Merge tag 'trace-printf-v6.13' of git://git.kernel.org/pub/scm/linux/kernel/git/trace...
[drm/drm-misc.git] / mm / slab.h
blob632fedd71fea7bebb3679ef51c8eb2825f39e77f
1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef MM_SLAB_H
3 #define MM_SLAB_H
5 #include <linux/reciprocal_div.h>
6 #include <linux/list_lru.h>
7 #include <linux/local_lock.h>
8 #include <linux/random.h>
9 #include <linux/kobject.h>
10 #include <linux/sched/mm.h>
11 #include <linux/memcontrol.h>
12 #include <linux/kfence.h>
13 #include <linux/kasan.h>
16 * Internal slab definitions
19 #ifdef CONFIG_64BIT
20 # ifdef system_has_cmpxchg128
21 # define system_has_freelist_aba() system_has_cmpxchg128()
22 # define try_cmpxchg_freelist try_cmpxchg128
23 # endif
24 #define this_cpu_try_cmpxchg_freelist this_cpu_try_cmpxchg128
25 typedef u128 freelist_full_t;
26 #else /* CONFIG_64BIT */
27 # ifdef system_has_cmpxchg64
28 # define system_has_freelist_aba() system_has_cmpxchg64()
29 # define try_cmpxchg_freelist try_cmpxchg64
30 # endif
31 #define this_cpu_try_cmpxchg_freelist this_cpu_try_cmpxchg64
32 typedef u64 freelist_full_t;
33 #endif /* CONFIG_64BIT */
35 #if defined(system_has_freelist_aba) && !defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
36 #undef system_has_freelist_aba
37 #endif
40 * Freelist pointer and counter to cmpxchg together, avoids the typical ABA
41 * problems with cmpxchg of just a pointer.
43 typedef union {
44 struct {
45 void *freelist;
46 unsigned long counter;
48 freelist_full_t full;
49 } freelist_aba_t;
51 /* Reuses the bits in struct page */
52 struct slab {
53 unsigned long __page_flags;
55 struct kmem_cache *slab_cache;
56 union {
57 struct {
58 union {
59 struct list_head slab_list;
60 #ifdef CONFIG_SLUB_CPU_PARTIAL
61 struct {
62 struct slab *next;
63 int slabs; /* Nr of slabs left */
65 #endif
67 /* Double-word boundary */
68 union {
69 struct {
70 void *freelist; /* first free object */
71 union {
72 unsigned long counters;
73 struct {
74 unsigned inuse:16;
75 unsigned objects:15;
77 * If slab debugging is enabled then the
78 * frozen bit can be reused to indicate
79 * that the slab was corrupted
81 unsigned frozen:1;
85 #ifdef system_has_freelist_aba
86 freelist_aba_t freelist_counter;
87 #endif
90 struct rcu_head rcu_head;
93 unsigned int __page_type;
94 atomic_t __page_refcount;
95 #ifdef CONFIG_SLAB_OBJ_EXT
96 unsigned long obj_exts;
97 #endif
100 #define SLAB_MATCH(pg, sl) \
101 static_assert(offsetof(struct page, pg) == offsetof(struct slab, sl))
102 SLAB_MATCH(flags, __page_flags);
103 SLAB_MATCH(compound_head, slab_cache); /* Ensure bit 0 is clear */
104 SLAB_MATCH(_refcount, __page_refcount);
105 #ifdef CONFIG_MEMCG
106 SLAB_MATCH(memcg_data, obj_exts);
107 #elif defined(CONFIG_SLAB_OBJ_EXT)
108 SLAB_MATCH(_unused_slab_obj_exts, obj_exts);
109 #endif
110 #undef SLAB_MATCH
111 static_assert(sizeof(struct slab) <= sizeof(struct page));
112 #if defined(system_has_freelist_aba)
113 static_assert(IS_ALIGNED(offsetof(struct slab, freelist), sizeof(freelist_aba_t)));
114 #endif
117 * folio_slab - Converts from folio to slab.
118 * @folio: The folio.
120 * Currently struct slab is a different representation of a folio where
121 * folio_test_slab() is true.
123 * Return: The slab which contains this folio.
125 #define folio_slab(folio) (_Generic((folio), \
126 const struct folio *: (const struct slab *)(folio), \
127 struct folio *: (struct slab *)(folio)))
130 * slab_folio - The folio allocated for a slab
131 * @slab: The slab.
133 * Slabs are allocated as folios that contain the individual objects and are
134 * using some fields in the first struct page of the folio - those fields are
135 * now accessed by struct slab. It is occasionally necessary to convert back to
136 * a folio in order to communicate with the rest of the mm. Please use this
137 * helper function instead of casting yourself, as the implementation may change
138 * in the future.
140 #define slab_folio(s) (_Generic((s), \
141 const struct slab *: (const struct folio *)s, \
142 struct slab *: (struct folio *)s))
145 * page_slab - Converts from first struct page to slab.
146 * @p: The first (either head of compound or single) page of slab.
148 * A temporary wrapper to convert struct page to struct slab in situations where
149 * we know the page is the compound head, or single order-0 page.
151 * Long-term ideally everything would work with struct slab directly or go
152 * through folio to struct slab.
154 * Return: The slab which contains this page
156 #define page_slab(p) (_Generic((p), \
157 const struct page *: (const struct slab *)(p), \
158 struct page *: (struct slab *)(p)))
161 * slab_page - The first struct page allocated for a slab
162 * @slab: The slab.
164 * A convenience wrapper for converting slab to the first struct page of the
165 * underlying folio, to communicate with code not yet converted to folio or
166 * struct slab.
168 #define slab_page(s) folio_page(slab_folio(s), 0)
171 * If network-based swap is enabled, sl*b must keep track of whether pages
172 * were allocated from pfmemalloc reserves.
174 static inline bool slab_test_pfmemalloc(const struct slab *slab)
176 return folio_test_active(slab_folio(slab));
179 static inline void slab_set_pfmemalloc(struct slab *slab)
181 folio_set_active(slab_folio(slab));
184 static inline void slab_clear_pfmemalloc(struct slab *slab)
186 folio_clear_active(slab_folio(slab));
189 static inline void __slab_clear_pfmemalloc(struct slab *slab)
191 __folio_clear_active(slab_folio(slab));
194 static inline void *slab_address(const struct slab *slab)
196 return folio_address(slab_folio(slab));
199 static inline int slab_nid(const struct slab *slab)
201 return folio_nid(slab_folio(slab));
204 static inline pg_data_t *slab_pgdat(const struct slab *slab)
206 return folio_pgdat(slab_folio(slab));
209 static inline struct slab *virt_to_slab(const void *addr)
211 struct folio *folio = virt_to_folio(addr);
213 if (!folio_test_slab(folio))
214 return NULL;
216 return folio_slab(folio);
219 static inline int slab_order(const struct slab *slab)
221 return folio_order(slab_folio(slab));
224 static inline size_t slab_size(const struct slab *slab)
226 return PAGE_SIZE << slab_order(slab);
229 #ifdef CONFIG_SLUB_CPU_PARTIAL
230 #define slub_percpu_partial(c) ((c)->partial)
232 #define slub_set_percpu_partial(c, p) \
233 ({ \
234 slub_percpu_partial(c) = (p)->next; \
237 #define slub_percpu_partial_read_once(c) READ_ONCE(slub_percpu_partial(c))
238 #else
239 #define slub_percpu_partial(c) NULL
241 #define slub_set_percpu_partial(c, p)
243 #define slub_percpu_partial_read_once(c) NULL
244 #endif // CONFIG_SLUB_CPU_PARTIAL
247 * Word size structure that can be atomically updated or read and that
248 * contains both the order and the number of objects that a slab of the
249 * given order would contain.
251 struct kmem_cache_order_objects {
252 unsigned int x;
256 * Slab cache management.
258 struct kmem_cache {
259 #ifndef CONFIG_SLUB_TINY
260 struct kmem_cache_cpu __percpu *cpu_slab;
261 #endif
262 /* Used for retrieving partial slabs, etc. */
263 slab_flags_t flags;
264 unsigned long min_partial;
265 unsigned int size; /* Object size including metadata */
266 unsigned int object_size; /* Object size without metadata */
267 struct reciprocal_value reciprocal_size;
268 unsigned int offset; /* Free pointer offset */
269 #ifdef CONFIG_SLUB_CPU_PARTIAL
270 /* Number of per cpu partial objects to keep around */
271 unsigned int cpu_partial;
272 /* Number of per cpu partial slabs to keep around */
273 unsigned int cpu_partial_slabs;
274 #endif
275 struct kmem_cache_order_objects oo;
277 /* Allocation and freeing of slabs */
278 struct kmem_cache_order_objects min;
279 gfp_t allocflags; /* gfp flags to use on each alloc */
280 int refcount; /* Refcount for slab cache destroy */
281 void (*ctor)(void *object); /* Object constructor */
282 unsigned int inuse; /* Offset to metadata */
283 unsigned int align; /* Alignment */
284 unsigned int red_left_pad; /* Left redzone padding size */
285 const char *name; /* Name (only for display!) */
286 struct list_head list; /* List of slab caches */
287 #ifdef CONFIG_SYSFS
288 struct kobject kobj; /* For sysfs */
289 #endif
290 #ifdef CONFIG_SLAB_FREELIST_HARDENED
291 unsigned long random;
292 #endif
294 #ifdef CONFIG_NUMA
296 * Defragmentation by allocating from a remote node.
298 unsigned int remote_node_defrag_ratio;
299 #endif
301 #ifdef CONFIG_SLAB_FREELIST_RANDOM
302 unsigned int *random_seq;
303 #endif
305 #ifdef CONFIG_KASAN_GENERIC
306 struct kasan_cache kasan_info;
307 #endif
309 #ifdef CONFIG_HARDENED_USERCOPY
310 unsigned int useroffset; /* Usercopy region offset */
311 unsigned int usersize; /* Usercopy region size */
312 #endif
314 struct kmem_cache_node *node[MAX_NUMNODES];
317 #if defined(CONFIG_SYSFS) && !defined(CONFIG_SLUB_TINY)
318 #define SLAB_SUPPORTS_SYSFS 1
319 void sysfs_slab_unlink(struct kmem_cache *s);
320 void sysfs_slab_release(struct kmem_cache *s);
321 #else
322 static inline void sysfs_slab_unlink(struct kmem_cache *s) { }
323 static inline void sysfs_slab_release(struct kmem_cache *s) { }
324 #endif
326 void *fixup_red_left(struct kmem_cache *s, void *p);
328 static inline void *nearest_obj(struct kmem_cache *cache,
329 const struct slab *slab, void *x)
331 void *object = x - (x - slab_address(slab)) % cache->size;
332 void *last_object = slab_address(slab) +
333 (slab->objects - 1) * cache->size;
334 void *result = (unlikely(object > last_object)) ? last_object : object;
336 result = fixup_red_left(cache, result);
337 return result;
340 /* Determine object index from a given position */
341 static inline unsigned int __obj_to_index(const struct kmem_cache *cache,
342 void *addr, void *obj)
344 return reciprocal_divide(kasan_reset_tag(obj) - addr,
345 cache->reciprocal_size);
348 static inline unsigned int obj_to_index(const struct kmem_cache *cache,
349 const struct slab *slab, void *obj)
351 if (is_kfence_address(obj))
352 return 0;
353 return __obj_to_index(cache, slab_address(slab), obj);
356 static inline int objs_per_slab(const struct kmem_cache *cache,
357 const struct slab *slab)
359 return slab->objects;
363 * State of the slab allocator.
365 * This is used to describe the states of the allocator during bootup.
366 * Allocators use this to gradually bootstrap themselves. Most allocators
367 * have the problem that the structures used for managing slab caches are
368 * allocated from slab caches themselves.
370 enum slab_state {
371 DOWN, /* No slab functionality yet */
372 PARTIAL, /* SLUB: kmem_cache_node available */
373 UP, /* Slab caches usable but not all extras yet */
374 FULL /* Everything is working */
377 extern enum slab_state slab_state;
379 /* The slab cache mutex protects the management structures during changes */
380 extern struct mutex slab_mutex;
382 /* The list of all slab caches on the system */
383 extern struct list_head slab_caches;
385 /* The slab cache that manages slab cache information */
386 extern struct kmem_cache *kmem_cache;
388 /* A table of kmalloc cache names and sizes */
389 extern const struct kmalloc_info_struct {
390 const char *name[NR_KMALLOC_TYPES];
391 unsigned int size;
392 } kmalloc_info[];
394 /* Kmalloc array related functions */
395 void setup_kmalloc_cache_index_table(void);
396 void create_kmalloc_caches(void);
398 extern u8 kmalloc_size_index[24];
400 static inline unsigned int size_index_elem(unsigned int bytes)
402 return (bytes - 1) / 8;
406 * Find the kmem_cache structure that serves a given size of
407 * allocation
409 * This assumes size is larger than zero and not larger than
410 * KMALLOC_MAX_CACHE_SIZE and the caller must check that.
412 static inline struct kmem_cache *
413 kmalloc_slab(size_t size, kmem_buckets *b, gfp_t flags, unsigned long caller)
415 unsigned int index;
417 if (!b)
418 b = &kmalloc_caches[kmalloc_type(flags, caller)];
419 if (size <= 192)
420 index = kmalloc_size_index[size_index_elem(size)];
421 else
422 index = fls(size - 1);
424 return (*b)[index];
427 gfp_t kmalloc_fix_flags(gfp_t flags);
429 /* Functions provided by the slab allocators */
430 int do_kmem_cache_create(struct kmem_cache *s, const char *name,
431 unsigned int size, struct kmem_cache_args *args,
432 slab_flags_t flags);
434 void __init kmem_cache_init(void);
435 extern void create_boot_cache(struct kmem_cache *, const char *name,
436 unsigned int size, slab_flags_t flags,
437 unsigned int useroffset, unsigned int usersize);
439 int slab_unmergeable(struct kmem_cache *s);
440 struct kmem_cache *find_mergeable(unsigned size, unsigned align,
441 slab_flags_t flags, const char *name, void (*ctor)(void *));
442 struct kmem_cache *
443 __kmem_cache_alias(const char *name, unsigned int size, unsigned int align,
444 slab_flags_t flags, void (*ctor)(void *));
446 slab_flags_t kmem_cache_flags(slab_flags_t flags, const char *name);
448 static inline bool is_kmalloc_cache(struct kmem_cache *s)
450 return (s->flags & SLAB_KMALLOC);
453 static inline bool is_kmalloc_normal(struct kmem_cache *s)
455 if (!is_kmalloc_cache(s))
456 return false;
457 return !(s->flags & (SLAB_CACHE_DMA|SLAB_ACCOUNT|SLAB_RECLAIM_ACCOUNT));
460 /* Legal flag mask for kmem_cache_create(), for various configurations */
461 #define SLAB_CORE_FLAGS (SLAB_HWCACHE_ALIGN | SLAB_CACHE_DMA | \
462 SLAB_CACHE_DMA32 | SLAB_PANIC | \
463 SLAB_TYPESAFE_BY_RCU | SLAB_DEBUG_OBJECTS )
465 #ifdef CONFIG_SLUB_DEBUG
466 #define SLAB_DEBUG_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
467 SLAB_TRACE | SLAB_CONSISTENCY_CHECKS)
468 #else
469 #define SLAB_DEBUG_FLAGS (0)
470 #endif
472 #define SLAB_CACHE_FLAGS (SLAB_NOLEAKTRACE | SLAB_RECLAIM_ACCOUNT | \
473 SLAB_TEMPORARY | SLAB_ACCOUNT | \
474 SLAB_NO_USER_FLAGS | SLAB_KMALLOC | SLAB_NO_MERGE)
476 /* Common flags available with current configuration */
477 #define CACHE_CREATE_MASK (SLAB_CORE_FLAGS | SLAB_DEBUG_FLAGS | SLAB_CACHE_FLAGS)
479 /* Common flags permitted for kmem_cache_create */
480 #define SLAB_FLAGS_PERMITTED (SLAB_CORE_FLAGS | \
481 SLAB_RED_ZONE | \
482 SLAB_POISON | \
483 SLAB_STORE_USER | \
484 SLAB_TRACE | \
485 SLAB_CONSISTENCY_CHECKS | \
486 SLAB_NOLEAKTRACE | \
487 SLAB_RECLAIM_ACCOUNT | \
488 SLAB_TEMPORARY | \
489 SLAB_ACCOUNT | \
490 SLAB_KMALLOC | \
491 SLAB_NO_MERGE | \
492 SLAB_NO_USER_FLAGS)
494 bool __kmem_cache_empty(struct kmem_cache *);
495 int __kmem_cache_shutdown(struct kmem_cache *);
496 void __kmem_cache_release(struct kmem_cache *);
497 int __kmem_cache_shrink(struct kmem_cache *);
498 void slab_kmem_cache_release(struct kmem_cache *);
500 struct seq_file;
501 struct file;
503 struct slabinfo {
504 unsigned long active_objs;
505 unsigned long num_objs;
506 unsigned long active_slabs;
507 unsigned long num_slabs;
508 unsigned long shared_avail;
509 unsigned int limit;
510 unsigned int batchcount;
511 unsigned int shared;
512 unsigned int objects_per_slab;
513 unsigned int cache_order;
516 void get_slabinfo(struct kmem_cache *s, struct slabinfo *sinfo);
518 #ifdef CONFIG_SLUB_DEBUG
519 #ifdef CONFIG_SLUB_DEBUG_ON
520 DECLARE_STATIC_KEY_TRUE(slub_debug_enabled);
521 #else
522 DECLARE_STATIC_KEY_FALSE(slub_debug_enabled);
523 #endif
524 extern void print_tracking(struct kmem_cache *s, void *object);
525 long validate_slab_cache(struct kmem_cache *s);
526 static inline bool __slub_debug_enabled(void)
528 return static_branch_unlikely(&slub_debug_enabled);
530 #else
531 static inline void print_tracking(struct kmem_cache *s, void *object)
534 static inline bool __slub_debug_enabled(void)
536 return false;
538 #endif
541 * Returns true if any of the specified slab_debug flags is enabled for the
542 * cache. Use only for flags parsed by setup_slub_debug() as it also enables
543 * the static key.
545 static inline bool kmem_cache_debug_flags(struct kmem_cache *s, slab_flags_t flags)
547 if (IS_ENABLED(CONFIG_SLUB_DEBUG))
548 VM_WARN_ON_ONCE(!(flags & SLAB_DEBUG_FLAGS));
549 if (__slub_debug_enabled())
550 return s->flags & flags;
551 return false;
554 #if IS_ENABLED(CONFIG_SLUB_DEBUG) && IS_ENABLED(CONFIG_KUNIT)
555 bool slab_in_kunit_test(void);
556 #else
557 static inline bool slab_in_kunit_test(void) { return false; }
558 #endif
560 #ifdef CONFIG_SLAB_OBJ_EXT
563 * slab_obj_exts - get the pointer to the slab object extension vector
564 * associated with a slab.
565 * @slab: a pointer to the slab struct
567 * Returns a pointer to the object extension vector associated with the slab,
568 * or NULL if no such vector has been associated yet.
570 static inline struct slabobj_ext *slab_obj_exts(struct slab *slab)
572 unsigned long obj_exts = READ_ONCE(slab->obj_exts);
574 #ifdef CONFIG_MEMCG
575 VM_BUG_ON_PAGE(obj_exts && !(obj_exts & MEMCG_DATA_OBJEXTS),
576 slab_page(slab));
577 VM_BUG_ON_PAGE(obj_exts & MEMCG_DATA_KMEM, slab_page(slab));
578 #endif
579 return (struct slabobj_ext *)(obj_exts & ~OBJEXTS_FLAGS_MASK);
582 int alloc_slab_obj_exts(struct slab *slab, struct kmem_cache *s,
583 gfp_t gfp, bool new_slab);
585 #else /* CONFIG_SLAB_OBJ_EXT */
587 static inline struct slabobj_ext *slab_obj_exts(struct slab *slab)
589 return NULL;
592 #endif /* CONFIG_SLAB_OBJ_EXT */
594 static inline enum node_stat_item cache_vmstat_idx(struct kmem_cache *s)
596 return (s->flags & SLAB_RECLAIM_ACCOUNT) ?
597 NR_SLAB_RECLAIMABLE_B : NR_SLAB_UNRECLAIMABLE_B;
600 #ifdef CONFIG_MEMCG
601 bool __memcg_slab_post_alloc_hook(struct kmem_cache *s, struct list_lru *lru,
602 gfp_t flags, size_t size, void **p);
603 void __memcg_slab_free_hook(struct kmem_cache *s, struct slab *slab,
604 void **p, int objects, struct slabobj_ext *obj_exts);
605 #endif
607 size_t __ksize(const void *objp);
609 static inline size_t slab_ksize(const struct kmem_cache *s)
611 #ifdef CONFIG_SLUB_DEBUG
613 * Debugging requires use of the padding between object
614 * and whatever may come after it.
616 if (s->flags & (SLAB_RED_ZONE | SLAB_POISON))
617 return s->object_size;
618 #endif
619 if (s->flags & SLAB_KASAN)
620 return s->object_size;
622 * If we have the need to store the freelist pointer
623 * back there or track user information then we can
624 * only use the space before that information.
626 if (s->flags & (SLAB_TYPESAFE_BY_RCU | SLAB_STORE_USER))
627 return s->inuse;
629 * Else we can use all the padding etc for the allocation
631 return s->size;
634 #ifdef CONFIG_SLUB_DEBUG
635 void dump_unreclaimable_slab(void);
636 #else
637 static inline void dump_unreclaimable_slab(void)
640 #endif
642 void ___cache_free(struct kmem_cache *cache, void *x, unsigned long addr);
644 #ifdef CONFIG_SLAB_FREELIST_RANDOM
645 int cache_random_seq_create(struct kmem_cache *cachep, unsigned int count,
646 gfp_t gfp);
647 void cache_random_seq_destroy(struct kmem_cache *cachep);
648 #else
649 static inline int cache_random_seq_create(struct kmem_cache *cachep,
650 unsigned int count, gfp_t gfp)
652 return 0;
654 static inline void cache_random_seq_destroy(struct kmem_cache *cachep) { }
655 #endif /* CONFIG_SLAB_FREELIST_RANDOM */
657 static inline bool slab_want_init_on_alloc(gfp_t flags, struct kmem_cache *c)
659 if (static_branch_maybe(CONFIG_INIT_ON_ALLOC_DEFAULT_ON,
660 &init_on_alloc)) {
661 if (c->ctor)
662 return false;
663 if (c->flags & (SLAB_TYPESAFE_BY_RCU | SLAB_POISON))
664 return flags & __GFP_ZERO;
665 return true;
667 return flags & __GFP_ZERO;
670 static inline bool slab_want_init_on_free(struct kmem_cache *c)
672 if (static_branch_maybe(CONFIG_INIT_ON_FREE_DEFAULT_ON,
673 &init_on_free))
674 return !(c->ctor ||
675 (c->flags & (SLAB_TYPESAFE_BY_RCU | SLAB_POISON)));
676 return false;
679 #if defined(CONFIG_DEBUG_FS) && defined(CONFIG_SLUB_DEBUG)
680 void debugfs_slab_release(struct kmem_cache *);
681 #else
682 static inline void debugfs_slab_release(struct kmem_cache *s) { }
683 #endif
685 #ifdef CONFIG_PRINTK
686 #define KS_ADDRS_COUNT 16
687 struct kmem_obj_info {
688 void *kp_ptr;
689 struct slab *kp_slab;
690 void *kp_objp;
691 unsigned long kp_data_offset;
692 struct kmem_cache *kp_slab_cache;
693 void *kp_ret;
694 void *kp_stack[KS_ADDRS_COUNT];
695 void *kp_free_stack[KS_ADDRS_COUNT];
697 void __kmem_obj_info(struct kmem_obj_info *kpp, void *object, struct slab *slab);
698 #endif
700 void __check_heap_object(const void *ptr, unsigned long n,
701 const struct slab *slab, bool to_user);
703 static inline bool slub_debug_orig_size(struct kmem_cache *s)
705 return (kmem_cache_debug_flags(s, SLAB_STORE_USER) &&
706 (s->flags & SLAB_KMALLOC));
709 #ifdef CONFIG_SLUB_DEBUG
710 void skip_orig_size_check(struct kmem_cache *s, const void *object);
711 #endif
713 #endif /* MM_SLAB_H */