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_DESTROY_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 #define SLAB_DESTROY_BY_RCU 0x00080000UL /* Defer freeing slabs to RCU */
66 #define SLAB_MEM_SPREAD 0x00100000UL /* Spread some memory over cpuset */
67 #define SLAB_TRACE 0x00200000UL /* Trace allocations and frees */
69 /* Flag to prevent checks on free */
70 #ifdef CONFIG_DEBUG_OBJECTS
71 # define SLAB_DEBUG_OBJECTS 0x00400000UL
73 # define SLAB_DEBUG_OBJECTS 0x00000000UL
76 #define SLAB_NOLEAKTRACE 0x00800000UL /* Avoid kmemleak tracing */
78 /* Don't track use of uninitialized memory */
79 #ifdef CONFIG_KMEMCHECK
80 # define SLAB_NOTRACK 0x01000000UL
82 # define SLAB_NOTRACK 0x00000000UL
84 #ifdef CONFIG_FAILSLAB
85 # define SLAB_FAILSLAB 0x02000000UL /* Fault injection mark */
87 # define SLAB_FAILSLAB 0x00000000UL
89 #if defined(CONFIG_MEMCG) && !defined(CONFIG_SLOB)
90 # define SLAB_ACCOUNT 0x04000000UL /* Account to memcg */
92 # define SLAB_ACCOUNT 0x00000000UL
96 #define SLAB_KASAN 0x08000000UL
98 #define SLAB_KASAN 0x00000000UL
101 /* The following flags affect the page allocator grouping pages by mobility */
102 #define SLAB_RECLAIM_ACCOUNT 0x00020000UL /* Objects are reclaimable */
103 #define SLAB_TEMPORARY SLAB_RECLAIM_ACCOUNT /* Objects are short-lived */
105 * ZERO_SIZE_PTR will be returned for zero sized kmalloc requests.
107 * Dereferencing ZERO_SIZE_PTR will lead to a distinct access fault.
109 * ZERO_SIZE_PTR can be passed to kfree though in the same way that NULL can.
110 * Both make kfree a no-op.
112 #define ZERO_SIZE_PTR ((void *)16)
114 #define ZERO_OR_NULL_PTR(x) ((unsigned long)(x) <= \
115 (unsigned long)ZERO_SIZE_PTR)
117 #include <linux/kmemleak.h>
118 #include <linux/kasan.h>
122 * struct kmem_cache related prototypes
124 void __init
kmem_cache_init(void);
125 bool slab_is_available(void);
127 struct kmem_cache
*kmem_cache_create(const char *, size_t, size_t,
130 void kmem_cache_destroy(struct kmem_cache
*);
131 int kmem_cache_shrink(struct kmem_cache
*);
133 void memcg_create_kmem_cache(struct mem_cgroup
*, struct kmem_cache
*);
134 void memcg_deactivate_kmem_caches(struct mem_cgroup
*);
135 void memcg_destroy_kmem_caches(struct mem_cgroup
*);
138 * Please use this macro to create slab caches. Simply specify the
139 * name of the structure and maybe some flags that are listed above.
141 * The alignment of the struct determines object alignment. If you
142 * f.e. add ____cacheline_aligned_in_smp to the struct declaration
143 * then the objects will be properly aligned in SMP configurations.
145 #define KMEM_CACHE(__struct, __flags) kmem_cache_create(#__struct,\
146 sizeof(struct __struct), __alignof__(struct __struct),\
150 * Common kmalloc functions provided by all allocators
152 void * __must_check
__krealloc(const void *, size_t, gfp_t
);
153 void * __must_check
krealloc(const void *, size_t, gfp_t
);
154 void kfree(const void *);
155 void kzfree(const void *);
156 size_t ksize(const void *);
159 * Some archs want to perform DMA into kmalloc caches and need a guaranteed
160 * alignment larger than the alignment of a 64-bit integer.
161 * Setting ARCH_KMALLOC_MINALIGN in arch headers allows that.
163 #if defined(ARCH_DMA_MINALIGN) && ARCH_DMA_MINALIGN > 8
164 #define ARCH_KMALLOC_MINALIGN ARCH_DMA_MINALIGN
165 #define KMALLOC_MIN_SIZE ARCH_DMA_MINALIGN
166 #define KMALLOC_SHIFT_LOW ilog2(ARCH_DMA_MINALIGN)
168 #define ARCH_KMALLOC_MINALIGN __alignof__(unsigned long long)
172 * Setting ARCH_SLAB_MINALIGN in arch headers allows a different alignment.
173 * Intended for arches that get misalignment faults even for 64 bit integer
176 #ifndef ARCH_SLAB_MINALIGN
177 #define ARCH_SLAB_MINALIGN __alignof__(unsigned long long)
181 * kmalloc and friends return ARCH_KMALLOC_MINALIGN aligned
182 * pointers. kmem_cache_alloc and friends return ARCH_SLAB_MINALIGN
185 #define __assume_kmalloc_alignment __assume_aligned(ARCH_KMALLOC_MINALIGN)
186 #define __assume_slab_alignment __assume_aligned(ARCH_SLAB_MINALIGN)
187 #define __assume_page_alignment __assume_aligned(PAGE_SIZE)
190 * Kmalloc array related definitions
195 * The largest kmalloc size supported by the SLAB allocators is
196 * 32 megabyte (2^25) or the maximum allocatable page order if that is
199 * WARNING: Its not easy to increase this value since the allocators have
200 * to do various tricks to work around compiler limitations in order to
201 * ensure proper constant folding.
203 #define KMALLOC_SHIFT_HIGH ((MAX_ORDER + PAGE_SHIFT - 1) <= 25 ? \
204 (MAX_ORDER + PAGE_SHIFT - 1) : 25)
205 #define KMALLOC_SHIFT_MAX KMALLOC_SHIFT_HIGH
206 #ifndef KMALLOC_SHIFT_LOW
207 #define KMALLOC_SHIFT_LOW 5
213 * SLUB directly allocates requests fitting in to an order-1 page
214 * (PAGE_SIZE*2). Larger requests are passed to the page allocator.
216 #define KMALLOC_SHIFT_HIGH (PAGE_SHIFT + 1)
217 #define KMALLOC_SHIFT_MAX (MAX_ORDER + PAGE_SHIFT)
218 #ifndef KMALLOC_SHIFT_LOW
219 #define KMALLOC_SHIFT_LOW 3
225 * SLOB passes all requests larger than one page to the page allocator.
226 * No kmalloc array is necessary since objects of different sizes can
227 * be allocated from the same page.
229 #define KMALLOC_SHIFT_HIGH PAGE_SHIFT
230 #define KMALLOC_SHIFT_MAX 30
231 #ifndef KMALLOC_SHIFT_LOW
232 #define KMALLOC_SHIFT_LOW 3
236 /* Maximum allocatable size */
237 #define KMALLOC_MAX_SIZE (1UL << KMALLOC_SHIFT_MAX)
238 /* Maximum size for which we actually use a slab cache */
239 #define KMALLOC_MAX_CACHE_SIZE (1UL << KMALLOC_SHIFT_HIGH)
240 /* Maximum order allocatable via the slab allocagtor */
241 #define KMALLOC_MAX_ORDER (KMALLOC_SHIFT_MAX - PAGE_SHIFT)
246 #ifndef KMALLOC_MIN_SIZE
247 #define KMALLOC_MIN_SIZE (1 << KMALLOC_SHIFT_LOW)
251 * This restriction comes from byte sized index implementation.
252 * Page size is normally 2^12 bytes and, in this case, if we want to use
253 * byte sized index which can represent 2^8 entries, the size of the object
254 * should be equal or greater to 2^12 / 2^8 = 2^4 = 16.
255 * If minimum size of kmalloc is less than 16, we use it as minimum object
256 * size and give up to use byte sized index.
258 #define SLAB_OBJ_MIN_SIZE (KMALLOC_MIN_SIZE < 16 ? \
259 (KMALLOC_MIN_SIZE) : 16)
262 extern struct kmem_cache
*kmalloc_caches
[KMALLOC_SHIFT_HIGH
+ 1];
263 #ifdef CONFIG_ZONE_DMA
264 extern struct kmem_cache
*kmalloc_dma_caches
[KMALLOC_SHIFT_HIGH
+ 1];
268 * Figure out which kmalloc slab an allocation of a certain size
272 * 2 = 129 .. 192 bytes
273 * n = 2^(n-1)+1 .. 2^n
275 static __always_inline
int kmalloc_index(size_t size
)
280 if (size
<= KMALLOC_MIN_SIZE
)
281 return KMALLOC_SHIFT_LOW
;
283 if (KMALLOC_MIN_SIZE
<= 32 && size
> 64 && size
<= 96)
285 if (KMALLOC_MIN_SIZE
<= 64 && size
> 128 && size
<= 192)
287 if (size
<= 8) return 3;
288 if (size
<= 16) return 4;
289 if (size
<= 32) return 5;
290 if (size
<= 64) return 6;
291 if (size
<= 128) return 7;
292 if (size
<= 256) return 8;
293 if (size
<= 512) return 9;
294 if (size
<= 1024) return 10;
295 if (size
<= 2 * 1024) return 11;
296 if (size
<= 4 * 1024) return 12;
297 if (size
<= 8 * 1024) return 13;
298 if (size
<= 16 * 1024) return 14;
299 if (size
<= 32 * 1024) return 15;
300 if (size
<= 64 * 1024) return 16;
301 if (size
<= 128 * 1024) return 17;
302 if (size
<= 256 * 1024) return 18;
303 if (size
<= 512 * 1024) return 19;
304 if (size
<= 1024 * 1024) return 20;
305 if (size
<= 2 * 1024 * 1024) return 21;
306 if (size
<= 4 * 1024 * 1024) return 22;
307 if (size
<= 8 * 1024 * 1024) return 23;
308 if (size
<= 16 * 1024 * 1024) return 24;
309 if (size
<= 32 * 1024 * 1024) return 25;
310 if (size
<= 64 * 1024 * 1024) return 26;
313 /* Will never be reached. Needed because the compiler may complain */
316 #endif /* !CONFIG_SLOB */
318 void *__kmalloc(size_t size
, gfp_t flags
) __assume_kmalloc_alignment
;
319 void *kmem_cache_alloc(struct kmem_cache
*, gfp_t flags
) __assume_slab_alignment
;
320 void kmem_cache_free(struct kmem_cache
*, void *);
323 * Bulk allocation and freeing operations. These are accelerated in an
324 * allocator specific way to avoid taking locks repeatedly or building
325 * metadata structures unnecessarily.
327 * Note that interrupts must be enabled when calling these functions.
329 void kmem_cache_free_bulk(struct kmem_cache
*, size_t, void **);
330 int kmem_cache_alloc_bulk(struct kmem_cache
*, gfp_t
, size_t, void **);
333 * Caller must not use kfree_bulk() on memory not originally allocated
334 * by kmalloc(), because the SLOB allocator cannot handle this.
336 static __always_inline
void kfree_bulk(size_t size
, void **p
)
338 kmem_cache_free_bulk(NULL
, size
, p
);
342 void *__kmalloc_node(size_t size
, gfp_t flags
, int node
) __assume_kmalloc_alignment
;
343 void *kmem_cache_alloc_node(struct kmem_cache
*, gfp_t flags
, int node
) __assume_slab_alignment
;
345 static __always_inline
void *__kmalloc_node(size_t size
, gfp_t flags
, int node
)
347 return __kmalloc(size
, flags
);
350 static __always_inline
void *kmem_cache_alloc_node(struct kmem_cache
*s
, gfp_t flags
, int node
)
352 return kmem_cache_alloc(s
, flags
);
356 #ifdef CONFIG_TRACING
357 extern void *kmem_cache_alloc_trace(struct kmem_cache
*, gfp_t
, size_t) __assume_slab_alignment
;
360 extern void *kmem_cache_alloc_node_trace(struct kmem_cache
*s
,
362 int node
, size_t size
) __assume_slab_alignment
;
364 static __always_inline
void *
365 kmem_cache_alloc_node_trace(struct kmem_cache
*s
,
367 int node
, size_t size
)
369 return kmem_cache_alloc_trace(s
, gfpflags
, size
);
371 #endif /* CONFIG_NUMA */
373 #else /* CONFIG_TRACING */
374 static __always_inline
void *kmem_cache_alloc_trace(struct kmem_cache
*s
,
375 gfp_t flags
, size_t size
)
377 void *ret
= kmem_cache_alloc(s
, flags
);
379 kasan_kmalloc(s
, ret
, size
, flags
);
383 static __always_inline
void *
384 kmem_cache_alloc_node_trace(struct kmem_cache
*s
,
386 int node
, size_t size
)
388 void *ret
= kmem_cache_alloc_node(s
, gfpflags
, node
);
390 kasan_kmalloc(s
, ret
, size
, gfpflags
);
393 #endif /* CONFIG_TRACING */
395 extern void *kmalloc_order(size_t size
, gfp_t flags
, unsigned int order
) __assume_page_alignment
;
397 #ifdef CONFIG_TRACING
398 extern void *kmalloc_order_trace(size_t size
, gfp_t flags
, unsigned int order
) __assume_page_alignment
;
400 static __always_inline
void *
401 kmalloc_order_trace(size_t size
, gfp_t flags
, unsigned int order
)
403 return kmalloc_order(size
, flags
, order
);
407 static __always_inline
void *kmalloc_large(size_t size
, gfp_t flags
)
409 unsigned int order
= get_order(size
);
410 return kmalloc_order_trace(size
, flags
, order
);
414 * kmalloc - allocate memory
415 * @size: how many bytes of memory are required.
416 * @flags: the type of memory to allocate.
418 * kmalloc is the normal method of allocating memory
419 * for objects smaller than page size in the kernel.
421 * The @flags argument may be one of:
423 * %GFP_USER - Allocate memory on behalf of user. May sleep.
425 * %GFP_KERNEL - Allocate normal kernel ram. May sleep.
427 * %GFP_ATOMIC - Allocation will not sleep. May use emergency pools.
428 * For example, use this inside interrupt handlers.
430 * %GFP_HIGHUSER - Allocate pages from high memory.
432 * %GFP_NOIO - Do not do any I/O at all while trying to get memory.
434 * %GFP_NOFS - Do not make any fs calls while trying to get memory.
436 * %GFP_NOWAIT - Allocation will not sleep.
438 * %__GFP_THISNODE - Allocate node-local memory only.
440 * %GFP_DMA - Allocation suitable for DMA.
441 * Should only be used for kmalloc() caches. Otherwise, use a
442 * slab created with SLAB_DMA.
444 * Also it is possible to set different flags by OR'ing
445 * in one or more of the following additional @flags:
447 * %__GFP_COLD - Request cache-cold pages instead of
448 * trying to return cache-warm pages.
450 * %__GFP_HIGH - This allocation has high priority and may use emergency pools.
452 * %__GFP_NOFAIL - Indicate that this allocation is in no way allowed to fail
453 * (think twice before using).
455 * %__GFP_NORETRY - If memory is not immediately available,
456 * then give up at once.
458 * %__GFP_NOWARN - If allocation fails, don't issue any warnings.
460 * %__GFP_REPEAT - If allocation fails initially, try once more before failing.
462 * There are other flags available as well, but these are not intended
463 * for general use, and so are not documented here. For a full list of
464 * potential flags, always refer to linux/gfp.h.
466 static __always_inline
void *kmalloc(size_t size
, gfp_t flags
)
468 if (__builtin_constant_p(size
)) {
469 if (size
> KMALLOC_MAX_CACHE_SIZE
)
470 return kmalloc_large(size
, flags
);
472 if (!(flags
& GFP_DMA
)) {
473 int index
= kmalloc_index(size
);
476 return ZERO_SIZE_PTR
;
478 return kmem_cache_alloc_trace(kmalloc_caches
[index
],
483 return __kmalloc(size
, flags
);
487 * Determine size used for the nth kmalloc cache.
488 * return size or 0 if a kmalloc cache for that
489 * size does not exist
491 static __always_inline
int kmalloc_size(int n
)
497 if (n
== 1 && KMALLOC_MIN_SIZE
<= 32)
500 if (n
== 2 && KMALLOC_MIN_SIZE
<= 64)
506 static __always_inline
void *kmalloc_node(size_t size
, gfp_t flags
, int node
)
509 if (__builtin_constant_p(size
) &&
510 size
<= KMALLOC_MAX_CACHE_SIZE
&& !(flags
& GFP_DMA
)) {
511 int i
= kmalloc_index(size
);
514 return ZERO_SIZE_PTR
;
516 return kmem_cache_alloc_node_trace(kmalloc_caches
[i
],
520 return __kmalloc_node(size
, flags
, node
);
523 struct memcg_cache_array
{
525 struct kmem_cache
*entries
[0];
529 * This is the main placeholder for memcg-related information in kmem caches.
530 * Both the root cache and the child caches will have it. For the root cache,
531 * this will hold a dynamically allocated array large enough to hold
532 * information about the currently limited memcgs in the system. To allow the
533 * array to be accessed without taking any locks, on relocation we free the old
534 * version only after a grace period.
536 * Child caches will hold extra metadata needed for its operation. Fields are:
538 * @memcg: pointer to the memcg this cache belongs to
539 * @root_cache: pointer to the global, root cache, this cache was derived from
541 * Both root and child caches of the same kind are linked into a list chained
544 struct memcg_cache_params
{
546 struct list_head list
;
548 struct memcg_cache_array __rcu
*memcg_caches
;
550 struct mem_cgroup
*memcg
;
551 struct kmem_cache
*root_cache
;
556 int memcg_update_all_caches(int num_memcgs
);
559 * kmalloc_array - allocate memory for an array.
560 * @n: number of elements.
561 * @size: element size.
562 * @flags: the type of memory to allocate (see kmalloc).
564 static inline void *kmalloc_array(size_t n
, size_t size
, gfp_t flags
)
566 if (size
!= 0 && n
> SIZE_MAX
/ size
)
568 return __kmalloc(n
* size
, flags
);
572 * kcalloc - allocate memory for an array. The memory is set to zero.
573 * @n: number of elements.
574 * @size: element size.
575 * @flags: the type of memory to allocate (see kmalloc).
577 static inline void *kcalloc(size_t n
, size_t size
, gfp_t flags
)
579 return kmalloc_array(n
, size
, flags
| __GFP_ZERO
);
583 * kmalloc_track_caller is a special version of kmalloc that records the
584 * calling function of the routine calling it for slab leak tracking instead
585 * of just the calling function (confusing, eh?).
586 * It's useful when the call to kmalloc comes from a widely-used standard
587 * allocator where we care about the real place the memory allocation
588 * request comes from.
590 extern void *__kmalloc_track_caller(size_t, gfp_t
, unsigned long);
591 #define kmalloc_track_caller(size, flags) \
592 __kmalloc_track_caller(size, flags, _RET_IP_)
595 extern void *__kmalloc_node_track_caller(size_t, gfp_t
, int, unsigned long);
596 #define kmalloc_node_track_caller(size, flags, node) \
597 __kmalloc_node_track_caller(size, flags, node, \
600 #else /* CONFIG_NUMA */
602 #define kmalloc_node_track_caller(size, flags, node) \
603 kmalloc_track_caller(size, flags)
605 #endif /* CONFIG_NUMA */
610 static inline void *kmem_cache_zalloc(struct kmem_cache
*k
, gfp_t flags
)
612 return kmem_cache_alloc(k
, flags
| __GFP_ZERO
);
616 * kzalloc - allocate memory. The memory is set to zero.
617 * @size: how many bytes of memory are required.
618 * @flags: the type of memory to allocate (see kmalloc).
620 static inline void *kzalloc(size_t size
, gfp_t flags
)
622 return kmalloc(size
, flags
| __GFP_ZERO
);
626 * kzalloc_node - allocate zeroed memory from a particular memory node.
627 * @size: how many bytes of memory are required.
628 * @flags: the type of memory to allocate (see kmalloc).
629 * @node: memory node from which to allocate
631 static inline void *kzalloc_node(size_t size
, gfp_t flags
, int node
)
633 return kmalloc_node(size
, flags
| __GFP_ZERO
, node
);
636 unsigned int kmem_cache_size(struct kmem_cache
*s
);
637 void __init
kmem_cache_init_late(void);
639 #endif /* _LINUX_SLAB_H */