]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blob - include/linux/mm_types.h
efi/arm: Fix boot crash with CONFIG_CPUMASK_OFFSTACK=y
[mirror_ubuntu-artful-kernel.git] / include / linux / mm_types.h
1 #ifndef _LINUX_MM_TYPES_H
2 #define _LINUX_MM_TYPES_H
3
4 #include <linux/auxvec.h>
5 #include <linux/types.h>
6 #include <linux/threads.h>
7 #include <linux/list.h>
8 #include <linux/spinlock.h>
9 #include <linux/rbtree.h>
10 #include <linux/rwsem.h>
11 #include <linux/completion.h>
12 #include <linux/cpumask.h>
13 #include <linux/uprobes.h>
14 #include <linux/page-flags-layout.h>
15 #include <linux/workqueue.h>
16 #include <asm/page.h>
17 #include <asm/mmu.h>
18
19 #ifndef AT_VECTOR_SIZE_ARCH
20 #define AT_VECTOR_SIZE_ARCH 0
21 #endif
22 #define AT_VECTOR_SIZE (2*(AT_VECTOR_SIZE_ARCH + AT_VECTOR_SIZE_BASE + 1))
23
24 struct address_space;
25 struct mem_cgroup;
26
27 #define USE_SPLIT_PTE_PTLOCKS (NR_CPUS >= CONFIG_SPLIT_PTLOCK_CPUS)
28 #define USE_SPLIT_PMD_PTLOCKS (USE_SPLIT_PTE_PTLOCKS && \
29 IS_ENABLED(CONFIG_ARCH_ENABLE_SPLIT_PMD_PTLOCK))
30 #define ALLOC_SPLIT_PTLOCKS (SPINLOCK_SIZE > BITS_PER_LONG/8)
31
32 /*
33 * Each physical page in the system has a struct page associated with
34 * it to keep track of whatever it is we are using the page for at the
35 * moment. Note that we have no way to track which tasks are using
36 * a page, though if it is a pagecache page, rmap structures can tell us
37 * who is mapping it.
38 *
39 * The objects in struct page are organized in double word blocks in
40 * order to allows us to use atomic double word operations on portions
41 * of struct page. That is currently only used by slub but the arrangement
42 * allows the use of atomic double word operations on the flags/mapping
43 * and lru list pointers also.
44 */
45 struct page {
46 /* First double word block */
47 unsigned long flags; /* Atomic flags, some possibly
48 * updated asynchronously */
49 union {
50 struct address_space *mapping; /* If low bit clear, points to
51 * inode address_space, or NULL.
52 * If page mapped as anonymous
53 * memory, low bit is set, and
54 * it points to anon_vma object:
55 * see PAGE_MAPPING_ANON below.
56 */
57 void *s_mem; /* slab first object */
58 atomic_t compound_mapcount; /* first tail page */
59 /* page_deferred_list().next -- second tail page */
60 };
61
62 /* Second double word */
63 union {
64 pgoff_t index; /* Our offset within mapping. */
65 void *freelist; /* sl[aou]b first free object */
66 /* page_deferred_list().prev -- second tail page */
67 };
68
69 union {
70 #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \
71 defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
72 /* Used for cmpxchg_double in slub */
73 unsigned long counters;
74 #else
75 /*
76 * Keep _refcount separate from slub cmpxchg_double data.
77 * As the rest of the double word is protected by slab_lock
78 * but _refcount is not.
79 */
80 unsigned counters;
81 #endif
82 struct {
83
84 union {
85 /*
86 * Count of ptes mapped in mms, to show when
87 * page is mapped & limit reverse map searches.
88 *
89 * Extra information about page type may be
90 * stored here for pages that are never mapped,
91 * in which case the value MUST BE <= -2.
92 * See page-flags.h for more details.
93 */
94 atomic_t _mapcount;
95
96 unsigned int active; /* SLAB */
97 struct { /* SLUB */
98 unsigned inuse:16;
99 unsigned objects:15;
100 unsigned frozen:1;
101 };
102 int units; /* SLOB */
103 };
104 /*
105 * Usage count, *USE WRAPPER FUNCTION* when manual
106 * accounting. See page_ref.h
107 */
108 atomic_t _refcount;
109 };
110 };
111
112 /*
113 * Third double word block
114 *
115 * WARNING: bit 0 of the first word encode PageTail(). That means
116 * the rest users of the storage space MUST NOT use the bit to
117 * avoid collision and false-positive PageTail().
118 */
119 union {
120 struct list_head lru; /* Pageout list, eg. active_list
121 * protected by zone_lru_lock !
122 * Can be used as a generic list
123 * by the page owner.
124 */
125 struct dev_pagemap *pgmap; /* ZONE_DEVICE pages are never on an
126 * lru or handled by a slab
127 * allocator, this points to the
128 * hosting device page map.
129 */
130 struct { /* slub per cpu partial pages */
131 struct page *next; /* Next partial slab */
132 #ifdef CONFIG_64BIT
133 int pages; /* Nr of partial slabs left */
134 int pobjects; /* Approximate # of objects */
135 #else
136 short int pages;
137 short int pobjects;
138 #endif
139 };
140
141 struct rcu_head rcu_head; /* Used by SLAB
142 * when destroying via RCU
143 */
144 /* Tail pages of compound page */
145 struct {
146 unsigned long compound_head; /* If bit zero is set */
147
148 /* First tail page only */
149 #ifdef CONFIG_64BIT
150 /*
151 * On 64 bit system we have enough space in struct page
152 * to encode compound_dtor and compound_order with
153 * unsigned int. It can help compiler generate better or
154 * smaller code on some archtectures.
155 */
156 unsigned int compound_dtor;
157 unsigned int compound_order;
158 #else
159 unsigned short int compound_dtor;
160 unsigned short int compound_order;
161 #endif
162 };
163
164 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && USE_SPLIT_PMD_PTLOCKS
165 struct {
166 unsigned long __pad; /* do not overlay pmd_huge_pte
167 * with compound_head to avoid
168 * possible bit 0 collision.
169 */
170 pgtable_t pmd_huge_pte; /* protected by page->ptl */
171 };
172 #endif
173 };
174
175 /* Remainder is not double word aligned */
176 union {
177 unsigned long private; /* Mapping-private opaque data:
178 * usually used for buffer_heads
179 * if PagePrivate set; used for
180 * swp_entry_t if PageSwapCache;
181 * indicates order in the buddy
182 * system if PG_buddy is set.
183 */
184 #if USE_SPLIT_PTE_PTLOCKS
185 #if ALLOC_SPLIT_PTLOCKS
186 spinlock_t *ptl;
187 #else
188 spinlock_t ptl;
189 #endif
190 #endif
191 struct kmem_cache *slab_cache; /* SL[AU]B: Pointer to slab */
192 };
193
194 #ifdef CONFIG_MEMCG
195 struct mem_cgroup *mem_cgroup;
196 #endif
197
198 /*
199 * On machines where all RAM is mapped into kernel address space,
200 * we can simply calculate the virtual address. On machines with
201 * highmem some memory is mapped into kernel virtual memory
202 * dynamically, so we need a place to store that address.
203 * Note that this field could be 16 bits on x86 ... ;)
204 *
205 * Architectures with slow multiplication can define
206 * WANT_PAGE_VIRTUAL in asm/page.h
207 */
208 #if defined(WANT_PAGE_VIRTUAL)
209 void *virtual; /* Kernel virtual address (NULL if
210 not kmapped, ie. highmem) */
211 #endif /* WANT_PAGE_VIRTUAL */
212
213 #ifdef CONFIG_KMEMCHECK
214 /*
215 * kmemcheck wants to track the status of each byte in a page; this
216 * is a pointer to such a status block. NULL if not tracked.
217 */
218 void *shadow;
219 #endif
220
221 #ifdef LAST_CPUPID_NOT_IN_PAGE_FLAGS
222 int _last_cpupid;
223 #endif
224 }
225 /*
226 * The struct page can be forced to be double word aligned so that atomic ops
227 * on double words work. The SLUB allocator can make use of such a feature.
228 */
229 #ifdef CONFIG_HAVE_ALIGNED_STRUCT_PAGE
230 __aligned(2 * sizeof(unsigned long))
231 #endif
232 ;
233
234 struct page_frag {
235 struct page *page;
236 #if (BITS_PER_LONG > 32) || (PAGE_SIZE >= 65536)
237 __u32 offset;
238 __u32 size;
239 #else
240 __u16 offset;
241 __u16 size;
242 #endif
243 };
244
245 #define PAGE_FRAG_CACHE_MAX_SIZE __ALIGN_MASK(32768, ~PAGE_MASK)
246 #define PAGE_FRAG_CACHE_MAX_ORDER get_order(PAGE_FRAG_CACHE_MAX_SIZE)
247
248 struct page_frag_cache {
249 void * va;
250 #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
251 __u16 offset;
252 __u16 size;
253 #else
254 __u32 offset;
255 #endif
256 /* we maintain a pagecount bias, so that we dont dirty cache line
257 * containing page->_refcount every time we allocate a fragment.
258 */
259 unsigned int pagecnt_bias;
260 bool pfmemalloc;
261 };
262
263 typedef unsigned long vm_flags_t;
264
265 /*
266 * A region containing a mapping of a non-memory backed file under NOMMU
267 * conditions. These are held in a global tree and are pinned by the VMAs that
268 * map parts of them.
269 */
270 struct vm_region {
271 struct rb_node vm_rb; /* link in global region tree */
272 vm_flags_t vm_flags; /* VMA vm_flags */
273 unsigned long vm_start; /* start address of region */
274 unsigned long vm_end; /* region initialised to here */
275 unsigned long vm_top; /* region allocated to here */
276 unsigned long vm_pgoff; /* the offset in vm_file corresponding to vm_start */
277 struct file *vm_file; /* the backing file or NULL */
278
279 int vm_usage; /* region usage count (access under nommu_region_sem) */
280 bool vm_icache_flushed : 1; /* true if the icache has been flushed for
281 * this region */
282 };
283
284 #ifdef CONFIG_USERFAULTFD
285 #define NULL_VM_UFFD_CTX ((struct vm_userfaultfd_ctx) { NULL, })
286 struct vm_userfaultfd_ctx {
287 struct userfaultfd_ctx *ctx;
288 };
289 #else /* CONFIG_USERFAULTFD */
290 #define NULL_VM_UFFD_CTX ((struct vm_userfaultfd_ctx) {})
291 struct vm_userfaultfd_ctx {};
292 #endif /* CONFIG_USERFAULTFD */
293
294 /*
295 * This struct defines a memory VMM memory area. There is one of these
296 * per VM-area/task. A VM area is any part of the process virtual memory
297 * space that has a special rule for the page-fault handlers (ie a shared
298 * library, the executable area etc).
299 */
300 struct vm_area_struct {
301 /* The first cache line has the info for VMA tree walking. */
302
303 unsigned long vm_start; /* Our start address within vm_mm. */
304 unsigned long vm_end; /* The first byte after our end address
305 within vm_mm. */
306
307 /* linked list of VM areas per task, sorted by address */
308 struct vm_area_struct *vm_next, *vm_prev;
309
310 struct rb_node vm_rb;
311
312 /*
313 * Largest free memory gap in bytes to the left of this VMA.
314 * Either between this VMA and vma->vm_prev, or between one of the
315 * VMAs below us in the VMA rbtree and its ->vm_prev. This helps
316 * get_unmapped_area find a free area of the right size.
317 */
318 unsigned long rb_subtree_gap;
319
320 /* Second cache line starts here. */
321
322 struct mm_struct *vm_mm; /* The address space we belong to. */
323 pgprot_t vm_page_prot; /* Access permissions of this VMA. */
324 unsigned long vm_flags; /* Flags, see mm.h. */
325
326 /*
327 * For areas with an address space and backing store,
328 * linkage into the address_space->i_mmap interval tree.
329 */
330 struct {
331 struct rb_node rb;
332 unsigned long rb_subtree_last;
333 } shared;
334
335 /*
336 * A file's MAP_PRIVATE vma can be in both i_mmap tree and anon_vma
337 * list, after a COW of one of the file pages. A MAP_SHARED vma
338 * can only be in the i_mmap tree. An anonymous MAP_PRIVATE, stack
339 * or brk vma (with NULL file) can only be in an anon_vma list.
340 */
341 struct list_head anon_vma_chain; /* Serialized by mmap_sem &
342 * page_table_lock */
343 struct anon_vma *anon_vma; /* Serialized by page_table_lock */
344
345 /* Function pointers to deal with this struct. */
346 const struct vm_operations_struct *vm_ops;
347
348 /* Information about our backing store: */
349 unsigned long vm_pgoff; /* Offset (within vm_file) in PAGE_SIZE
350 units */
351 struct file * vm_file; /* File we map to (can be NULL). */
352 void * vm_private_data; /* was vm_pte (shared mem) */
353
354 #ifndef CONFIG_MMU
355 struct vm_region *vm_region; /* NOMMU mapping region */
356 #endif
357 #ifdef CONFIG_NUMA
358 struct mempolicy *vm_policy; /* NUMA policy for the VMA */
359 #endif
360 struct vm_userfaultfd_ctx vm_userfaultfd_ctx;
361 };
362
363 struct core_thread {
364 struct task_struct *task;
365 struct core_thread *next;
366 };
367
368 struct core_state {
369 atomic_t nr_threads;
370 struct core_thread dumper;
371 struct completion startup;
372 };
373
374 enum {
375 MM_FILEPAGES, /* Resident file mapping pages */
376 MM_ANONPAGES, /* Resident anonymous pages */
377 MM_SWAPENTS, /* Anonymous swap entries */
378 MM_SHMEMPAGES, /* Resident shared memory pages */
379 NR_MM_COUNTERS
380 };
381
382 #if USE_SPLIT_PTE_PTLOCKS && defined(CONFIG_MMU)
383 #define SPLIT_RSS_COUNTING
384 /* per-thread cached information, */
385 struct task_rss_stat {
386 int events; /* for synchronization threshold */
387 int count[NR_MM_COUNTERS];
388 };
389 #endif /* USE_SPLIT_PTE_PTLOCKS */
390
391 struct mm_rss_stat {
392 atomic_long_t count[NR_MM_COUNTERS];
393 };
394
395 struct kioctx_table;
396 struct mm_struct {
397 struct vm_area_struct *mmap; /* list of VMAs */
398 struct rb_root mm_rb;
399 u32 vmacache_seqnum; /* per-thread vmacache */
400 #ifdef CONFIG_MMU
401 unsigned long (*get_unmapped_area) (struct file *filp,
402 unsigned long addr, unsigned long len,
403 unsigned long pgoff, unsigned long flags);
404 #endif
405 unsigned long mmap_base; /* base of mmap area */
406 unsigned long mmap_legacy_base; /* base of mmap area in bottom-up allocations */
407 unsigned long task_size; /* size of task vm space */
408 unsigned long highest_vm_end; /* highest vma end address */
409 pgd_t * pgd;
410
411 /**
412 * @mm_users: The number of users including userspace.
413 *
414 * Use mmget()/mmget_not_zero()/mmput() to modify. When this drops
415 * to 0 (i.e. when the task exits and there are no other temporary
416 * reference holders), we also release a reference on @mm_count
417 * (which may then free the &struct mm_struct if @mm_count also
418 * drops to 0).
419 */
420 atomic_t mm_users;
421
422 /**
423 * @mm_count: The number of references to &struct mm_struct
424 * (@mm_users count as 1).
425 *
426 * Use mmgrab()/mmdrop() to modify. When this drops to 0, the
427 * &struct mm_struct is freed.
428 */
429 atomic_t mm_count;
430
431 atomic_long_t nr_ptes; /* PTE page table pages */
432 #if CONFIG_PGTABLE_LEVELS > 2
433 atomic_long_t nr_pmds; /* PMD page table pages */
434 #endif
435 int map_count; /* number of VMAs */
436
437 spinlock_t page_table_lock; /* Protects page tables and some counters */
438 struct rw_semaphore mmap_sem;
439
440 struct list_head mmlist; /* List of maybe swapped mm's. These are globally strung
441 * together off init_mm.mmlist, and are protected
442 * by mmlist_lock
443 */
444
445
446 unsigned long hiwater_rss; /* High-watermark of RSS usage */
447 unsigned long hiwater_vm; /* High-water virtual memory usage */
448
449 unsigned long total_vm; /* Total pages mapped */
450 unsigned long locked_vm; /* Pages that have PG_mlocked set */
451 unsigned long pinned_vm; /* Refcount permanently increased */
452 unsigned long data_vm; /* VM_WRITE & ~VM_SHARED & ~VM_STACK */
453 unsigned long exec_vm; /* VM_EXEC & ~VM_WRITE & ~VM_STACK */
454 unsigned long stack_vm; /* VM_STACK */
455 unsigned long def_flags;
456 unsigned long start_code, end_code, start_data, end_data;
457 unsigned long start_brk, brk, start_stack;
458 unsigned long arg_start, arg_end, env_start, env_end;
459
460 unsigned long saved_auxv[AT_VECTOR_SIZE]; /* for /proc/PID/auxv */
461
462 /*
463 * Special counters, in some configurations protected by the
464 * page_table_lock, in other configurations by being atomic.
465 */
466 struct mm_rss_stat rss_stat;
467
468 struct linux_binfmt *binfmt;
469
470 cpumask_var_t cpu_vm_mask_var;
471
472 /* Architecture-specific MM context */
473 mm_context_t context;
474
475 unsigned long flags; /* Must use atomic bitops to access the bits */
476
477 struct core_state *core_state; /* coredumping support */
478 #ifdef CONFIG_AIO
479 spinlock_t ioctx_lock;
480 struct kioctx_table __rcu *ioctx_table;
481 #endif
482 #ifdef CONFIG_MEMCG
483 /*
484 * "owner" points to a task that is regarded as the canonical
485 * user/owner of this mm. All of the following must be true in
486 * order for it to be changed:
487 *
488 * current == mm->owner
489 * current->mm != mm
490 * new_owner->mm == mm
491 * new_owner->alloc_lock is held
492 */
493 struct task_struct __rcu *owner;
494 #endif
495 struct user_namespace *user_ns;
496
497 /* store ref to file /proc/<pid>/exe symlink points to */
498 struct file __rcu *exe_file;
499 #ifdef CONFIG_MMU_NOTIFIER
500 struct mmu_notifier_mm *mmu_notifier_mm;
501 #endif
502 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
503 pgtable_t pmd_huge_pte; /* protected by page_table_lock */
504 #endif
505 #ifdef CONFIG_CPUMASK_OFFSTACK
506 struct cpumask cpumask_allocation;
507 #endif
508 #ifdef CONFIG_NUMA_BALANCING
509 /*
510 * numa_next_scan is the next time that the PTEs will be marked
511 * pte_numa. NUMA hinting faults will gather statistics and migrate
512 * pages to new nodes if necessary.
513 */
514 unsigned long numa_next_scan;
515
516 /* Restart point for scanning and setting pte_numa */
517 unsigned long numa_scan_offset;
518
519 /* numa_scan_seq prevents two threads setting pte_numa */
520 int numa_scan_seq;
521 #endif
522 #if defined(CONFIG_NUMA_BALANCING) || defined(CONFIG_COMPACTION)
523 /*
524 * An operation with batched TLB flushing is going on. Anything that
525 * can move process memory needs to flush the TLB when moving a
526 * PROT_NONE or PROT_NUMA mapped page.
527 */
528 bool tlb_flush_pending;
529 #endif
530 struct uprobes_state uprobes_state;
531 #ifdef CONFIG_HUGETLB_PAGE
532 atomic_long_t hugetlb_usage;
533 #endif
534 struct work_struct async_put_work;
535 };
536
537 static inline void mm_init_cpumask(struct mm_struct *mm)
538 {
539 #ifdef CONFIG_CPUMASK_OFFSTACK
540 mm->cpu_vm_mask_var = &mm->cpumask_allocation;
541 #endif
542 cpumask_clear(mm->cpu_vm_mask_var);
543 }
544
545 /* Future-safe accessor for struct mm_struct's cpu_vm_mask. */
546 static inline cpumask_t *mm_cpumask(struct mm_struct *mm)
547 {
548 return mm->cpu_vm_mask_var;
549 }
550
551 #if defined(CONFIG_NUMA_BALANCING) || defined(CONFIG_COMPACTION)
552 /*
553 * Memory barriers to keep this state in sync are graciously provided by
554 * the page table locks, outside of which no page table modifications happen.
555 * The barriers below prevent the compiler from re-ordering the instructions
556 * around the memory barriers that are already present in the code.
557 */
558 static inline bool mm_tlb_flush_pending(struct mm_struct *mm)
559 {
560 barrier();
561 return mm->tlb_flush_pending;
562 }
563 static inline void set_tlb_flush_pending(struct mm_struct *mm)
564 {
565 mm->tlb_flush_pending = true;
566
567 /*
568 * Guarantee that the tlb_flush_pending store does not leak into the
569 * critical section updating the page tables
570 */
571 smp_mb__before_spinlock();
572 }
573 /* Clearing is done after a TLB flush, which also provides a barrier. */
574 static inline void clear_tlb_flush_pending(struct mm_struct *mm)
575 {
576 barrier();
577 mm->tlb_flush_pending = false;
578 }
579 #else
580 static inline bool mm_tlb_flush_pending(struct mm_struct *mm)
581 {
582 return false;
583 }
584 static inline void set_tlb_flush_pending(struct mm_struct *mm)
585 {
586 }
587 static inline void clear_tlb_flush_pending(struct mm_struct *mm)
588 {
589 }
590 #endif
591
592 struct vm_fault;
593
594 struct vm_special_mapping {
595 const char *name; /* The name, e.g. "[vdso]". */
596
597 /*
598 * If .fault is not provided, this points to a
599 * NULL-terminated array of pages that back the special mapping.
600 *
601 * This must not be NULL unless .fault is provided.
602 */
603 struct page **pages;
604
605 /*
606 * If non-NULL, then this is called to resolve page faults
607 * on the special mapping. If used, .pages is not checked.
608 */
609 int (*fault)(const struct vm_special_mapping *sm,
610 struct vm_area_struct *vma,
611 struct vm_fault *vmf);
612
613 int (*mremap)(const struct vm_special_mapping *sm,
614 struct vm_area_struct *new_vma);
615 };
616
617 enum tlb_flush_reason {
618 TLB_FLUSH_ON_TASK_SWITCH,
619 TLB_REMOTE_SHOOTDOWN,
620 TLB_LOCAL_SHOOTDOWN,
621 TLB_LOCAL_MM_SHOOTDOWN,
622 TLB_REMOTE_SEND_IPI,
623 NR_TLB_FLUSH_REASONS,
624 };
625
626 /*
627 * A swap entry has to fit into a "unsigned long", as the entry is hidden
628 * in the "index" field of the swapper address space.
629 */
630 typedef struct {
631 unsigned long val;
632 } swp_entry_t;
633
634 #endif /* _LINUX_MM_TYPES_H */