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 * See also the comment on struct slab_rcu in mm/slab.c.
58 #define SLAB_DESTROY_BY_RCU 0x00080000UL /* Defer freeing slabs to RCU */
59 #define SLAB_MEM_SPREAD 0x00100000UL /* Spread some memory over cpuset */
60 #define SLAB_TRACE 0x00200000UL /* Trace allocations and frees */
62 /* Flag to prevent checks on free */
63 #ifdef CONFIG_DEBUG_OBJECTS
64 # define SLAB_DEBUG_OBJECTS 0x00400000UL
66 # define SLAB_DEBUG_OBJECTS 0x00000000UL
69 #define SLAB_NOLEAKTRACE 0x00800000UL /* Avoid kmemleak tracing */
71 /* Don't track use of uninitialized memory */
72 #ifdef CONFIG_KMEMCHECK
73 # define SLAB_NOTRACK 0x01000000UL
75 # define SLAB_NOTRACK 0x00000000UL
77 #ifdef CONFIG_FAILSLAB
78 # define SLAB_FAILSLAB 0x02000000UL /* Fault injection mark */
80 # define SLAB_FAILSLAB 0x00000000UL
83 /* The following flags affect the page allocator grouping pages by mobility */
84 #define SLAB_RECLAIM_ACCOUNT 0x00020000UL /* Objects are reclaimable */
85 #define SLAB_TEMPORARY SLAB_RECLAIM_ACCOUNT /* Objects are short-lived */
87 * ZERO_SIZE_PTR will be returned for zero sized kmalloc requests.
89 * Dereferencing ZERO_SIZE_PTR will lead to a distinct access fault.
91 * ZERO_SIZE_PTR can be passed to kfree though in the same way that NULL can.
92 * Both make kfree a no-op.
94 #define ZERO_SIZE_PTR ((void *)16)
96 #define ZERO_OR_NULL_PTR(x) ((unsigned long)(x) <= \
97 (unsigned long)ZERO_SIZE_PTR)
99 #include <linux/kmemleak.h>
103 * struct kmem_cache related prototypes
105 void __init kmem_cache_init(void);
106 int slab_is_available(void);
108 struct kmem_cache *kmem_cache_create(const char *, size_t, size_t,
112 kmem_cache_create_memcg(struct mem_cgroup *, const char *, size_t, size_t,
113 unsigned long, void (*)(void *), struct kmem_cache *);
114 void kmem_cache_destroy(struct kmem_cache *);
115 int kmem_cache_shrink(struct kmem_cache *);
116 void kmem_cache_free(struct kmem_cache *, void *);
119 * Please use this macro to create slab caches. Simply specify the
120 * name of the structure and maybe some flags that are listed above.
122 * The alignment of the struct determines object alignment. If you
123 * f.e. add ____cacheline_aligned_in_smp to the struct declaration
124 * then the objects will be properly aligned in SMP configurations.
126 #define KMEM_CACHE(__struct, __flags) kmem_cache_create(#__struct,\
127 sizeof(struct __struct), __alignof__(struct __struct),\
131 * Common kmalloc functions provided by all allocators
133 void * __must_check __krealloc(const void *, size_t, gfp_t);
134 void * __must_check krealloc(const void *, size_t, gfp_t);
135 void kfree(const void *);
136 void kzfree(const void *);
137 size_t ksize(const void *);
140 * Some archs want to perform DMA into kmalloc caches and need a guaranteed
141 * alignment larger than the alignment of a 64-bit integer.
142 * Setting ARCH_KMALLOC_MINALIGN in arch headers allows that.
144 #if defined(ARCH_DMA_MINALIGN) && ARCH_DMA_MINALIGN > 8
145 #define ARCH_KMALLOC_MINALIGN ARCH_DMA_MINALIGN
146 #define KMALLOC_MIN_SIZE ARCH_DMA_MINALIGN
147 #define KMALLOC_SHIFT_LOW ilog2(ARCH_DMA_MINALIGN)
149 #define ARCH_KMALLOC_MINALIGN __alignof__(unsigned long long)
154 * Common fields provided in kmem_cache by all slab allocators
155 * This struct is either used directly by the allocator (SLOB)
156 * or the allocator must include definitions for all fields
157 * provided in kmem_cache_common in their definition of kmem_cache.
159 * Once we can do anonymous structs (C11 standard) we could put a
160 * anonymous struct definition in these allocators so that the
161 * separate allocations in the kmem_cache structure of SLAB and
162 * SLUB is no longer needed.
165 unsigned int object_size;/* The original size of the object */
166 unsigned int size; /* The aligned/padded/added on size */
167 unsigned int align; /* Alignment as calculated */
168 unsigned long flags; /* Active flags on the slab */
169 const char *name; /* Slab name for sysfs */
170 int refcount; /* Use counter */
171 void (*ctor)(void *); /* Called on object slot creation */
172 struct list_head list; /* List of all slab caches on the system */
175 #endif /* CONFIG_SLOB */
178 * Kmalloc array related definitions
183 * The largest kmalloc size supported by the SLAB allocators is
184 * 32 megabyte (2^25) or the maximum allocatable page order if that is
187 * WARNING: Its not easy to increase this value since the allocators have
188 * to do various tricks to work around compiler limitations in order to
189 * ensure proper constant folding.
191 #define KMALLOC_SHIFT_HIGH ((MAX_ORDER + PAGE_SHIFT - 1) <= 25 ? \
192 (MAX_ORDER + PAGE_SHIFT - 1) : 25)
193 #define KMALLOC_SHIFT_MAX KMALLOC_SHIFT_HIGH
194 #ifndef KMALLOC_SHIFT_LOW
195 #define KMALLOC_SHIFT_LOW 5
201 * SLUB allocates up to order 2 pages directly and otherwise
202 * passes the request to the page allocator.
204 #define KMALLOC_SHIFT_HIGH (PAGE_SHIFT + 1)
205 #define KMALLOC_SHIFT_MAX (MAX_ORDER + PAGE_SHIFT)
206 #ifndef KMALLOC_SHIFT_LOW
207 #define KMALLOC_SHIFT_LOW 3
213 * SLOB passes all page size and larger requests to the page allocator.
214 * No kmalloc array is necessary since objects of different sizes can
215 * be allocated from the same page.
217 #define KMALLOC_SHIFT_MAX 30
218 #define KMALLOC_SHIFT_HIGH PAGE_SHIFT
219 #ifndef KMALLOC_SHIFT_LOW
220 #define KMALLOC_SHIFT_LOW 3
224 /* Maximum allocatable size */
225 #define KMALLOC_MAX_SIZE (1UL << KMALLOC_SHIFT_MAX)
226 /* Maximum size for which we actually use a slab cache */
227 #define KMALLOC_MAX_CACHE_SIZE (1UL << KMALLOC_SHIFT_HIGH)
228 /* Maximum order allocatable via the slab allocagtor */
229 #define KMALLOC_MAX_ORDER (KMALLOC_SHIFT_MAX - PAGE_SHIFT)
234 #ifndef KMALLOC_MIN_SIZE
235 #define KMALLOC_MIN_SIZE (1 << KMALLOC_SHIFT_LOW)
239 extern struct kmem_cache *kmalloc_caches[KMALLOC_SHIFT_HIGH + 1];
240 #ifdef CONFIG_ZONE_DMA
241 extern struct kmem_cache *kmalloc_dma_caches[KMALLOC_SHIFT_HIGH + 1];
245 * Figure out which kmalloc slab an allocation of a certain size
249 * 2 = 120 .. 192 bytes
250 * n = 2^(n-1) .. 2^n -1
252 static __always_inline int kmalloc_index(size_t size)
257 if (size <= KMALLOC_MIN_SIZE)
258 return KMALLOC_SHIFT_LOW;
260 if (KMALLOC_MIN_SIZE <= 32 && size > 64 && size <= 96)
262 if (KMALLOC_MIN_SIZE <= 64 && size > 128 && size <= 192)
264 if (size <= 8) return 3;
265 if (size <= 16) return 4;
266 if (size <= 32) return 5;
267 if (size <= 64) return 6;
268 if (size <= 128) return 7;
269 if (size <= 256) return 8;
270 if (size <= 512) return 9;
271 if (size <= 1024) return 10;
272 if (size <= 2 * 1024) return 11;
273 if (size <= 4 * 1024) return 12;
274 if (size <= 8 * 1024) return 13;
275 if (size <= 16 * 1024) return 14;
276 if (size <= 32 * 1024) return 15;
277 if (size <= 64 * 1024) return 16;
278 if (size <= 128 * 1024) return 17;
279 if (size <= 256 * 1024) return 18;
280 if (size <= 512 * 1024) return 19;
281 if (size <= 1024 * 1024) return 20;
282 if (size <= 2 * 1024 * 1024) return 21;
283 if (size <= 4 * 1024 * 1024) return 22;
284 if (size <= 8 * 1024 * 1024) return 23;
285 if (size <= 16 * 1024 * 1024) return 24;
286 if (size <= 32 * 1024 * 1024) return 25;
287 if (size <= 64 * 1024 * 1024) return 26;
290 /* Will never be reached. Needed because the compiler may complain */
293 #endif /* !CONFIG_SLOB */
295 void *__kmalloc(size_t size, gfp_t flags);
296 void *kmem_cache_alloc(struct kmem_cache *, gfp_t flags);
299 void *__kmalloc_node(size_t size, gfp_t flags, int node);
300 void *kmem_cache_alloc_node(struct kmem_cache *, gfp_t flags, int node);
302 static __always_inline void *__kmalloc_node(size_t size, gfp_t flags, int node)
304 return __kmalloc(size, flags);
307 static __always_inline void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t flags, int node)
309 return kmem_cache_alloc(s, flags);
313 #ifdef CONFIG_TRACING
314 extern void *kmem_cache_alloc_trace(struct kmem_cache *, gfp_t, size_t);
317 extern void *kmem_cache_alloc_node_trace(struct kmem_cache *s,
319 int node, size_t size);
321 static __always_inline void *
322 kmem_cache_alloc_node_trace(struct kmem_cache *s,
324 int node, size_t size)
326 return kmem_cache_alloc_trace(s, gfpflags, size);
328 #endif /* CONFIG_NUMA */
330 #else /* CONFIG_TRACING */
331 static __always_inline void *kmem_cache_alloc_trace(struct kmem_cache *s,
332 gfp_t flags, size_t size)
334 return kmem_cache_alloc(s, flags);
337 static __always_inline void *
338 kmem_cache_alloc_node_trace(struct kmem_cache *s,
340 int node, size_t size)
342 return kmem_cache_alloc_node(s, gfpflags, node);
344 #endif /* CONFIG_TRACING */
347 #include <linux/slab_def.h>
351 #include <linux/slub_def.h>
354 static __always_inline void *
355 kmalloc_order(size_t size, gfp_t flags, unsigned int order)
359 flags |= (__GFP_COMP | __GFP_KMEMCG);
360 ret = (void *) __get_free_pages(flags, order);
361 kmemleak_alloc(ret, size, 1, flags);
365 #ifdef CONFIG_TRACING
366 extern void *kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order);
368 static __always_inline void *
369 kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order)
371 return kmalloc_order(size, flags, order);
375 static __always_inline void *kmalloc_large(size_t size, gfp_t flags)
377 unsigned int order = get_order(size);
378 return kmalloc_order_trace(size, flags, order);
382 * kmalloc - allocate memory
383 * @size: how many bytes of memory are required.
384 * @flags: the type of memory to allocate (see kcalloc).
386 * kmalloc is the normal method of allocating memory
387 * for objects smaller than page size in the kernel.
389 static __always_inline void *kmalloc(size_t size, gfp_t flags)
391 if (__builtin_constant_p(size)) {
392 if (size > KMALLOC_MAX_CACHE_SIZE)
393 return kmalloc_large(size, flags);
395 if (!(flags & GFP_DMA)) {
396 int index = kmalloc_index(size);
399 return ZERO_SIZE_PTR;
401 return kmem_cache_alloc_trace(kmalloc_caches[index],
406 return __kmalloc(size, flags);
410 * Determine size used for the nth kmalloc cache.
411 * return size or 0 if a kmalloc cache for that
412 * size does not exist
414 static __always_inline int kmalloc_size(int n)
420 if (n == 1 && KMALLOC_MIN_SIZE <= 32)
423 if (n == 2 && KMALLOC_MIN_SIZE <= 64)
429 static __always_inline void *kmalloc_node(size_t size, gfp_t flags, int node)
432 if (__builtin_constant_p(size) &&
433 size <= KMALLOC_MAX_CACHE_SIZE && !(flags & GFP_DMA)) {
434 int i = kmalloc_index(size);
437 return ZERO_SIZE_PTR;
439 return kmem_cache_alloc_node_trace(kmalloc_caches[i],
443 return __kmalloc_node(size, flags, node);
447 * Setting ARCH_SLAB_MINALIGN in arch headers allows a different alignment.
448 * Intended for arches that get misalignment faults even for 64 bit integer
451 #ifndef ARCH_SLAB_MINALIGN
452 #define ARCH_SLAB_MINALIGN __alignof__(unsigned long long)
455 * This is the main placeholder for memcg-related information in kmem caches.
456 * struct kmem_cache will hold a pointer to it, so the memory cost while
457 * disabled is 1 pointer. The runtime cost while enabled, gets bigger than it
458 * would otherwise be if that would be bundled in kmem_cache: we'll need an
459 * extra pointer chase. But the trade off clearly lays in favor of not
460 * penalizing non-users.
462 * Both the root cache and the child caches will have it. For the root cache,
463 * this will hold a dynamically allocated array large enough to hold
464 * information about the currently limited memcgs in the system.
466 * Child caches will hold extra metadata needed for its operation. Fields are:
468 * @memcg: pointer to the memcg this cache belongs to
469 * @list: list_head for the list of all caches in this memcg
470 * @root_cache: pointer to the global, root cache, this cache was derived from
471 * @dead: set to true after the memcg dies; the cache may still be around.
472 * @nr_pages: number of pages that belongs to this cache.
473 * @destroy: worker to be called whenever we are ready, or believe we may be
474 * ready, to destroy this cache.
476 struct memcg_cache_params {
479 struct kmem_cache *memcg_caches[0];
481 struct mem_cgroup *memcg;
482 struct list_head list;
483 struct kmem_cache *root_cache;
486 struct work_struct destroy;
491 int memcg_update_all_caches(int num_memcgs);
494 int cache_show(struct kmem_cache *s, struct seq_file *m);
495 void print_slabinfo_header(struct seq_file *m);
498 * kmalloc - allocate memory
499 * @size: how many bytes of memory are required.
500 * @flags: the type of memory to allocate.
502 * The @flags argument may be one of:
504 * %GFP_USER - Allocate memory on behalf of user. May sleep.
506 * %GFP_KERNEL - Allocate normal kernel ram. May sleep.
508 * %GFP_ATOMIC - Allocation will not sleep. May use emergency pools.
509 * For example, use this inside interrupt handlers.
511 * %GFP_HIGHUSER - Allocate pages from high memory.
513 * %GFP_NOIO - Do not do any I/O at all while trying to get memory.
515 * %GFP_NOFS - Do not make any fs calls while trying to get memory.
517 * %GFP_NOWAIT - Allocation will not sleep.
519 * %GFP_THISNODE - Allocate node-local memory only.
521 * %GFP_DMA - Allocation suitable for DMA.
522 * Should only be used for kmalloc() caches. Otherwise, use a
523 * slab created with SLAB_DMA.
525 * Also it is possible to set different flags by OR'ing
526 * in one or more of the following additional @flags:
528 * %__GFP_COLD - Request cache-cold pages instead of
529 * trying to return cache-warm pages.
531 * %__GFP_HIGH - This allocation has high priority and may use emergency pools.
533 * %__GFP_NOFAIL - Indicate that this allocation is in no way allowed to fail
534 * (think twice before using).
536 * %__GFP_NORETRY - If memory is not immediately available,
537 * then give up at once.
539 * %__GFP_NOWARN - If allocation fails, don't issue any warnings.
541 * %__GFP_REPEAT - If allocation fails initially, try once more before failing.
543 * There are other flags available as well, but these are not intended
544 * for general use, and so are not documented here. For a full list of
545 * potential flags, always refer to linux/gfp.h.
547 * kmalloc is the normal method of allocating memory
550 static __always_inline void *kmalloc(size_t size, gfp_t flags);
553 * kmalloc_array - allocate memory for an array.
554 * @n: number of elements.
555 * @size: element size.
556 * @flags: the type of memory to allocate (see kmalloc).
558 static inline void *kmalloc_array(size_t n, size_t size, gfp_t flags)
560 if (size != 0 && n > SIZE_MAX / size)
562 return __kmalloc(n * size, flags);
566 * kcalloc - allocate memory for an array. The memory is set to zero.
567 * @n: number of elements.
568 * @size: element size.
569 * @flags: the type of memory to allocate (see kmalloc).
571 static inline void *kcalloc(size_t n, size_t size, gfp_t flags)
573 return kmalloc_array(n, size, flags | __GFP_ZERO);
577 * kmalloc_track_caller is a special version of kmalloc that records the
578 * calling function of the routine calling it for slab leak tracking instead
579 * of just the calling function (confusing, eh?).
580 * It's useful when the call to kmalloc comes from a widely-used standard
581 * allocator where we care about the real place the memory allocation
582 * request comes from.
584 #if defined(CONFIG_DEBUG_SLAB) || defined(CONFIG_SLUB) || \
585 (defined(CONFIG_SLAB) && defined(CONFIG_TRACING)) || \
586 (defined(CONFIG_SLOB) && defined(CONFIG_TRACING))
587 extern void *__kmalloc_track_caller(size_t, gfp_t, unsigned long);
588 #define kmalloc_track_caller(size, flags) \
589 __kmalloc_track_caller(size, flags, _RET_IP_)
591 #define kmalloc_track_caller(size, flags) \
592 __kmalloc(size, flags)
593 #endif /* DEBUG_SLAB */
597 * kmalloc_node_track_caller is a special version of kmalloc_node that
598 * records the calling function of the routine calling it for slab leak
599 * tracking instead of just the calling function (confusing, eh?).
600 * It's useful when the call to kmalloc_node comes from a widely-used
601 * standard allocator where we care about the real place the memory
602 * allocation request comes from.
604 #if defined(CONFIG_DEBUG_SLAB) || defined(CONFIG_SLUB) || \
605 (defined(CONFIG_SLAB) && defined(CONFIG_TRACING)) || \
606 (defined(CONFIG_SLOB) && defined(CONFIG_TRACING))
607 extern void *__kmalloc_node_track_caller(size_t, gfp_t, int, unsigned long);
608 #define kmalloc_node_track_caller(size, flags, node) \
609 __kmalloc_node_track_caller(size, flags, node, \
612 #define kmalloc_node_track_caller(size, flags, node) \
613 __kmalloc_node(size, flags, node)
616 #else /* CONFIG_NUMA */
618 #define kmalloc_node_track_caller(size, flags, node) \
619 kmalloc_track_caller(size, flags)
621 #endif /* CONFIG_NUMA */
626 static inline void *kmem_cache_zalloc(struct kmem_cache *k, gfp_t flags)
628 return kmem_cache_alloc(k, flags | __GFP_ZERO);
632 * kzalloc - allocate memory. The memory is set to zero.
633 * @size: how many bytes of memory are required.
634 * @flags: the type of memory to allocate (see kmalloc).
636 static inline void *kzalloc(size_t size, gfp_t flags)
638 return kmalloc(size, flags | __GFP_ZERO);
642 * kzalloc_node - allocate zeroed memory from a particular memory node.
643 * @size: how many bytes of memory are required.
644 * @flags: the type of memory to allocate (see kmalloc).
645 * @node: memory node from which to allocate
647 static inline void *kzalloc_node(size_t size, gfp_t flags, int node)
649 return kmalloc_node(size, flags | __GFP_ZERO, node);
653 * Determine the size of a slab object
655 static inline unsigned int kmem_cache_size(struct kmem_cache *s)
657 return s->object_size;
660 void __init kmem_cache_init_late(void);
662 #endif /* _LINUX_SLAB_H */