */
261 unsigned long vm_top; /* region allocated to here */
262 unsigned long vm_pgoff; /* the offset in vm_file corresponding to vm_start */
263 struct file *vm_file; /* the backing file or NULL */
264
265 int vm_usage; /* region usage count (access under nommu_region_sem) */
266 bool vm_icache_flushed : 1; /* true if the icache has been flushed for
267 * this region */
268 };
269
270 #ifdef CONFIG_USERFAULTFD
271 #define NULL_VM_UFFD_CTX ((struct vm_userfaultfd_ctx) { NULL, })
272 struct vm_userfaultfd_ctx {
273 struct userfaultfd_ctx *ctx;
274 };
275 #else /* CONFIG_USERFAULTFD */
276 #define NULL_VM_UFFD_CTX ((struct vm_userfaultfd_ctx) {})
277 struct vm_userfaultfd_ctx {};
278 #endif /* CONFIG_USERFAULTFD */
279
280 /*
281 * This struct defines a memory VMM memory area. There is one of these
282 * per VM-area/task. A VM area is any part of the process virtual memory
283 * space that has a special rule for the page-fault handlers (ie a shared
284 * library, the executable area etc).
285 */
286 struct vm_area_struct {
287 /* The first cache line has the info for VMA tree walking. */
288
289 unsigned long vm_start; /* Our start address within vm_mm. */
290 unsigned long vm_end; /* The first byte after our end address
291 within vm_mm. */
292
293 /* linked list of VM areas per task, sorted by address */
294 struct vm_area_struct *vm_next, *vm_prev;
295
296 struct rb_node vm_rb;
297
298 /*
299 * Largest free memory gap in bytes to the left of this VMA.
300 * Either between this VMA and vma->vm_prev, or between one of the
301 * VMAs below us in the VMA rbtree and its ->vm_prev. This helps
302 * get_unmapped_area find a free area of the right size.
303 */
304 unsigned long rb_subtree_gap;
305
306 /* Second cache line starts here. */
307
308 struct mm_struct *vm_mm; /* The address space we belong to. */
309 pgprot_t vm_page_prot; /* Access permissions of this VMA. */
310 unsigned long vm_flags; /* Flags, see mm.h. */
311
312 /*
313 * For areas with an address space and backing store,
314 * linkage into the address_space->i_mmap interval tree.
315 */
316 struct {
317 struct rb_node rb;
318 unsigned long rb_subtree_last;
319 } shared;
320
321 /*
322 * A file's MAP_PRIVATE vma can be in both i_mmap tree and anon_vma
323 * list, after a COW of one of the file pages. A MAP_SHARED vma
324 * can only be in the i_mmap tree. An anonymous MAP_PRIVATE, stack
325 * or brk vma (with NULL file) can only be in an anon_vma list.
326 */
327 struct list_head anon_vma_chain; /* Serialized by mmap_sem &
328 * page_table_lock */
329 struct anon_vma *anon_vma; /* Serialized by page_table_lock */
330
331 /* Function pointers to deal with this struct. */
332 const struct vm_operations_struct *vm_ops;
333
334 /* Information about our backing store: */
335 unsigned long vm_pgoff; /* Offset (within vm_file) in PAGE_SIZE
336 units */
337 struct file * vm_file; /* File we map to (can be NULL). */
338 void * vm_private_data; /* was vm_pte (shared mem) */
339
340 atomic_long_t swap_readahead_info;
341 #ifndef CONFIG_MMU
342 struct vm_region *vm_region; /* NOMMU mapping region */
343 #endif
344 #ifdef CONFIG_NUMA
345 struct mempolicy *vm_policy; /* NUMA policy for the VMA */
346 #endif
347 struct vm_userfaultfd_ctx vm_userfaultfd_ctx;
348 } __randomize_layout;
349
350 struct core_thread {
351 struct task_struct *task;
352 struct core_thread *next;
353 };
354
355 struct core_state {
356 atomic_t nr_threads;
357 struct core_thread dumper;
358 struct completion startup;
359 };
360
361 struct kioctx_table;
362 struct mm_struct {
363 struct vm_area_struct *mmap; /* list of VMAs */
364 struct rb_root mm_rb;
365 u32 vmacache_seqnum; /* per-thread vmacache */
366 #ifdef CONFIG_MMU
367 unsigned long (*get_unmapped_area) (struct file *filp,
368 unsigned long addr, unsigned long len,
369 unsigned long pgoff, unsigned long flags);
370 #endif
371 unsigned long mmap_base; /* base of mmap area */
372 unsigned long mmap_legacy_base; /* base of mmap area in bottom-up allocations */
373 #ifdef CONFIG_HAVE_ARCH_COMPAT_MMAP_BASES
374 /* Base adresses for compatible mmap() */
375 unsigned long mmap_compat_base;
376 unsigned long mmap_compat_legacy_base;
377 #endif
378 unsigned long task_size; /* size of task vm space */
379 unsigned long highest_vm_end; /* highest vma end address */
380 pgd_t * pgd;
381
382 /**
383 * @mm_users: The number of users including userspace.
384 *
385 * Use mmget()/mmget_not_zero()/mmput() to modify. When this drops
386 * to 0 (i.e. when the task exits and there are no other temporary
387 * reference holders), we also release a reference on @mm_count
388 * (which may then free the &struct mm_struct if @mm_count also
389 * drops to 0).
390 */
391 atomic_t mm_users;
392
393 /**
394 * @mm_count: The number of references to &struct mm_struct
395 * (@mm_users count as 1).
396 *
397 * Use mmgrab()/mmdrop() to modify. When this drops to 0, the
398 * &struct mm_struct is freed.
399 */
400 atomic_t mm_count;
401
402 atomic_long_t nr_ptes; /* PTE page table pages */
403 #if CONFIG_PGTABLE_LEVELS > 2
404 atomic_long_t nr_pmds; /* PMD page table pages */
405 #endif
406 int map_count; /* number of VMAs */
407
408 spinlock_t page_table_lock; /* Protects page tables and some counters */
409 struct rw_semaphore mmap_sem;
410
411 struct list_head mmlist; /* List of maybe swapped mm's. These are globally strung
412 * together off init_mm.mmlist, and are protected
413 * by mmlist_lock
414 */
415
416
417 unsigned long hiwater_rss; /* High-watermark of RSS usage */
418 unsigned long hiwater_vm; /* High-water virtual memory usage */
419
420 unsigned long total_vm; /* Total pages mapped */
421 unsigned long locked_vm; /* Pages that have PG_mlocked set */
422 unsigned long pinned_vm; /* Refcount permanently increased */
423 unsigned long data_vm; /* VM_WRITE & ~VM_SHARED & ~VM_STACK */
424 unsigned long exec_vm; /* VM_EXEC & ~VM_WRITE & ~VM_STACK */
425 unsigned long stack_vm; /* VM_STACK */
426 unsigned long def_flags;
427 unsigned long start_code, end_code, start_data, end_data;
428 unsigned long start_brk, brk, start_stack;
429 unsigned long arg_start, arg_end, env_start, env_end;
430
431 unsigned long saved_auxv[AT_VECTOR_SIZE]; /* for /proc/PID/auxv */
432
433 /*
434 * Special counters, in some configurations protected by the
435 * page_table_lock, in other configurations by being atomic.
436 */
437 struct mm_rss_stat rss_stat;
438
439 struct linux_binfmt *binfmt;
440
441 cpumask_var_t cpu_vm_mask_var;
442
443 /* Architecture-specific MM context */
444 mm_context_t context;
445
446 unsigned long flags; /* Must use atomic bitops to access the bits */
447
448 struct core_state *core_state; /* coredumping support */
449 #ifdef CONFIG_MEMBARRIER
450 atomic_t membarrier_state;
451 #endif
452 #ifdef CONFIG_AIO
453 spinlock_t ioctx_lock;
454 struct kioctx_table __rcu *ioctx_table;
455 #endif
456 #ifdef CONFIG_MEMCG
457 /*
458 * "owner" points to a task that is regarded as the canonical
459 * user/owner of this mm. All of the following must be true in
460 * order for it to be changed:
461 *
462 * current == mm->owner
463 * current->mm != mm
464 * new_owner->mm == mm
465 * new_owner->alloc_lock is held
466 */
467 struct task_struct __rcu *owner;
468 #endif
469 struct user_namespace *user_ns;
470
471 /* store ref to file /proc/<pid>/exe symlink points to */
472 struct file __rcu *exe_file;
473 #ifdef CONFIG_MMU_NOTIFIER
474 struct mmu_notifier_mm *mmu_notifier_mm;
475 #endif
476 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
477 pgtable_t pmd_huge_pte; /* protected by page_table_lock */
478 #endif
479 #ifdef CONFIG_CPUMASK_OFFSTACK
480 struct cpumask cpumask_allocation;
481 #endif
482 #ifdef CONFIG_NUMA_BALANCING
483 /*
484 * numa_next_scan is the next time that the PTEs will be marked
485 * pte_numa. NUMA hinting faults will gather statistics and migrate
486 * pages to new nodes if necessary.
487 */
488 unsigned long numa_next_scan;
489
490 /* Restart point for scanning and setting pte_numa */
491 unsigned long numa_scan_offset;
492
493 /* numa_scan_seq prevents two threads setting pte_numa */
494 int numa_scan_seq;
495 #endif
496 /*
497 * An operation with batched TLB flushing is going on. Anything that
498 * can move process memory needs to flush the TLB when moving a
499 * PROT_NONE or PROT_NUMA mapped page.
500 */
501 atomic_t tlb_flush_pending;
502 #ifdef CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH
503 /* See flush_tlb_batched_pending() */
504 bool tlb_flush_batched;
505 #endif
506 struct uprobes_state uprobes_state;
507 #ifdef CONFIG_HUGETLB_PAGE
508 atomic_long_t hugetlb_usage;
509 #endif
510 struct work_struct async_put_work;
511
512 #if IS_ENABLED(CONFIG_HMM)
513 /* HMM needs to track a few things per mm */
514 struct hmm *hmm;
515 #endif
516 } __randomize_layout;
517
518 extern struct mm_struct init_mm;
519
520 static inline void mm_init_cpumask(struct mm_struct *mm)
521 {
522 #ifdef CONFIG_CPUMASK_OFFSTACK
523 mm->cpu_vm_mask_var = &mm->cpumask_allocation;
524 #endif
525 cpumask_clear(mm->cpu_vm_mask_var);
526 }
527
528 /* Future-safe accessor for struct mm_struct's cpu_vm_mask. */
529 static inline cpumask_t *mm_cpumask(struct mm_struct *mm)
530 {
531 return mm->cpu_vm_mask_var;
532 }
533
534 struct mmu_gather;
535 extern void tlb_gather_mmu(struct mmu_gather *tlb, struct mm_struct *mm,
536 unsigned long start, unsigned long end);
537 extern void tlb_finish_mmu(struct mmu_gather *tlb,
538 unsigned long start, unsigned long end);
539
540 static inline void init_tlb_flush_pending(struct mm_struct *mm)
541 {
542 atomic_set(&mm->tlb_flush_pending, 0);
543 }
544
545 static inline void inc_tlb_flush_pending(struct mm_struct *mm)
546 {
547 atomic_inc(&mm->tlb_flush_pending);
548 /*
549 * The only time this value is relevant is when there are indeed pages
550 * to flush. And we'll only flush pages after changing them, which
551 * requires the PTL.
552 *
553 * So the ordering here is:
554 *
555 * atomic_inc(&mm->tlb_flush_pending);
556 * spin_lock(&ptl);
557 * ...
558 * set_pte_at();
559 * spin_unlock(&ptl);
560 *
561 * spin_lock(&ptl)
562 * mm_tlb_flush_pending();
563 * ....
564 * spin_unlock(&ptl);
565 *
566 * flush_tlb_range();
567 * atomic_dec(&mm->tlb_flush_pending);
568 *
569 * Where the increment if constrained by the PTL unlock, it thus
570 * ensures that the increment is visible if the PTE modification is
571 * visible. After all, if there is no PTE modification, nobody cares
572 * about TLB flushes either.
573 *
574 * This very much relies on users (mm_tlb_flush_pending() and
575 * mm_tlb_flush_nested()) only caring about _specific_ PTEs (and
576 * therefore specific PTLs), because with SPLIT_PTE_PTLOCKS and RCpc
577 * locks (PPC) the unlock of one doesn't order against the lock of
578 * another PTL.
579 *
580 * The decrement is ordered by the flush_tlb_range(), such that
581 * mm_tlb_flush_pending() will not return false unless all flushes have
582 * completed.
583 */
584 }
585
586 static inline void dec_tlb_flush_pending(struct mm_struct *mm)
587 {
588 /*
589 * See inc_tlb_flush_pending().
590 *
591 * This cannot be smp_mb__before_atomic() because smp_mb() simply does
592 * not order against TLB invalidate completion, which is what we need.
593 *
594 * Therefore we must rely on tlb_flush_*() to guarantee order.
595 */
596 atomic_dec(&mm->tlb_flush_pending);
597 }
598
599 static inline bool mm_tlb_flush_pending(struct mm_struct *mm)
600 {
601 /*
602 * Must be called after having acquired the PTL; orders against that
603 * PTLs release and therefore ensures that if we observe the modified
604 * PTE we must also observe the increment from inc_tlb_flush_pending().
605 *
606 * That is, it only guarantees to return true if there is a flush
607 * pending for _this_ PTL.
608 */
609 return atomic_read(&mm->tlb_flush_pending);
610 }
611
612 static inline bool mm_tlb_flush_nested(struct mm_struct *mm)
613 {
614 /*
615 * Similar to mm_tlb_flush_pending(), we must have acquired the PTL
616 * for which there is a TLB flush pending in order to guarantee
617 * we've seen both that PTE modification and the increment.
618 *
619 * (no requirement on actually still holding the PTL, that is irrelevant)
620 */
621 return atomic_read(&mm->tlb_flush_pending) > 1;
622 }
623
624 struct vm_fault;
625
626 struct vm_special_mapping {
627 const char *name; /* The name, e.g. "[vdso]". */
628
629 /*
630 * If .fault is not provided, this points to a
631 * NULL-terminated array of pages that back the special mapping.
632 *
633 * This must not be NULL unless .fault is provided.
634 */
635 struct page **pages;
636
637 /*
638 * If non-NULL, then this is called to resolve page faults
639 * on the special mapping. If used, .pages is not checked.
640 */
641 int (*fault)(const struct vm_special_mapping *sm,
642 struct vm_area_struct *vma,
643 struct vm_fault *vmf);
644
645 int (*mremap)(const struct vm_special_mapping *sm,
646 struct vm_area_struct *new_vma);
647 };
648
649 enum tlb_flush_reason {
650 TLB_FLUSH_ON_TASK_SWITCH,
651 TLB_REMOTE_SHOOTDOWN,
652 TLB_LOCAL_SHOOTDOWN,
653 TLB_LOCAL_MM_SHOOTDOWN,
654 TLB_REMOTE_SEND_IPI,
655 NR_TLB_FLUSH_REASONS,
656 };
657
658 /*
659 * A swap entry has to fit into a "unsigned long", as the entry is hidden
660 * in the "index" field of the swapper address space.
661 */
662 typedef struct {
663 unsigned long val;
664 } swp_entry_t;
665
666 #endif /* _LINUX_MM_TYPES_H */