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_SLAB_DEBUG is set.
23 #define SLAB_DEBUG_FREE 0x00000100UL /* DEBUG: Perform (expensive) checks on 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
90 /* The following flags affect the page allocator grouping pages by mobility */
91 #define SLAB_RECLAIM_ACCOUNT 0x00020000UL /* Objects are reclaimable */
92 #define SLAB_TEMPORARY SLAB_RECLAIM_ACCOUNT /* Objects are short-lived */
94 * ZERO_SIZE_PTR will be returned for zero sized kmalloc requests.
96 * Dereferencing ZERO_SIZE_PTR will lead to a distinct access fault.
98 * ZERO_SIZE_PTR can be passed to kfree though in the same way that NULL can.
99 * Both make kfree a no-op.
101 #define ZERO_SIZE_PTR ((void *)16)
103 #define ZERO_OR_NULL_PTR(x) ((unsigned long)(x) <= \
104 (unsigned long)ZERO_SIZE_PTR)
106 #include <linux/kmemleak.h>
107 #include <linux/kasan.h>
111 * struct kmem_cache related prototypes
113 void __init kmem_cache_init(void);
114 int slab_is_available(void);
116 struct kmem_cache *kmem_cache_create(const char *, size_t, size_t,
119 void kmem_cache_destroy(struct kmem_cache *);
120 int kmem_cache_shrink(struct kmem_cache *);
122 void memcg_create_kmem_cache(struct mem_cgroup *, struct kmem_cache *);
123 void memcg_deactivate_kmem_caches(struct mem_cgroup *);
124 void memcg_destroy_kmem_caches(struct mem_cgroup *);
127 * Please use this macro to create slab caches. Simply specify the
128 * name of the structure and maybe some flags that are listed above.
130 * The alignment of the struct determines object alignment. If you
131 * f.e. add ____cacheline_aligned_in_smp to the struct declaration
132 * then the objects will be properly aligned in SMP configurations.
134 #define KMEM_CACHE(__struct, __flags) kmem_cache_create(#__struct,\
135 sizeof(struct __struct), __alignof__(struct __struct),\
139 * Common kmalloc functions provided by all allocators
141 void * __must_check __krealloc(const void *, size_t, gfp_t);
142 void * __must_check krealloc(const void *, size_t, gfp_t);
143 void kfree(const void *);
144 void kzfree(const void *);
145 size_t ksize(const void *);
148 * Some archs want to perform DMA into kmalloc caches and need a guaranteed
149 * alignment larger than the alignment of a 64-bit integer.
150 * Setting ARCH_KMALLOC_MINALIGN in arch headers allows that.
152 #if defined(ARCH_DMA_MINALIGN) && ARCH_DMA_MINALIGN > 8
153 #define ARCH_KMALLOC_MINALIGN ARCH_DMA_MINALIGN
154 #define KMALLOC_MIN_SIZE ARCH_DMA_MINALIGN
155 #define KMALLOC_SHIFT_LOW ilog2(ARCH_DMA_MINALIGN)
157 #define ARCH_KMALLOC_MINALIGN __alignof__(unsigned long long)
161 * Kmalloc array related definitions
166 * The largest kmalloc size supported by the SLAB allocators is
167 * 32 megabyte (2^25) or the maximum allocatable page order if that is
170 * WARNING: Its not easy to increase this value since the allocators have
171 * to do various tricks to work around compiler limitations in order to
172 * ensure proper constant folding.
174 #define KMALLOC_SHIFT_HIGH ((MAX_ORDER + PAGE_SHIFT - 1) <= 25 ? \
175 (MAX_ORDER + PAGE_SHIFT - 1) : 25)
176 #define KMALLOC_SHIFT_MAX KMALLOC_SHIFT_HIGH
177 #ifndef KMALLOC_SHIFT_LOW
178 #define KMALLOC_SHIFT_LOW 5
184 * SLUB directly allocates requests fitting in to an order-1 page
185 * (PAGE_SIZE*2). Larger requests are passed to the page allocator.
187 #define KMALLOC_SHIFT_HIGH (PAGE_SHIFT + 1)
188 #define KMALLOC_SHIFT_MAX (MAX_ORDER + PAGE_SHIFT)
189 #ifndef KMALLOC_SHIFT_LOW
190 #define KMALLOC_SHIFT_LOW 3
196 * SLOB passes all requests larger than one page to the page allocator.
197 * No kmalloc array is necessary since objects of different sizes can
198 * be allocated from the same page.
200 #define KMALLOC_SHIFT_HIGH PAGE_SHIFT
201 #define KMALLOC_SHIFT_MAX 30
202 #ifndef KMALLOC_SHIFT_LOW
203 #define KMALLOC_SHIFT_LOW 3
207 /* Maximum allocatable size */
208 #define KMALLOC_MAX_SIZE (1UL << KMALLOC_SHIFT_MAX)
209 /* Maximum size for which we actually use a slab cache */
210 #define KMALLOC_MAX_CACHE_SIZE (1UL << KMALLOC_SHIFT_HIGH)
211 /* Maximum order allocatable via the slab allocagtor */
212 #define KMALLOC_MAX_ORDER (KMALLOC_SHIFT_MAX - PAGE_SHIFT)
217 #ifndef KMALLOC_MIN_SIZE
218 #define KMALLOC_MIN_SIZE (1 << KMALLOC_SHIFT_LOW)
222 * This restriction comes from byte sized index implementation.
223 * Page size is normally 2^12 bytes and, in this case, if we want to use
224 * byte sized index which can represent 2^8 entries, the size of the object
225 * should be equal or greater to 2^12 / 2^8 = 2^4 = 16.
226 * If minimum size of kmalloc is less than 16, we use it as minimum object
227 * size and give up to use byte sized index.
229 #define SLAB_OBJ_MIN_SIZE (KMALLOC_MIN_SIZE < 16 ? \
230 (KMALLOC_MIN_SIZE) : 16)
233 extern struct kmem_cache *kmalloc_caches[KMALLOC_SHIFT_HIGH + 1];
234 #ifdef CONFIG_ZONE_DMA
235 extern struct kmem_cache *kmalloc_dma_caches[KMALLOC_SHIFT_HIGH + 1];
239 * Figure out which kmalloc slab an allocation of a certain size
243 * 2 = 120 .. 192 bytes
244 * n = 2^(n-1) .. 2^n -1
246 static __always_inline int kmalloc_index(size_t size)
251 if (size <= KMALLOC_MIN_SIZE)
252 return KMALLOC_SHIFT_LOW;
254 if (KMALLOC_MIN_SIZE <= 32 && size > 64 && size <= 96)
256 if (KMALLOC_MIN_SIZE <= 64 && size > 128 && size <= 192)
258 if (size <= 8) return 3;
259 if (size <= 16) return 4;
260 if (size <= 32) return 5;
261 if (size <= 64) return 6;
262 if (size <= 128) return 7;
263 if (size <= 256) return 8;
264 if (size <= 512) return 9;
265 if (size <= 1024) return 10;
266 if (size <= 2 * 1024) return 11;
267 if (size <= 4 * 1024) return 12;
268 if (size <= 8 * 1024) return 13;
269 if (size <= 16 * 1024) return 14;
270 if (size <= 32 * 1024) return 15;
271 if (size <= 64 * 1024) return 16;
272 if (size <= 128 * 1024) return 17;
273 if (size <= 256 * 1024) return 18;
274 if (size <= 512 * 1024) return 19;
275 if (size <= 1024 * 1024) return 20;
276 if (size <= 2 * 1024 * 1024) return 21;
277 if (size <= 4 * 1024 * 1024) return 22;
278 if (size <= 8 * 1024 * 1024) return 23;
279 if (size <= 16 * 1024 * 1024) return 24;
280 if (size <= 32 * 1024 * 1024) return 25;
281 if (size <= 64 * 1024 * 1024) return 26;
284 /* Will never be reached. Needed because the compiler may complain */
287 #endif /* !CONFIG_SLOB */
289 void *__kmalloc(size_t size, gfp_t flags);
290 void *kmem_cache_alloc(struct kmem_cache *, gfp_t flags);
291 void kmem_cache_free(struct kmem_cache *, void *);
294 void *__kmalloc_node(size_t size, gfp_t flags, int node);
295 void *kmem_cache_alloc_node(struct kmem_cache *, gfp_t flags, int node);
297 static __always_inline void *__kmalloc_node(size_t size, gfp_t flags, int node)
299 return __kmalloc(size, flags);
302 static __always_inline void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t flags, int node)
304 return kmem_cache_alloc(s, flags);
308 #ifdef CONFIG_TRACING
309 extern void *kmem_cache_alloc_trace(struct kmem_cache *, gfp_t, size_t);
312 extern void *kmem_cache_alloc_node_trace(struct kmem_cache *s,
314 int node, size_t size);
316 static __always_inline void *
317 kmem_cache_alloc_node_trace(struct kmem_cache *s,
319 int node, size_t size)
321 return kmem_cache_alloc_trace(s, gfpflags, size);
323 #endif /* CONFIG_NUMA */
325 #else /* CONFIG_TRACING */
326 static __always_inline void *kmem_cache_alloc_trace(struct kmem_cache *s,
327 gfp_t flags, size_t size)
329 void *ret = kmem_cache_alloc(s, flags);
331 kasan_kmalloc(s, ret, size);
335 static __always_inline void *
336 kmem_cache_alloc_node_trace(struct kmem_cache *s,
338 int node, size_t size)
340 void *ret = kmem_cache_alloc_node(s, gfpflags, node);
342 kasan_kmalloc(s, ret, size);
345 #endif /* CONFIG_TRACING */
347 extern void *kmalloc_order(size_t size, gfp_t flags, unsigned int order);
349 #ifdef CONFIG_TRACING
350 extern void *kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order);
352 static __always_inline void *
353 kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order)
355 return kmalloc_order(size, flags, order);
359 static __always_inline void *kmalloc_large(size_t size, gfp_t flags)
361 unsigned int order = get_order(size);
362 return kmalloc_order_trace(size, flags, order);
366 * kmalloc - allocate memory
367 * @size: how many bytes of memory are required.
368 * @flags: the type of memory to allocate.
370 * kmalloc is the normal method of allocating memory
371 * for objects smaller than page size in the kernel.
373 * The @flags argument may be one of:
375 * %GFP_USER - Allocate memory on behalf of user. May sleep.
377 * %GFP_KERNEL - Allocate normal kernel ram. May sleep.
379 * %GFP_ATOMIC - Allocation will not sleep. May use emergency pools.
380 * For example, use this inside interrupt handlers.
382 * %GFP_HIGHUSER - Allocate pages from high memory.
384 * %GFP_NOIO - Do not do any I/O at all while trying to get memory.
386 * %GFP_NOFS - Do not make any fs calls while trying to get memory.
388 * %GFP_NOWAIT - Allocation will not sleep.
390 * %__GFP_THISNODE - Allocate node-local memory only.
392 * %GFP_DMA - Allocation suitable for DMA.
393 * Should only be used for kmalloc() caches. Otherwise, use a
394 * slab created with SLAB_DMA.
396 * Also it is possible to set different flags by OR'ing
397 * in one or more of the following additional @flags:
399 * %__GFP_COLD - Request cache-cold pages instead of
400 * trying to return cache-warm pages.
402 * %__GFP_HIGH - This allocation has high priority and may use emergency pools.
404 * %__GFP_NOFAIL - Indicate that this allocation is in no way allowed to fail
405 * (think twice before using).
407 * %__GFP_NORETRY - If memory is not immediately available,
408 * then give up at once.
410 * %__GFP_NOWARN - If allocation fails, don't issue any warnings.
412 * %__GFP_REPEAT - If allocation fails initially, try once more before failing.
414 * There are other flags available as well, but these are not intended
415 * for general use, and so are not documented here. For a full list of
416 * potential flags, always refer to linux/gfp.h.
418 static __always_inline void *kmalloc(size_t size, gfp_t flags)
420 if (__builtin_constant_p(size)) {
421 if (size > KMALLOC_MAX_CACHE_SIZE)
422 return kmalloc_large(size, flags);
424 if (!(flags & GFP_DMA)) {
425 int index = kmalloc_index(size);
428 return ZERO_SIZE_PTR;
430 return kmem_cache_alloc_trace(kmalloc_caches[index],
435 return __kmalloc(size, flags);
439 * Determine size used for the nth kmalloc cache.
440 * return size or 0 if a kmalloc cache for that
441 * size does not exist
443 static __always_inline int kmalloc_size(int n)
449 if (n == 1 && KMALLOC_MIN_SIZE <= 32)
452 if (n == 2 && KMALLOC_MIN_SIZE <= 64)
458 static __always_inline void *kmalloc_node(size_t size, gfp_t flags, int node)
461 if (__builtin_constant_p(size) &&
462 size <= KMALLOC_MAX_CACHE_SIZE && !(flags & GFP_DMA)) {
463 int i = kmalloc_index(size);
466 return ZERO_SIZE_PTR;
468 return kmem_cache_alloc_node_trace(kmalloc_caches[i],
472 return __kmalloc_node(size, flags, node);
476 * Setting ARCH_SLAB_MINALIGN in arch headers allows a different alignment.
477 * Intended for arches that get misalignment faults even for 64 bit integer
480 #ifndef ARCH_SLAB_MINALIGN
481 #define ARCH_SLAB_MINALIGN __alignof__(unsigned long long)
484 struct memcg_cache_array {
486 struct kmem_cache *entries[0];
490 * This is the main placeholder for memcg-related information in kmem caches.
491 * Both the root cache and the child caches will have it. For the root cache,
492 * this will hold a dynamically allocated array large enough to hold
493 * information about the currently limited memcgs in the system. To allow the
494 * array to be accessed without taking any locks, on relocation we free the old
495 * version only after a grace period.
497 * Child caches will hold extra metadata needed for its operation. Fields are:
499 * @memcg: pointer to the memcg this cache belongs to
500 * @root_cache: pointer to the global, root cache, this cache was derived from
502 * Both root and child caches of the same kind are linked into a list chained
505 struct memcg_cache_params {
507 struct list_head list;
509 struct memcg_cache_array __rcu *memcg_caches;
511 struct mem_cgroup *memcg;
512 struct kmem_cache *root_cache;
517 int memcg_update_all_caches(int num_memcgs);
520 * kmalloc_array - allocate memory for an array.
521 * @n: number of elements.
522 * @size: element size.
523 * @flags: the type of memory to allocate (see kmalloc).
525 static inline void *kmalloc_array(size_t n, size_t size, gfp_t flags)
527 if (size != 0 && n > SIZE_MAX / size)
529 return __kmalloc(n * size, flags);
533 * kcalloc - allocate memory for an array. The memory is set to zero.
534 * @n: number of elements.
535 * @size: element size.
536 * @flags: the type of memory to allocate (see kmalloc).
538 static inline void *kcalloc(size_t n, size_t size, gfp_t flags)
540 return kmalloc_array(n, size, flags | __GFP_ZERO);
544 * kmalloc_track_caller is a special version of kmalloc that records the
545 * calling function of the routine calling it for slab leak tracking instead
546 * of just the calling function (confusing, eh?).
547 * It's useful when the call to kmalloc comes from a widely-used standard
548 * allocator where we care about the real place the memory allocation
549 * request comes from.
551 extern void *__kmalloc_track_caller(size_t, gfp_t, unsigned long);
552 #define kmalloc_track_caller(size, flags) \
553 __kmalloc_track_caller(size, flags, _RET_IP_)
556 extern void *__kmalloc_node_track_caller(size_t, gfp_t, int, unsigned long);
557 #define kmalloc_node_track_caller(size, flags, node) \
558 __kmalloc_node_track_caller(size, flags, node, \
561 #else /* CONFIG_NUMA */
563 #define kmalloc_node_track_caller(size, flags, node) \
564 kmalloc_track_caller(size, flags)
566 #endif /* CONFIG_NUMA */
571 static inline void *kmem_cache_zalloc(struct kmem_cache *k, gfp_t flags)
573 return kmem_cache_alloc(k, flags | __GFP_ZERO);
577 * kzalloc - allocate memory. The memory is set to zero.
578 * @size: how many bytes of memory are required.
579 * @flags: the type of memory to allocate (see kmalloc).
581 static inline void *kzalloc(size_t size, gfp_t flags)
583 return kmalloc(size, flags | __GFP_ZERO);
587 * kzalloc_node - allocate zeroed memory from a particular memory node.
588 * @size: how many bytes of memory are required.
589 * @flags: the type of memory to allocate (see kmalloc).
590 * @node: memory node from which to allocate
592 static inline void *kzalloc_node(size_t size, gfp_t flags, int node)
594 return kmalloc_node(size, flags | __GFP_ZERO, node);
597 unsigned int kmem_cache_size(struct kmem_cache *s);
598 void __init kmem_cache_init_late(void);
600 #endif /* _LINUX_SLAB_H */