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_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
89 #if defined(CONFIG_MEMCG) && !defined(CONFIG_SLOB)
90 # define SLAB_ACCOUNT 0x04000000UL /* Account to memcg */
92 # define SLAB_ACCOUNT 0x00000000UL
95 /* The following flags affect the page allocator grouping pages by mobility */
96 #define SLAB_RECLAIM_ACCOUNT 0x00020000UL /* Objects are reclaimable */
97 #define SLAB_TEMPORARY SLAB_RECLAIM_ACCOUNT /* Objects are short-lived */
99 * ZERO_SIZE_PTR will be returned for zero sized kmalloc requests.
101 * Dereferencing ZERO_SIZE_PTR will lead to a distinct access fault.
103 * ZERO_SIZE_PTR can be passed to kfree though in the same way that NULL can.
104 * Both make kfree a no-op.
106 #define ZERO_SIZE_PTR ((void *)16)
108 #define ZERO_OR_NULL_PTR(x) ((unsigned long)(x) <= \
109 (unsigned long)ZERO_SIZE_PTR)
111 #include <linux/kmemleak.h>
112 #include <linux/kasan.h>
116 * struct kmem_cache related prototypes
118 void __init kmem_cache_init(void);
119 bool slab_is_available(void);
121 struct kmem_cache *kmem_cache_create(const char *, size_t, size_t,
124 void kmem_cache_destroy(struct kmem_cache *);
125 int kmem_cache_shrink(struct kmem_cache *);
127 void memcg_create_kmem_cache(struct mem_cgroup *, struct kmem_cache *);
128 void memcg_deactivate_kmem_caches(struct mem_cgroup *);
129 void memcg_destroy_kmem_caches(struct mem_cgroup *);
132 * Please use this macro to create slab caches. Simply specify the
133 * name of the structure and maybe some flags that are listed above.
135 * The alignment of the struct determines object alignment. If you
136 * f.e. add ____cacheline_aligned_in_smp to the struct declaration
137 * then the objects will be properly aligned in SMP configurations.
139 #define KMEM_CACHE(__struct, __flags) kmem_cache_create(#__struct,\
140 sizeof(struct __struct), __alignof__(struct __struct),\
144 * Common kmalloc functions provided by all allocators
146 void * __must_check __krealloc(const void *, size_t, gfp_t);
147 void * __must_check krealloc(const void *, size_t, gfp_t);
148 void kfree(const void *);
149 void kzfree(const void *);
150 size_t ksize(const void *);
153 * Some archs want to perform DMA into kmalloc caches and need a guaranteed
154 * alignment larger than the alignment of a 64-bit integer.
155 * Setting ARCH_KMALLOC_MINALIGN in arch headers allows that.
157 #if defined(ARCH_DMA_MINALIGN) && ARCH_DMA_MINALIGN > 8
158 #define ARCH_KMALLOC_MINALIGN ARCH_DMA_MINALIGN
159 #define KMALLOC_MIN_SIZE ARCH_DMA_MINALIGN
160 #define KMALLOC_SHIFT_LOW ilog2(ARCH_DMA_MINALIGN)
162 #define ARCH_KMALLOC_MINALIGN __alignof__(unsigned long long)
166 * Setting ARCH_SLAB_MINALIGN in arch headers allows a different alignment.
167 * Intended for arches that get misalignment faults even for 64 bit integer
170 #ifndef ARCH_SLAB_MINALIGN
171 #define ARCH_SLAB_MINALIGN __alignof__(unsigned long long)
175 * kmalloc and friends return ARCH_KMALLOC_MINALIGN aligned
176 * pointers. kmem_cache_alloc and friends return ARCH_SLAB_MINALIGN
179 #define __assume_kmalloc_alignment __assume_aligned(ARCH_KMALLOC_MINALIGN)
180 #define __assume_slab_alignment __assume_aligned(ARCH_SLAB_MINALIGN)
181 #define __assume_page_alignment __assume_aligned(PAGE_SIZE)
184 * Kmalloc array related definitions
189 * The largest kmalloc size supported by the SLAB allocators is
190 * 32 megabyte (2^25) or the maximum allocatable page order if that is
193 * WARNING: Its not easy to increase this value since the allocators have
194 * to do various tricks to work around compiler limitations in order to
195 * ensure proper constant folding.
197 #define KMALLOC_SHIFT_HIGH ((MAX_ORDER + PAGE_SHIFT - 1) <= 25 ? \
198 (MAX_ORDER + PAGE_SHIFT - 1) : 25)
199 #define KMALLOC_SHIFT_MAX KMALLOC_SHIFT_HIGH
200 #ifndef KMALLOC_SHIFT_LOW
201 #define KMALLOC_SHIFT_LOW 5
207 * SLUB directly allocates requests fitting in to an order-1 page
208 * (PAGE_SIZE*2). Larger requests are passed to the page allocator.
210 #define KMALLOC_SHIFT_HIGH (PAGE_SHIFT + 1)
211 #define KMALLOC_SHIFT_MAX (MAX_ORDER + PAGE_SHIFT)
212 #ifndef KMALLOC_SHIFT_LOW
213 #define KMALLOC_SHIFT_LOW 3
219 * SLOB passes all requests larger than one page to the page allocator.
220 * No kmalloc array is necessary since objects of different sizes can
221 * be allocated from the same page.
223 #define KMALLOC_SHIFT_HIGH PAGE_SHIFT
224 #define KMALLOC_SHIFT_MAX 30
225 #ifndef KMALLOC_SHIFT_LOW
226 #define KMALLOC_SHIFT_LOW 3
230 /* Maximum allocatable size */
231 #define KMALLOC_MAX_SIZE (1UL << KMALLOC_SHIFT_MAX)
232 /* Maximum size for which we actually use a slab cache */
233 #define KMALLOC_MAX_CACHE_SIZE (1UL << KMALLOC_SHIFT_HIGH)
234 /* Maximum order allocatable via the slab allocagtor */
235 #define KMALLOC_MAX_ORDER (KMALLOC_SHIFT_MAX - PAGE_SHIFT)
240 #ifndef KMALLOC_MIN_SIZE
241 #define KMALLOC_MIN_SIZE (1 << KMALLOC_SHIFT_LOW)
245 * This restriction comes from byte sized index implementation.
246 * Page size is normally 2^12 bytes and, in this case, if we want to use
247 * byte sized index which can represent 2^8 entries, the size of the object
248 * should be equal or greater to 2^12 / 2^8 = 2^4 = 16.
249 * If minimum size of kmalloc is less than 16, we use it as minimum object
250 * size and give up to use byte sized index.
252 #define SLAB_OBJ_MIN_SIZE (KMALLOC_MIN_SIZE < 16 ? \
253 (KMALLOC_MIN_SIZE) : 16)
256 extern struct kmem_cache *kmalloc_caches[KMALLOC_SHIFT_HIGH + 1];
257 #ifdef CONFIG_ZONE_DMA
258 extern struct kmem_cache *kmalloc_dma_caches[KMALLOC_SHIFT_HIGH + 1];
262 * Figure out which kmalloc slab an allocation of a certain size
266 * 2 = 129 .. 192 bytes
267 * n = 2^(n-1)+1 .. 2^n
269 static __always_inline int kmalloc_index(size_t size)
274 if (size <= KMALLOC_MIN_SIZE)
275 return KMALLOC_SHIFT_LOW;
277 if (KMALLOC_MIN_SIZE <= 32 && size > 64 && size <= 96)
279 if (KMALLOC_MIN_SIZE <= 64 && size > 128 && size <= 192)
281 if (size <= 8) return 3;
282 if (size <= 16) return 4;
283 if (size <= 32) return 5;
284 if (size <= 64) return 6;
285 if (size <= 128) return 7;
286 if (size <= 256) return 8;
287 if (size <= 512) return 9;
288 if (size <= 1024) return 10;
289 if (size <= 2 * 1024) return 11;
290 if (size <= 4 * 1024) return 12;
291 if (size <= 8 * 1024) return 13;
292 if (size <= 16 * 1024) return 14;
293 if (size <= 32 * 1024) return 15;
294 if (size <= 64 * 1024) return 16;
295 if (size <= 128 * 1024) return 17;
296 if (size <= 256 * 1024) return 18;
297 if (size <= 512 * 1024) return 19;
298 if (size <= 1024 * 1024) return 20;
299 if (size <= 2 * 1024 * 1024) return 21;
300 if (size <= 4 * 1024 * 1024) return 22;
301 if (size <= 8 * 1024 * 1024) return 23;
302 if (size <= 16 * 1024 * 1024) return 24;
303 if (size <= 32 * 1024 * 1024) return 25;
304 if (size <= 64 * 1024 * 1024) return 26;
307 /* Will never be reached. Needed because the compiler may complain */
310 #endif /* !CONFIG_SLOB */
312 void *__kmalloc(size_t size, gfp_t flags) __assume_kmalloc_alignment;
313 void *kmem_cache_alloc(struct kmem_cache *, gfp_t flags) __assume_slab_alignment;
314 void kmem_cache_free(struct kmem_cache *, void *);
317 * Bulk allocation and freeing operations. These are accellerated in an
318 * allocator specific way to avoid taking locks repeatedly or building
319 * metadata structures unnecessarily.
321 * Note that interrupts must be enabled when calling these functions.
323 void kmem_cache_free_bulk(struct kmem_cache *, size_t, void **);
324 int kmem_cache_alloc_bulk(struct kmem_cache *, gfp_t, size_t, void **);
327 void *__kmalloc_node(size_t size, gfp_t flags, int node) __assume_kmalloc_alignment;
328 void *kmem_cache_alloc_node(struct kmem_cache *, gfp_t flags, int node) __assume_slab_alignment;
330 static __always_inline void *__kmalloc_node(size_t size, gfp_t flags, int node)
332 return __kmalloc(size, flags);
335 static __always_inline void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t flags, int node)
337 return kmem_cache_alloc(s, flags);
341 #ifdef CONFIG_TRACING
342 extern void *kmem_cache_alloc_trace(struct kmem_cache *, gfp_t, size_t) __assume_slab_alignment;
345 extern void *kmem_cache_alloc_node_trace(struct kmem_cache *s,
347 int node, size_t size) __assume_slab_alignment;
349 static __always_inline void *
350 kmem_cache_alloc_node_trace(struct kmem_cache *s,
352 int node, size_t size)
354 return kmem_cache_alloc_trace(s, gfpflags, size);
356 #endif /* CONFIG_NUMA */
358 #else /* CONFIG_TRACING */
359 static __always_inline void *kmem_cache_alloc_trace(struct kmem_cache *s,
360 gfp_t flags, size_t size)
362 void *ret = kmem_cache_alloc(s, flags);
364 kasan_kmalloc(s, ret, size);
368 static __always_inline void *
369 kmem_cache_alloc_node_trace(struct kmem_cache *s,
371 int node, size_t size)
373 void *ret = kmem_cache_alloc_node(s, gfpflags, node);
375 kasan_kmalloc(s, ret, size);
378 #endif /* CONFIG_TRACING */
380 extern void *kmalloc_order(size_t size, gfp_t flags, unsigned int order) __assume_page_alignment;
382 #ifdef CONFIG_TRACING
383 extern void *kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order) __assume_page_alignment;
385 static __always_inline void *
386 kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order)
388 return kmalloc_order(size, flags, order);
392 static __always_inline void *kmalloc_large(size_t size, gfp_t flags)
394 unsigned int order = get_order(size);
395 return kmalloc_order_trace(size, flags, order);
399 * kmalloc - allocate memory
400 * @size: how many bytes of memory are required.
401 * @flags: the type of memory to allocate.
403 * kmalloc is the normal method of allocating memory
404 * for objects smaller than page size in the kernel.
406 * The @flags argument may be one of:
408 * %GFP_USER - Allocate memory on behalf of user. May sleep.
410 * %GFP_KERNEL - Allocate normal kernel ram. May sleep.
412 * %GFP_ATOMIC - Allocation will not sleep. May use emergency pools.
413 * For example, use this inside interrupt handlers.
415 * %GFP_HIGHUSER - Allocate pages from high memory.
417 * %GFP_NOIO - Do not do any I/O at all while trying to get memory.
419 * %GFP_NOFS - Do not make any fs calls while trying to get memory.
421 * %GFP_NOWAIT - Allocation will not sleep.
423 * %__GFP_THISNODE - Allocate node-local memory only.
425 * %GFP_DMA - Allocation suitable for DMA.
426 * Should only be used for kmalloc() caches. Otherwise, use a
427 * slab created with SLAB_DMA.
429 * Also it is possible to set different flags by OR'ing
430 * in one or more of the following additional @flags:
432 * %__GFP_COLD - Request cache-cold pages instead of
433 * trying to return cache-warm pages.
435 * %__GFP_HIGH - This allocation has high priority and may use emergency pools.
437 * %__GFP_NOFAIL - Indicate that this allocation is in no way allowed to fail
438 * (think twice before using).
440 * %__GFP_NORETRY - If memory is not immediately available,
441 * then give up at once.
443 * %__GFP_NOWARN - If allocation fails, don't issue any warnings.
445 * %__GFP_REPEAT - If allocation fails initially, try once more before failing.
447 * There are other flags available as well, but these are not intended
448 * for general use, and so are not documented here. For a full list of
449 * potential flags, always refer to linux/gfp.h.
451 static __always_inline void *kmalloc(size_t size, gfp_t flags)
453 if (__builtin_constant_p(size)) {
454 if (size > KMALLOC_MAX_CACHE_SIZE)
455 return kmalloc_large(size, flags);
457 if (!(flags & GFP_DMA)) {
458 int index = kmalloc_index(size);
461 return ZERO_SIZE_PTR;
463 return kmem_cache_alloc_trace(kmalloc_caches[index],
468 return __kmalloc(size, flags);
472 * Determine size used for the nth kmalloc cache.
473 * return size or 0 if a kmalloc cache for that
474 * size does not exist
476 static __always_inline int kmalloc_size(int n)
482 if (n == 1 && KMALLOC_MIN_SIZE <= 32)
485 if (n == 2 && KMALLOC_MIN_SIZE <= 64)
491 static __always_inline void *kmalloc_node(size_t size, gfp_t flags, int node)
494 if (__builtin_constant_p(size) &&
495 size <= KMALLOC_MAX_CACHE_SIZE && !(flags & GFP_DMA)) {
496 int i = kmalloc_index(size);
499 return ZERO_SIZE_PTR;
501 return kmem_cache_alloc_node_trace(kmalloc_caches[i],
505 return __kmalloc_node(size, flags, node);
508 struct memcg_cache_array {
510 struct kmem_cache *entries[0];
514 * This is the main placeholder for memcg-related information in kmem caches.
515 * Both the root cache and the child caches will have it. For the root cache,
516 * this will hold a dynamically allocated array large enough to hold
517 * information about the currently limited memcgs in the system. To allow the
518 * array to be accessed without taking any locks, on relocation we free the old
519 * version only after a grace period.
521 * Child caches will hold extra metadata needed for its operation. Fields are:
523 * @memcg: pointer to the memcg this cache belongs to
524 * @root_cache: pointer to the global, root cache, this cache was derived from
526 * Both root and child caches of the same kind are linked into a list chained
529 struct memcg_cache_params {
531 struct list_head list;
533 struct memcg_cache_array __rcu *memcg_caches;
535 struct mem_cgroup *memcg;
536 struct kmem_cache *root_cache;
541 int memcg_update_all_caches(int num_memcgs);
544 * kmalloc_array - allocate memory for an array.
545 * @n: number of elements.
546 * @size: element size.
547 * @flags: the type of memory to allocate (see kmalloc).
549 static inline void *kmalloc_array(size_t n, size_t size, gfp_t flags)
551 if (size != 0 && n > SIZE_MAX / size)
553 return __kmalloc(n * size, flags);
557 * kcalloc - allocate memory for an array. The memory is set to zero.
558 * @n: number of elements.
559 * @size: element size.
560 * @flags: the type of memory to allocate (see kmalloc).
562 static inline void *kcalloc(size_t n, size_t size, gfp_t flags)
564 return kmalloc_array(n, size, flags | __GFP_ZERO);
568 * kmalloc_track_caller is a special version of kmalloc that records the
569 * calling function of the routine calling it for slab leak tracking instead
570 * of just the calling function (confusing, eh?).
571 * It's useful when the call to kmalloc comes from a widely-used standard
572 * allocator where we care about the real place the memory allocation
573 * request comes from.
575 extern void *__kmalloc_track_caller(size_t, gfp_t, unsigned long);
576 #define kmalloc_track_caller(size, flags) \
577 __kmalloc_track_caller(size, flags, _RET_IP_)
580 extern void *__kmalloc_node_track_caller(size_t, gfp_t, int, unsigned long);
581 #define kmalloc_node_track_caller(size, flags, node) \
582 __kmalloc_node_track_caller(size, flags, node, \
585 #else /* CONFIG_NUMA */
587 #define kmalloc_node_track_caller(size, flags, node) \
588 kmalloc_track_caller(size, flags)
590 #endif /* CONFIG_NUMA */
595 static inline void *kmem_cache_zalloc(struct kmem_cache *k, gfp_t flags)
597 return kmem_cache_alloc(k, flags | __GFP_ZERO);
601 * kzalloc - allocate memory. The memory is set to zero.
602 * @size: how many bytes of memory are required.
603 * @flags: the type of memory to allocate (see kmalloc).
605 static inline void *kzalloc(size_t size, gfp_t flags)
607 return kmalloc(size, flags | __GFP_ZERO);
611 * kzalloc_node - allocate zeroed memory from a particular memory node.
612 * @size: how many bytes of memory are required.
613 * @flags: the type of memory to allocate (see kmalloc).
614 * @node: memory node from which to allocate
616 static inline void *kzalloc_node(size_t size, gfp_t flags, int node)
618 return kmalloc_node(size, flags | __GFP_ZERO, node);
621 unsigned int kmem_cache_size(struct kmem_cache *s);
622 void __init kmem_cache_init_late(void);
624 #endif /* _LINUX_SLAB_H */