2 * Written by Mark Hemment, 1996 (markhe@nextd.demon.co.uk).
4 * (C) SGI 2006, Christoph Lameter
5 * Cleaned up and restructured to ease the addition of alternative
6 * implementations of SLAB allocators.
7 * (C) Linux Foundation 2008-2013
8 * Unified interface for all slab allocators
14 #include <linux/gfp.h>
15 #include <linux/types.h>
16 #include <linux/workqueue.h>
20 * Flags to pass to kmem_cache_create().
21 * The ones marked DEBUG are only valid if CONFIG_DEBUG_SLAB is set.
23 #define SLAB_CONSISTENCY_CHECKS 0x00000100UL /* DEBUG: Perform (expensive) checks on alloc/free */
24 #define SLAB_RED_ZONE 0x00000400UL /* DEBUG: Red zone objs in a cache */
25 #define SLAB_POISON 0x00000800UL /* DEBUG: Poison objects */
26 #define SLAB_HWCACHE_ALIGN 0x00002000UL /* Align objs on cache lines */
27 #define SLAB_CACHE_DMA 0x00004000UL /* Use GFP_DMA memory */
28 #define SLAB_STORE_USER 0x00010000UL /* DEBUG: Store the last owner for bug hunting */
29 #define SLAB_PANIC 0x00040000UL /* Panic if kmem_cache_create() fails */
31 * SLAB_TYPESAFE_BY_RCU - **WARNING** READ THIS!
33 * This delays freeing the SLAB page by a grace period, it does _NOT_
34 * delay object freeing. This means that if you do kmem_cache_free()
35 * that memory location is free to be reused at any time. Thus it may
36 * be possible to see another object there in the same RCU grace period.
38 * This feature only ensures the memory location backing the object
39 * stays valid, the trick to using this is relying on an independent
40 * object validation pass. Something like:
44 * obj = lockless_lookup(key);
46 * if (!try_get_ref(obj)) // might fail for free objects
49 * if (obj->key != key) { // not the object we expected
56 * This is useful if we need to approach a kernel structure obliquely,
57 * from its address obtained without the usual locking. We can lock
58 * the structure to stabilize it and check it's still at the given address,
59 * only if we can be sure that the memory has not been meanwhile reused
60 * for some other kind of object (which our subsystem's lock might corrupt).
62 * rcu_read_lock before reading the address, then rcu_read_unlock after
63 * taking the spinlock within the structure expected at that address.
65 * Note that SLAB_TYPESAFE_BY_RCU was originally named SLAB_DESTROY_BY_RCU.
67 #define SLAB_TYPESAFE_BY_RCU 0x00080000UL /* Defer freeing slabs to RCU */
68 #define SLAB_MEM_SPREAD 0x00100000UL /* Spread some memory over cpuset */
69 #define SLAB_TRACE 0x00200000UL /* Trace allocations and frees */
71 /* Flag to prevent checks on free */
72 #ifdef CONFIG_DEBUG_OBJECTS
73 # define SLAB_DEBUG_OBJECTS 0x00400000UL
75 # define SLAB_DEBUG_OBJECTS 0x00000000UL
78 #define SLAB_NOLEAKTRACE 0x00800000UL /* Avoid kmemleak tracing */
80 /* Don't track use of uninitialized memory */
81 #ifdef CONFIG_KMEMCHECK
82 # define SLAB_NOTRACK 0x01000000UL
84 # define SLAB_NOTRACK 0x00000000UL
86 #ifdef CONFIG_FAILSLAB
87 # define SLAB_FAILSLAB 0x02000000UL /* Fault injection mark */
89 # define SLAB_FAILSLAB 0x00000000UL
91 #if defined(CONFIG_MEMCG) && !defined(CONFIG_SLOB)
92 # define SLAB_ACCOUNT 0x04000000UL /* Account to memcg */
94 # define SLAB_ACCOUNT 0x00000000UL
98 #define SLAB_KASAN 0x08000000UL
100 #define SLAB_KASAN 0x00000000UL
103 /* The following flags affect the page allocator grouping pages by mobility */
104 #define SLAB_RECLAIM_ACCOUNT 0x00020000UL /* Objects are reclaimable */
105 #define SLAB_TEMPORARY SLAB_RECLAIM_ACCOUNT /* Objects are short-lived */
107 * ZERO_SIZE_PTR will be returned for zero sized kmalloc requests.
109 * Dereferencing ZERO_SIZE_PTR will lead to a distinct access fault.
111 * ZERO_SIZE_PTR can be passed to kfree though in the same way that NULL can.
112 * Both make kfree a no-op.
114 #define ZERO_SIZE_PTR ((void *)16)
116 #define ZERO_OR_NULL_PTR(x) ((unsigned long)(x) <= \
117 (unsigned long)ZERO_SIZE_PTR)
119 #include <linux/kmemleak.h>
120 #include <linux/kasan.h>
124 * struct kmem_cache related prototypes
126 void __init
kmem_cache_init(void);
127 bool slab_is_available(void);
129 struct kmem_cache
*kmem_cache_create(const char *, size_t, size_t,
132 void kmem_cache_destroy(struct kmem_cache
*);
133 int kmem_cache_shrink(struct kmem_cache
*);
135 void memcg_create_kmem_cache(struct mem_cgroup
*, struct kmem_cache
*);
136 void memcg_deactivate_kmem_caches(struct mem_cgroup
*);
137 void memcg_destroy_kmem_caches(struct mem_cgroup
*);
140 * Please use this macro to create slab caches. Simply specify the
141 * name of the structure and maybe some flags that are listed above.
143 * The alignment of the struct determines object alignment. If you
144 * f.e. add ____cacheline_aligned_in_smp to the struct declaration
145 * then the objects will be properly aligned in SMP configurations.
147 #define KMEM_CACHE(__struct, __flags) kmem_cache_create(#__struct,\
148 sizeof(struct __struct), __alignof__(struct __struct),\
152 * Common kmalloc functions provided by all allocators
154 void * __must_check
__krealloc(const void *, size_t, gfp_t
);
155 void * __must_check
krealloc(const void *, size_t, gfp_t
);
156 void kfree(const void *);
157 void kzfree(const void *);
158 size_t ksize(const void *);
160 #ifdef CONFIG_HAVE_HARDENED_USERCOPY_ALLOCATOR
161 const char *__check_heap_object(const void *ptr
, unsigned long n
,
164 static inline const char *__check_heap_object(const void *ptr
,
173 * Some archs want to perform DMA into kmalloc caches and need a guaranteed
174 * alignment larger than the alignment of a 64-bit integer.
175 * Setting ARCH_KMALLOC_MINALIGN in arch headers allows that.
177 #if defined(ARCH_DMA_MINALIGN) && ARCH_DMA_MINALIGN > 8
178 #define ARCH_KMALLOC_MINALIGN ARCH_DMA_MINALIGN
179 #define KMALLOC_MIN_SIZE ARCH_DMA_MINALIGN
180 #define KMALLOC_SHIFT_LOW ilog2(ARCH_DMA_MINALIGN)
182 #define ARCH_KMALLOC_MINALIGN __alignof__(unsigned long long)
186 * Setting ARCH_SLAB_MINALIGN in arch headers allows a different alignment.
187 * Intended for arches that get misalignment faults even for 64 bit integer
190 #ifndef ARCH_SLAB_MINALIGN
191 #define ARCH_SLAB_MINALIGN __alignof__(unsigned long long)
195 * kmalloc and friends return ARCH_KMALLOC_MINALIGN aligned
196 * pointers. kmem_cache_alloc and friends return ARCH_SLAB_MINALIGN
199 #define __assume_kmalloc_alignment __assume_aligned(ARCH_KMALLOC_MINALIGN)
200 #define __assume_slab_alignment __assume_aligned(ARCH_SLAB_MINALIGN)
201 #define __assume_page_alignment __assume_aligned(PAGE_SIZE)
204 * Kmalloc array related definitions
209 * The largest kmalloc size supported by the SLAB allocators is
210 * 32 megabyte (2^25) or the maximum allocatable page order if that is
213 * WARNING: Its not easy to increase this value since the allocators have
214 * to do various tricks to work around compiler limitations in order to
215 * ensure proper constant folding.
217 #define KMALLOC_SHIFT_HIGH ((MAX_ORDER + PAGE_SHIFT - 1) <= 25 ? \
218 (MAX_ORDER + PAGE_SHIFT - 1) : 25)
219 #define KMALLOC_SHIFT_MAX KMALLOC_SHIFT_HIGH
220 #ifndef KMALLOC_SHIFT_LOW
221 #define KMALLOC_SHIFT_LOW 5
227 * SLUB directly allocates requests fitting in to an order-1 page
228 * (PAGE_SIZE*2). Larger requests are passed to the page allocator.
230 #define KMALLOC_SHIFT_HIGH (PAGE_SHIFT + 1)
231 #define KMALLOC_SHIFT_MAX (MAX_ORDER + PAGE_SHIFT - 1)
232 #ifndef KMALLOC_SHIFT_LOW
233 #define KMALLOC_SHIFT_LOW 3
239 * SLOB passes all requests larger than one page to the page allocator.
240 * No kmalloc array is necessary since objects of different sizes can
241 * be allocated from the same page.
243 #define KMALLOC_SHIFT_HIGH PAGE_SHIFT
244 #define KMALLOC_SHIFT_MAX (MAX_ORDER + PAGE_SHIFT - 1)
245 #ifndef KMALLOC_SHIFT_LOW
246 #define KMALLOC_SHIFT_LOW 3
250 /* Maximum allocatable size */
251 #define KMALLOC_MAX_SIZE (1UL << KMALLOC_SHIFT_MAX)
252 /* Maximum size for which we actually use a slab cache */
253 #define KMALLOC_MAX_CACHE_SIZE (1UL << KMALLOC_SHIFT_HIGH)
254 /* Maximum order allocatable via the slab allocagtor */
255 #define KMALLOC_MAX_ORDER (KMALLOC_SHIFT_MAX - PAGE_SHIFT)
260 #ifndef KMALLOC_MIN_SIZE
261 #define KMALLOC_MIN_SIZE (1 << KMALLOC_SHIFT_LOW)
265 * This restriction comes from byte sized index implementation.
266 * Page size is normally 2^12 bytes and, in this case, if we want to use
267 * byte sized index which can represent 2^8 entries, the size of the object
268 * should be equal or greater to 2^12 / 2^8 = 2^4 = 16.
269 * If minimum size of kmalloc is less than 16, we use it as minimum object
270 * size and give up to use byte sized index.
272 #define SLAB_OBJ_MIN_SIZE (KMALLOC_MIN_SIZE < 16 ? \
273 (KMALLOC_MIN_SIZE) : 16)
276 extern struct kmem_cache
*kmalloc_caches
[KMALLOC_SHIFT_HIGH
+ 1];
277 #ifdef CONFIG_ZONE_DMA
278 extern struct kmem_cache
*kmalloc_dma_caches
[KMALLOC_SHIFT_HIGH
+ 1];
282 * Figure out which kmalloc slab an allocation of a certain size
286 * 2 = 129 .. 192 bytes
287 * n = 2^(n-1)+1 .. 2^n
289 static __always_inline
int kmalloc_index(size_t size
)
294 if (size
<= KMALLOC_MIN_SIZE
)
295 return KMALLOC_SHIFT_LOW
;
297 if (KMALLOC_MIN_SIZE
<= 32 && size
> 64 && size
<= 96)
299 if (KMALLOC_MIN_SIZE
<= 64 && size
> 128 && size
<= 192)
301 if (size
<= 8) return 3;
302 if (size
<= 16) return 4;
303 if (size
<= 32) return 5;
304 if (size
<= 64) return 6;
305 if (size
<= 128) return 7;
306 if (size
<= 256) return 8;
307 if (size
<= 512) return 9;
308 if (size
<= 1024) return 10;
309 if (size
<= 2 * 1024) return 11;
310 if (size
<= 4 * 1024) return 12;
311 if (size
<= 8 * 1024) return 13;
312 if (size
<= 16 * 1024) return 14;
313 if (size
<= 32 * 1024) return 15;
314 if (size
<= 64 * 1024) return 16;
315 if (size
<= 128 * 1024) return 17;
316 if (size
<= 256 * 1024) return 18;
317 if (size
<= 512 * 1024) return 19;
318 if (size
<= 1024 * 1024) return 20;
319 if (size
<= 2 * 1024 * 1024) return 21;
320 if (size
<= 4 * 1024 * 1024) return 22;
321 if (size
<= 8 * 1024 * 1024) return 23;
322 if (size
<= 16 * 1024 * 1024) return 24;
323 if (size
<= 32 * 1024 * 1024) return 25;
324 if (size
<= 64 * 1024 * 1024) return 26;
327 /* Will never be reached. Needed because the compiler may complain */
330 #endif /* !CONFIG_SLOB */
332 void *__kmalloc(size_t size
, gfp_t flags
) __assume_kmalloc_alignment __malloc
;
333 void *kmem_cache_alloc(struct kmem_cache
*, gfp_t flags
) __assume_slab_alignment __malloc
;
334 void kmem_cache_free(struct kmem_cache
*, void *);
337 * Bulk allocation and freeing operations. These are accelerated in an
338 * allocator specific way to avoid taking locks repeatedly or building
339 * metadata structures unnecessarily.
341 * Note that interrupts must be enabled when calling these functions.
343 void kmem_cache_free_bulk(struct kmem_cache
*, size_t, void **);
344 int kmem_cache_alloc_bulk(struct kmem_cache
*, gfp_t
, size_t, void **);
347 * Caller must not use kfree_bulk() on memory not originally allocated
348 * by kmalloc(), because the SLOB allocator cannot handle this.
350 static __always_inline
void kfree_bulk(size_t size
, void **p
)
352 kmem_cache_free_bulk(NULL
, size
, p
);
356 void *__kmalloc_node(size_t size
, gfp_t flags
, int node
) __assume_kmalloc_alignment __malloc
;
357 void *kmem_cache_alloc_node(struct kmem_cache
*, gfp_t flags
, int node
) __assume_slab_alignment __malloc
;
359 static __always_inline
void *__kmalloc_node(size_t size
, gfp_t flags
, int node
)
361 return __kmalloc(size
, flags
);
364 static __always_inline
void *kmem_cache_alloc_node(struct kmem_cache
*s
, gfp_t flags
, int node
)
366 return kmem_cache_alloc(s
, flags
);
370 #ifdef CONFIG_TRACING
371 extern void *kmem_cache_alloc_trace(struct kmem_cache
*, gfp_t
, size_t) __assume_slab_alignment __malloc
;
374 extern void *kmem_cache_alloc_node_trace(struct kmem_cache
*s
,
376 int node
, size_t size
) __assume_slab_alignment __malloc
;
378 static __always_inline
void *
379 kmem_cache_alloc_node_trace(struct kmem_cache
*s
,
381 int node
, size_t size
)
383 return kmem_cache_alloc_trace(s
, gfpflags
, size
);
385 #endif /* CONFIG_NUMA */
387 #else /* CONFIG_TRACING */
388 static __always_inline
void *kmem_cache_alloc_trace(struct kmem_cache
*s
,
389 gfp_t flags
, size_t size
)
391 void *ret
= kmem_cache_alloc(s
, flags
);
393 kasan_kmalloc(s
, ret
, size
, flags
);
397 static __always_inline
void *
398 kmem_cache_alloc_node_trace(struct kmem_cache
*s
,
400 int node
, size_t size
)
402 void *ret
= kmem_cache_alloc_node(s
, gfpflags
, node
);
404 kasan_kmalloc(s
, ret
, size
, gfpflags
);
407 #endif /* CONFIG_TRACING */
409 extern void *kmalloc_order(size_t size
, gfp_t flags
, unsigned int order
) __assume_page_alignment __malloc
;
411 #ifdef CONFIG_TRACING
412 extern void *kmalloc_order_trace(size_t size
, gfp_t flags
, unsigned int order
) __assume_page_alignment __malloc
;
414 static __always_inline
void *
415 kmalloc_order_trace(size_t size
, gfp_t flags
, unsigned int order
)
417 return kmalloc_order(size
, flags
, order
);
421 static __always_inline
void *kmalloc_large(size_t size
, gfp_t flags
)
423 unsigned int order
= get_order(size
);
424 return kmalloc_order_trace(size
, flags
, order
);
428 * kmalloc - allocate memory
429 * @size: how many bytes of memory are required.
430 * @flags: the type of memory to allocate.
432 * kmalloc is the normal method of allocating memory
433 * for objects smaller than page size in the kernel.
435 * The @flags argument may be one of:
437 * %GFP_USER - Allocate memory on behalf of user. May sleep.
439 * %GFP_KERNEL - Allocate normal kernel ram. May sleep.
441 * %GFP_ATOMIC - Allocation will not sleep. May use emergency pools.
442 * For example, use this inside interrupt handlers.
444 * %GFP_HIGHUSER - Allocate pages from high memory.
446 * %GFP_NOIO - Do not do any I/O at all while trying to get memory.
448 * %GFP_NOFS - Do not make any fs calls while trying to get memory.
450 * %GFP_NOWAIT - Allocation will not sleep.
452 * %__GFP_THISNODE - Allocate node-local memory only.
454 * %GFP_DMA - Allocation suitable for DMA.
455 * Should only be used for kmalloc() caches. Otherwise, use a
456 * slab created with SLAB_DMA.
458 * Also it is possible to set different flags by OR'ing
459 * in one or more of the following additional @flags:
461 * %__GFP_COLD - Request cache-cold pages instead of
462 * trying to return cache-warm pages.
464 * %__GFP_HIGH - This allocation has high priority and may use emergency pools.
466 * %__GFP_NOFAIL - Indicate that this allocation is in no way allowed to fail
467 * (think twice before using).
469 * %__GFP_NORETRY - If memory is not immediately available,
470 * then give up at once.
472 * %__GFP_NOWARN - If allocation fails, don't issue any warnings.
474 * %__GFP_RETRY_MAYFAIL - Try really hard to succeed the allocation but fail
477 * There are other flags available as well, but these are not intended
478 * for general use, and so are not documented here. For a full list of
479 * potential flags, always refer to linux/gfp.h.
481 static __always_inline
void *kmalloc(size_t size
, gfp_t flags
)
483 if (__builtin_constant_p(size
)) {
484 if (size
> KMALLOC_MAX_CACHE_SIZE
)
485 return kmalloc_large(size
, flags
);
487 if (!(flags
& GFP_DMA
)) {
488 int index
= kmalloc_index(size
);
491 return ZERO_SIZE_PTR
;
493 return kmem_cache_alloc_trace(kmalloc_caches
[index
],
498 return __kmalloc(size
, flags
);
502 * Determine size used for the nth kmalloc cache.
503 * return size or 0 if a kmalloc cache for that
504 * size does not exist
506 static __always_inline
int kmalloc_size(int n
)
512 if (n
== 1 && KMALLOC_MIN_SIZE
<= 32)
515 if (n
== 2 && KMALLOC_MIN_SIZE
<= 64)
521 static __always_inline
void *kmalloc_node(size_t size
, gfp_t flags
, int node
)
524 if (__builtin_constant_p(size
) &&
525 size
<= KMALLOC_MAX_CACHE_SIZE
&& !(flags
& GFP_DMA
)) {
526 int i
= kmalloc_index(size
);
529 return ZERO_SIZE_PTR
;
531 return kmem_cache_alloc_node_trace(kmalloc_caches
[i
],
535 return __kmalloc_node(size
, flags
, node
);
538 struct memcg_cache_array
{
540 struct kmem_cache
*entries
[0];
544 * This is the main placeholder for memcg-related information in kmem caches.
545 * Both the root cache and the child caches will have it. For the root cache,
546 * this will hold a dynamically allocated array large enough to hold
547 * information about the currently limited memcgs in the system. To allow the
548 * array to be accessed without taking any locks, on relocation we free the old
549 * version only after a grace period.
551 * Root and child caches hold different metadata.
553 * @root_cache: Common to root and child caches. NULL for root, pointer to
554 * the root cache for children.
556 * The following fields are specific to root caches.
558 * @memcg_caches: kmemcg ID indexed table of child caches. This table is
559 * used to index child cachces during allocation and cleared
560 * early during shutdown.
562 * @root_caches_node: List node for slab_root_caches list.
564 * @children: List of all child caches. While the child caches are also
565 * reachable through @memcg_caches, a child cache remains on
566 * this list until it is actually destroyed.
568 * The following fields are specific to child caches.
570 * @memcg: Pointer to the memcg this cache belongs to.
572 * @children_node: List node for @root_cache->children list.
574 * @kmem_caches_node: List node for @memcg->kmem_caches list.
576 struct memcg_cache_params
{
577 struct kmem_cache
*root_cache
;
580 struct memcg_cache_array __rcu
*memcg_caches
;
581 struct list_head __root_caches_node
;
582 struct list_head children
;
585 struct mem_cgroup
*memcg
;
586 struct list_head children_node
;
587 struct list_head kmem_caches_node
;
589 void (*deact_fn
)(struct kmem_cache
*);
591 struct rcu_head deact_rcu_head
;
592 struct work_struct deact_work
;
598 int memcg_update_all_caches(int num_memcgs
);
601 * kmalloc_array - allocate memory for an array.
602 * @n: number of elements.
603 * @size: element size.
604 * @flags: the type of memory to allocate (see kmalloc).
606 static inline void *kmalloc_array(size_t n
, size_t size
, gfp_t flags
)
608 if (size
!= 0 && n
> SIZE_MAX
/ size
)
610 if (__builtin_constant_p(n
) && __builtin_constant_p(size
))
611 return kmalloc(n
* size
, flags
);
612 return __kmalloc(n
* size
, flags
);
616 * kcalloc - allocate memory for an array. The memory is set to zero.
617 * @n: number of elements.
618 * @size: element size.
619 * @flags: the type of memory to allocate (see kmalloc).
621 static inline void *kcalloc(size_t n
, size_t size
, gfp_t flags
)
623 return kmalloc_array(n
, size
, flags
| __GFP_ZERO
);
627 * kmalloc_track_caller is a special version of kmalloc that records the
628 * calling function of the routine calling it for slab leak tracking instead
629 * of just the calling function (confusing, eh?).
630 * It's useful when the call to kmalloc comes from a widely-used standard
631 * allocator where we care about the real place the memory allocation
632 * request comes from.
634 extern void *__kmalloc_track_caller(size_t, gfp_t
, unsigned long);
635 #define kmalloc_track_caller(size, flags) \
636 __kmalloc_track_caller(size, flags, _RET_IP_)
639 extern void *__kmalloc_node_track_caller(size_t, gfp_t
, int, unsigned long);
640 #define kmalloc_node_track_caller(size, flags, node) \
641 __kmalloc_node_track_caller(size, flags, node, \
644 #else /* CONFIG_NUMA */
646 #define kmalloc_node_track_caller(size, flags, node) \
647 kmalloc_track_caller(size, flags)
649 #endif /* CONFIG_NUMA */
654 static inline void *kmem_cache_zalloc(struct kmem_cache
*k
, gfp_t flags
)
656 return kmem_cache_alloc(k
, flags
| __GFP_ZERO
);
660 * kzalloc - allocate memory. The memory is set to zero.
661 * @size: how many bytes of memory are required.
662 * @flags: the type of memory to allocate (see kmalloc).
664 static inline void *kzalloc(size_t size
, gfp_t flags
)
666 return kmalloc(size
, flags
| __GFP_ZERO
);
670 * kzalloc_node - allocate zeroed memory from a particular memory node.
671 * @size: how many bytes of memory are required.
672 * @flags: the type of memory to allocate (see kmalloc).
673 * @node: memory node from which to allocate
675 static inline void *kzalloc_node(size_t size
, gfp_t flags
, int node
)
677 return kmalloc_node(size
, flags
| __GFP_ZERO
, node
);
680 unsigned int kmem_cache_size(struct kmem_cache
*s
);
681 void __init
kmem_cache_init_late(void);
683 #if defined(CONFIG_SMP) && defined(CONFIG_SLAB)
684 int slab_prepare_cpu(unsigned int cpu
);
685 int slab_dead_cpu(unsigned int cpu
);
687 #define slab_prepare_cpu NULL
688 #define slab_dead_cpu NULL
691 #endif /* _LINUX_SLAB_H */