1 /* SPDX-License-Identifier: GPL-2.0 */
3 * Written by Mark Hemment, 1996 (markhe@nextd.demon.co.uk).
5 * (C) SGI 2006, Christoph Lameter
6 * Cleaned up and restructured to ease the addition of alternative
7 * implementations of SLAB allocators.
8 * (C) Linux Foundation 2008-2013
9 * Unified interface for all slab allocators
15 #include <linux/gfp.h>
16 #include <linux/types.h>
17 #include <linux/workqueue.h>
21 * Flags to pass to kmem_cache_create().
22 * The ones marked DEBUG are only valid if CONFIG_DEBUG_SLAB is set.
24 /* DEBUG: Perform (expensive) checks on alloc/free */
25 #define SLAB_CONSISTENCY_CHECKS ((slab_flags_t __force)0x00000100U)
26 /* DEBUG: Red zone objs in a cache */
27 #define SLAB_RED_ZONE ((slab_flags_t __force)0x00000400U)
28 /* DEBUG: Poison objects */
29 #define SLAB_POISON ((slab_flags_t __force)0x00000800U)
30 /* Align objs on cache lines */
31 #define SLAB_HWCACHE_ALIGN ((slab_flags_t __force)0x00002000U)
32 /* Use GFP_DMA memory */
33 #define SLAB_CACHE_DMA ((slab_flags_t __force)0x00004000U)
34 /* DEBUG: Store the last owner for bug hunting */
35 #define SLAB_STORE_USER ((slab_flags_t __force)0x00010000U)
36 /* Panic if kmem_cache_create() fails */
37 #define SLAB_PANIC ((slab_flags_t __force)0x00040000U)
39 * SLAB_TYPESAFE_BY_RCU - **WARNING** READ THIS!
41 * This delays freeing the SLAB page by a grace period, it does _NOT_
42 * delay object freeing. This means that if you do kmem_cache_free()
43 * that memory location is free to be reused at any time. Thus it may
44 * be possible to see another object there in the same RCU grace period.
46 * This feature only ensures the memory location backing the object
47 * stays valid, the trick to using this is relying on an independent
48 * object validation pass. Something like:
52 * obj = lockless_lookup(key);
54 * if (!try_get_ref(obj)) // might fail for free objects
57 * if (obj->key != key) { // not the object we expected
64 * This is useful if we need to approach a kernel structure obliquely,
65 * from its address obtained without the usual locking. We can lock
66 * the structure to stabilize it and check it's still at the given address,
67 * only if we can be sure that the memory has not been meanwhile reused
68 * for some other kind of object (which our subsystem's lock might corrupt).
70 * rcu_read_lock before reading the address, then rcu_read_unlock after
71 * taking the spinlock within the structure expected at that address.
73 * Note that SLAB_TYPESAFE_BY_RCU was originally named SLAB_DESTROY_BY_RCU.
75 /* Defer freeing slabs to RCU */
76 #define SLAB_TYPESAFE_BY_RCU ((slab_flags_t __force)0x00080000U)
77 /* Spread some memory over cpuset */
78 #define SLAB_MEM_SPREAD ((slab_flags_t __force)0x00100000U)
79 /* Trace allocations and frees */
80 #define SLAB_TRACE ((slab_flags_t __force)0x00200000U)
82 /* Flag to prevent checks on free */
83 #ifdef CONFIG_DEBUG_OBJECTS
84 # define SLAB_DEBUG_OBJECTS ((slab_flags_t __force)0x00400000U)
86 # define SLAB_DEBUG_OBJECTS 0
89 /* Avoid kmemleak tracing */
90 #define SLAB_NOLEAKTRACE ((slab_flags_t __force)0x00800000U)
92 /* Fault injection mark */
93 #ifdef CONFIG_FAILSLAB
94 # define SLAB_FAILSLAB ((slab_flags_t __force)0x02000000U)
96 # define SLAB_FAILSLAB 0
98 /* Account to memcg */
99 #if defined(CONFIG_MEMCG) && !defined(CONFIG_SLOB)
100 # define SLAB_ACCOUNT ((slab_flags_t __force)0x04000000U)
102 # define SLAB_ACCOUNT 0
106 #define SLAB_KASAN ((slab_flags_t __force)0x08000000U)
111 /* The following flags affect the page allocator grouping pages by mobility */
112 /* Objects are reclaimable */
113 #define SLAB_RECLAIM_ACCOUNT ((slab_flags_t __force)0x00020000U)
114 #define SLAB_TEMPORARY SLAB_RECLAIM_ACCOUNT /* Objects are short-lived */
116 * ZERO_SIZE_PTR will be returned for zero sized kmalloc requests.
118 * Dereferencing ZERO_SIZE_PTR will lead to a distinct access fault.
120 * ZERO_SIZE_PTR can be passed to kfree though in the same way that NULL can.
121 * Both make kfree a no-op.
123 #define ZERO_SIZE_PTR ((void *)16)
125 #define ZERO_OR_NULL_PTR(x) ((unsigned long)(x) <= \
126 (unsigned long)ZERO_SIZE_PTR)
128 #include <linux/kmemleak.h>
129 #include <linux/kasan.h>
133 * struct kmem_cache related prototypes
135 void __init kmem_cache_init(void);
136 bool slab_is_available(void);
138 struct kmem_cache *kmem_cache_create(const char *, size_t, size_t,
141 void kmem_cache_destroy(struct kmem_cache *);
142 int kmem_cache_shrink(struct kmem_cache *);
144 void memcg_create_kmem_cache(struct mem_cgroup *, struct kmem_cache *);
145 void memcg_deactivate_kmem_caches(struct mem_cgroup *);
146 void memcg_destroy_kmem_caches(struct mem_cgroup *);
149 * Please use this macro to create slab caches. Simply specify the
150 * name of the structure and maybe some flags that are listed above.
152 * The alignment of the struct determines object alignment. If you
153 * f.e. add ____cacheline_aligned_in_smp to the struct declaration
154 * then the objects will be properly aligned in SMP configurations.
156 #define KMEM_CACHE(__struct, __flags) kmem_cache_create(#__struct,\
157 sizeof(struct __struct), __alignof__(struct __struct),\
161 * Common kmalloc functions provided by all allocators
163 void * __must_check __krealloc(const void *, size_t, gfp_t);
164 void * __must_check krealloc(const void *, size_t, gfp_t);
165 void kfree(const void *);
166 void kzfree(const void *);
167 size_t ksize(const void *);
169 #ifdef CONFIG_HAVE_HARDENED_USERCOPY_ALLOCATOR
170 const char *__check_heap_object(const void *ptr, unsigned long n,
173 static inline const char *__check_heap_object(const void *ptr,
182 * Some archs want to perform DMA into kmalloc caches and need a guaranteed
183 * alignment larger than the alignment of a 64-bit integer.
184 * Setting ARCH_KMALLOC_MINALIGN in arch headers allows that.
186 #if defined(ARCH_DMA_MINALIGN) && ARCH_DMA_MINALIGN > 8
187 #define ARCH_KMALLOC_MINALIGN ARCH_DMA_MINALIGN
188 #define KMALLOC_MIN_SIZE ARCH_DMA_MINALIGN
189 #define KMALLOC_SHIFT_LOW ilog2(ARCH_DMA_MINALIGN)
191 #define ARCH_KMALLOC_MINALIGN __alignof__(unsigned long long)
195 * Setting ARCH_SLAB_MINALIGN in arch headers allows a different alignment.
196 * Intended for arches that get misalignment faults even for 64 bit integer
199 #ifndef ARCH_SLAB_MINALIGN
200 #define ARCH_SLAB_MINALIGN __alignof__(unsigned long long)
204 * kmalloc and friends return ARCH_KMALLOC_MINALIGN aligned
205 * pointers. kmem_cache_alloc and friends return ARCH_SLAB_MINALIGN
208 #define __assume_kmalloc_alignment __assume_aligned(ARCH_KMALLOC_MINALIGN)
209 #define __assume_slab_alignment __assume_aligned(ARCH_SLAB_MINALIGN)
210 #define __assume_page_alignment __assume_aligned(PAGE_SIZE)
213 * Kmalloc array related definitions
218 * The largest kmalloc size supported by the SLAB allocators is
219 * 32 megabyte (2^25) or the maximum allocatable page order if that is
222 * WARNING: Its not easy to increase this value since the allocators have
223 * to do various tricks to work around compiler limitations in order to
224 * ensure proper constant folding.
226 #define KMALLOC_SHIFT_HIGH ((MAX_ORDER + PAGE_SHIFT - 1) <= 25 ? \
227 (MAX_ORDER + PAGE_SHIFT - 1) : 25)
228 #define KMALLOC_SHIFT_MAX KMALLOC_SHIFT_HIGH
229 #ifndef KMALLOC_SHIFT_LOW
230 #define KMALLOC_SHIFT_LOW 5
236 * SLUB directly allocates requests fitting in to an order-1 page
237 * (PAGE_SIZE*2). Larger requests are passed to the page allocator.
239 #define KMALLOC_SHIFT_HIGH (PAGE_SHIFT + 1)
240 #define KMALLOC_SHIFT_MAX (MAX_ORDER + PAGE_SHIFT - 1)
241 #ifndef KMALLOC_SHIFT_LOW
242 #define KMALLOC_SHIFT_LOW 3
248 * SLOB passes all requests larger than one page to the page allocator.
249 * No kmalloc array is necessary since objects of different sizes can
250 * be allocated from the same page.
252 #define KMALLOC_SHIFT_HIGH PAGE_SHIFT
253 #define KMALLOC_SHIFT_MAX (MAX_ORDER + PAGE_SHIFT - 1)
254 #ifndef KMALLOC_SHIFT_LOW
255 #define KMALLOC_SHIFT_LOW 3
259 /* Maximum allocatable size */
260 #define KMALLOC_MAX_SIZE (1UL << KMALLOC_SHIFT_MAX)
261 /* Maximum size for which we actually use a slab cache */
262 #define KMALLOC_MAX_CACHE_SIZE (1UL << KMALLOC_SHIFT_HIGH)
263 /* Maximum order allocatable via the slab allocagtor */
264 #define KMALLOC_MAX_ORDER (KMALLOC_SHIFT_MAX - PAGE_SHIFT)
269 #ifndef KMALLOC_MIN_SIZE
270 #define KMALLOC_MIN_SIZE (1 << KMALLOC_SHIFT_LOW)
274 * This restriction comes from byte sized index implementation.
275 * Page size is normally 2^12 bytes and, in this case, if we want to use
276 * byte sized index which can represent 2^8 entries, the size of the object
277 * should be equal or greater to 2^12 / 2^8 = 2^4 = 16.
278 * If minimum size of kmalloc is less than 16, we use it as minimum object
279 * size and give up to use byte sized index.
281 #define SLAB_OBJ_MIN_SIZE (KMALLOC_MIN_SIZE < 16 ? \
282 (KMALLOC_MIN_SIZE) : 16)
285 extern struct kmem_cache *kmalloc_caches[KMALLOC_SHIFT_HIGH + 1];
286 #ifdef CONFIG_ZONE_DMA
287 extern struct kmem_cache *kmalloc_dma_caches[KMALLOC_SHIFT_HIGH + 1];
291 * Figure out which kmalloc slab an allocation of a certain size
295 * 2 = 129 .. 192 bytes
296 * n = 2^(n-1)+1 .. 2^n
298 static __always_inline int kmalloc_index(size_t size)
303 if (size <= KMALLOC_MIN_SIZE)
304 return KMALLOC_SHIFT_LOW;
306 if (KMALLOC_MIN_SIZE <= 32 && size > 64 && size <= 96)
308 if (KMALLOC_MIN_SIZE <= 64 && size > 128 && size <= 192)
310 if (size <= 8) return 3;
311 if (size <= 16) return 4;
312 if (size <= 32) return 5;
313 if (size <= 64) return 6;
314 if (size <= 128) return 7;
315 if (size <= 256) return 8;
316 if (size <= 512) return 9;
317 if (size <= 1024) return 10;
318 if (size <= 2 * 1024) return 11;
319 if (size <= 4 * 1024) return 12;
320 if (size <= 8 * 1024) return 13;
321 if (size <= 16 * 1024) return 14;
322 if (size <= 32 * 1024) return 15;
323 if (size <= 64 * 1024) return 16;
324 if (size <= 128 * 1024) return 17;
325 if (size <= 256 * 1024) return 18;
326 if (size <= 512 * 1024) return 19;
327 if (size <= 1024 * 1024) return 20;
328 if (size <= 2 * 1024 * 1024) return 21;
329 if (size <= 4 * 1024 * 1024) return 22;
330 if (size <= 8 * 1024 * 1024) return 23;
331 if (size <= 16 * 1024 * 1024) return 24;
332 if (size <= 32 * 1024 * 1024) return 25;
333 if (size <= 64 * 1024 * 1024) return 26;
336 /* Will never be reached. Needed because the compiler may complain */
339 #endif /* !CONFIG_SLOB */
341 void *__kmalloc(size_t size, gfp_t flags) __assume_kmalloc_alignment __malloc;
342 void *kmem_cache_alloc(struct kmem_cache *, gfp_t flags) __assume_slab_alignment __malloc;
343 void kmem_cache_free(struct kmem_cache *, void *);
346 * Bulk allocation and freeing operations. These are accelerated in an
347 * allocator specific way to avoid taking locks repeatedly or building
348 * metadata structures unnecessarily.
350 * Note that interrupts must be enabled when calling these functions.
352 void kmem_cache_free_bulk(struct kmem_cache *, size_t, void **);
353 int kmem_cache_alloc_bulk(struct kmem_cache *, gfp_t, size_t, void **);
356 * Caller must not use kfree_bulk() on memory not originally allocated
357 * by kmalloc(), because the SLOB allocator cannot handle this.
359 static __always_inline void kfree_bulk(size_t size, void **p)
361 kmem_cache_free_bulk(NULL, size, p);
365 void *__kmalloc_node(size_t size, gfp_t flags, int node) __assume_kmalloc_alignment __malloc;
366 void *kmem_cache_alloc_node(struct kmem_cache *, gfp_t flags, int node) __assume_slab_alignment __malloc;
368 static __always_inline void *__kmalloc_node(size_t size, gfp_t flags, int node)
370 return __kmalloc(size, flags);
373 static __always_inline void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t flags, int node)
375 return kmem_cache_alloc(s, flags);
379 #ifdef CONFIG_TRACING
380 extern void *kmem_cache_alloc_trace(struct kmem_cache *, gfp_t, size_t) __assume_slab_alignment __malloc;
383 extern void *kmem_cache_alloc_node_trace(struct kmem_cache *s,
385 int node, size_t size) __assume_slab_alignment __malloc;
387 static __always_inline void *
388 kmem_cache_alloc_node_trace(struct kmem_cache *s,
390 int node, size_t size)
392 return kmem_cache_alloc_trace(s, gfpflags, size);
394 #endif /* CONFIG_NUMA */
396 #else /* CONFIG_TRACING */
397 static __always_inline void *kmem_cache_alloc_trace(struct kmem_cache *s,
398 gfp_t flags, size_t size)
400 void *ret = kmem_cache_alloc(s, flags);
402 kasan_kmalloc(s, ret, size, flags);
406 static __always_inline void *
407 kmem_cache_alloc_node_trace(struct kmem_cache *s,
409 int node, size_t size)
411 void *ret = kmem_cache_alloc_node(s, gfpflags, node);
413 kasan_kmalloc(s, ret, size, gfpflags);
416 #endif /* CONFIG_TRACING */
418 extern void *kmalloc_order(size_t size, gfp_t flags, unsigned int order) __assume_page_alignment __malloc;
420 #ifdef CONFIG_TRACING
421 extern void *kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order) __assume_page_alignment __malloc;
423 static __always_inline void *
424 kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order)
426 return kmalloc_order(size, flags, order);
430 static __always_inline void *kmalloc_large(size_t size, gfp_t flags)
432 unsigned int order = get_order(size);
433 return kmalloc_order_trace(size, flags, order);
437 * kmalloc - allocate memory
438 * @size: how many bytes of memory are required.
439 * @flags: the type of memory to allocate.
441 * kmalloc is the normal method of allocating memory
442 * for objects smaller than page size in the kernel.
444 * The @flags argument may be one of:
446 * %GFP_USER - Allocate memory on behalf of user. May sleep.
448 * %GFP_KERNEL - Allocate normal kernel ram. May sleep.
450 * %GFP_ATOMIC - Allocation will not sleep. May use emergency pools.
451 * For example, use this inside interrupt handlers.
453 * %GFP_HIGHUSER - Allocate pages from high memory.
455 * %GFP_NOIO - Do not do any I/O at all while trying to get memory.
457 * %GFP_NOFS - Do not make any fs calls while trying to get memory.
459 * %GFP_NOWAIT - Allocation will not sleep.
461 * %__GFP_THISNODE - Allocate node-local memory only.
463 * %GFP_DMA - Allocation suitable for DMA.
464 * Should only be used for kmalloc() caches. Otherwise, use a
465 * slab created with SLAB_DMA.
467 * Also it is possible to set different flags by OR'ing
468 * in one or more of the following additional @flags:
470 * %__GFP_HIGH - This allocation has high priority and may use emergency pools.
472 * %__GFP_NOFAIL - Indicate that this allocation is in no way allowed to fail
473 * (think twice before using).
475 * %__GFP_NORETRY - If memory is not immediately available,
476 * then give up at once.
478 * %__GFP_NOWARN - If allocation fails, don't issue any warnings.
480 * %__GFP_RETRY_MAYFAIL - Try really hard to succeed the allocation but fail
483 * There are other flags available as well, but these are not intended
484 * for general use, and so are not documented here. For a full list of
485 * potential flags, always refer to linux/gfp.h.
487 static __always_inline void *kmalloc(size_t size, gfp_t flags)
489 if (__builtin_constant_p(size)) {
490 if (size > KMALLOC_MAX_CACHE_SIZE)
491 return kmalloc_large(size, flags);
493 if (!(flags & GFP_DMA)) {
494 int index = kmalloc_index(size);
497 return ZERO_SIZE_PTR;
499 return kmem_cache_alloc_trace(kmalloc_caches[index],
504 return __kmalloc(size, flags);
508 * Determine size used for the nth kmalloc cache.
509 * return size or 0 if a kmalloc cache for that
510 * size does not exist
512 static __always_inline int kmalloc_size(int n)
518 if (n == 1 && KMALLOC_MIN_SIZE <= 32)
521 if (n == 2 && KMALLOC_MIN_SIZE <= 64)
527 static __always_inline void *kmalloc_node(size_t size, gfp_t flags, int node)
530 if (__builtin_constant_p(size) &&
531 size <= KMALLOC_MAX_CACHE_SIZE && !(flags & GFP_DMA)) {
532 int i = kmalloc_index(size);
535 return ZERO_SIZE_PTR;
537 return kmem_cache_alloc_node_trace(kmalloc_caches[i],
541 return __kmalloc_node(size, flags, node);
544 struct memcg_cache_array {
546 struct kmem_cache *entries[0];
550 * This is the main placeholder for memcg-related information in kmem caches.
551 * Both the root cache and the child caches will have it. For the root cache,
552 * this will hold a dynamically allocated array large enough to hold
553 * information about the currently limited memcgs in the system. To allow the
554 * array to be accessed without taking any locks, on relocation we free the old
555 * version only after a grace period.
557 * Root and child caches hold different metadata.
559 * @root_cache: Common to root and child caches. NULL for root, pointer to
560 * the root cache for children.
562 * The following fields are specific to root caches.
564 * @memcg_caches: kmemcg ID indexed table of child caches. This table is
565 * used to index child cachces during allocation and cleared
566 * early during shutdown.
568 * @root_caches_node: List node for slab_root_caches list.
570 * @children: List of all child caches. While the child caches are also
571 * reachable through @memcg_caches, a child cache remains on
572 * this list until it is actually destroyed.
574 * The following fields are specific to child caches.
576 * @memcg: Pointer to the memcg this cache belongs to.
578 * @children_node: List node for @root_cache->children list.
580 * @kmem_caches_node: List node for @memcg->kmem_caches list.
582 struct memcg_cache_params {
583 struct kmem_cache *root_cache;
586 struct memcg_cache_array __rcu *memcg_caches;
587 struct list_head __root_caches_node;
588 struct list_head children;
591 struct mem_cgroup *memcg;
592 struct list_head children_node;
593 struct list_head kmem_caches_node;
595 void (*deact_fn)(struct kmem_cache *);
597 struct rcu_head deact_rcu_head;
598 struct work_struct deact_work;
604 int memcg_update_all_caches(int num_memcgs);
607 * kmalloc_array - allocate memory for an array.
608 * @n: number of elements.
609 * @size: element size.
610 * @flags: the type of memory to allocate (see kmalloc).
612 static inline void *kmalloc_array(size_t n, size_t size, gfp_t flags)
614 if (size != 0 && n > SIZE_MAX / size)
616 if (__builtin_constant_p(n) && __builtin_constant_p(size))
617 return kmalloc(n * size, flags);
618 return __kmalloc(n * size, flags);
622 * kcalloc - allocate memory for an array. The memory is set to zero.
623 * @n: number of elements.
624 * @size: element size.
625 * @flags: the type of memory to allocate (see kmalloc).
627 static inline void *kcalloc(size_t n, size_t size, gfp_t flags)
629 return kmalloc_array(n, size, flags | __GFP_ZERO);
633 * kmalloc_track_caller is a special version of kmalloc that records the
634 * calling function of the routine calling it for slab leak tracking instead
635 * of just the calling function (confusing, eh?).
636 * It's useful when the call to kmalloc comes from a widely-used standard
637 * allocator where we care about the real place the memory allocation
638 * request comes from.
640 extern void *__kmalloc_track_caller(size_t, gfp_t, unsigned long);
641 #define kmalloc_track_caller(size, flags) \
642 __kmalloc_track_caller(size, flags, _RET_IP_)
644 static inline void *kmalloc_array_node(size_t n, size_t size, gfp_t flags,
647 if (size != 0 && n > SIZE_MAX / size)
649 if (__builtin_constant_p(n) && __builtin_constant_p(size))
650 return kmalloc_node(n * size, flags, node);
651 return __kmalloc_node(n * size, flags, node);
654 static inline void *kcalloc_node(size_t n, size_t size, gfp_t flags, int node)
656 return kmalloc_array_node(n, size, flags | __GFP_ZERO, node);
661 extern void *__kmalloc_node_track_caller(size_t, gfp_t, int, unsigned long);
662 #define kmalloc_node_track_caller(size, flags, node) \
663 __kmalloc_node_track_caller(size, flags, node, \
666 #else /* CONFIG_NUMA */
668 #define kmalloc_node_track_caller(size, flags, node) \
669 kmalloc_track_caller(size, flags)
671 #endif /* CONFIG_NUMA */
676 static inline void *kmem_cache_zalloc(struct kmem_cache *k, gfp_t flags)
678 return kmem_cache_alloc(k, flags | __GFP_ZERO);
682 * kzalloc - allocate memory. The memory is set to zero.
683 * @size: how many bytes of memory are required.
684 * @flags: the type of memory to allocate (see kmalloc).
686 static inline void *kzalloc(size_t size, gfp_t flags)
688 return kmalloc(size, flags | __GFP_ZERO);
692 * kzalloc_node - allocate zeroed memory from a particular memory node.
693 * @size: how many bytes of memory are required.
694 * @flags: the type of memory to allocate (see kmalloc).
695 * @node: memory node from which to allocate
697 static inline void *kzalloc_node(size_t size, gfp_t flags, int node)
699 return kmalloc_node(size, flags | __GFP_ZERO, node);
702 unsigned int kmem_cache_size(struct kmem_cache *s);
703 void __init kmem_cache_init_late(void);
705 #if defined(CONFIG_SMP) && defined(CONFIG_SLAB)
706 int slab_prepare_cpu(unsigned int cpu);
707 int slab_dead_cpu(unsigned int cpu);
709 #define slab_prepare_cpu NULL
710 #define slab_dead_cpu NULL
713 #endif /* _LINUX_SLAB_H */