1 /* SPDX-License-Identifier: GPL-2.0 */
3 * Written by Mark Hemment, 1996 (markhe@nextd.demon.co.uk).
5 * (C) SGI 2006, Christoph Lameter
6 * Cleaned up and restructured to ease the addition of alternative
7 * implementations of SLAB allocators.
8 * (C) Linux Foundation 2008-2013
9 * Unified interface for all slab allocators
15 #include <linux/gfp.h>
16 #include <linux/types.h>
17 #include <linux/workqueue.h>
21 * Flags to pass to kmem_cache_create().
22 * The ones marked DEBUG are only valid if CONFIG_DEBUG_SLAB is set.
24 /* DEBUG: Perform (expensive) checks on alloc/free */
25 #define SLAB_CONSISTENCY_CHECKS ((slab_flags_t __force)0x00000100UL)
26 /* DEBUG: Red zone objs in a cache */
27 #define SLAB_RED_ZONE ((slab_flags_t __force)0x00000400UL)
28 /* DEBUG: Poison objects */
29 #define SLAB_POISON ((slab_flags_t __force)0x00000800UL)
30 /* Align objs on cache lines */
31 #define SLAB_HWCACHE_ALIGN ((slab_flags_t __force)0x00002000UL)
32 /* Use GFP_DMA memory */
33 #define SLAB_CACHE_DMA ((slab_flags_t __force)0x00004000UL)
34 /* DEBUG: Store the last owner for bug hunting */
35 #define SLAB_STORE_USER ((slab_flags_t __force)0x00010000UL)
36 /* Panic if kmem_cache_create() fails */
37 #define SLAB_PANIC ((slab_flags_t __force)0x00040000UL)
39 * SLAB_TYPESAFE_BY_RCU - **WARNING** READ THIS!
41 * This delays freeing the SLAB page by a grace period, it does _NOT_
42 * delay object freeing. This means that if you do kmem_cache_free()
43 * that memory location is free to be reused at any time. Thus it may
44 * be possible to see another object there in the same RCU grace period.
46 * This feature only ensures the memory location backing the object
47 * stays valid, the trick to using this is relying on an independent
48 * object validation pass. Something like:
52 * obj = lockless_lookup(key);
54 * if (!try_get_ref(obj)) // might fail for free objects
57 * if (obj->key != key) { // not the object we expected
64 * This is useful if we need to approach a kernel structure obliquely,
65 * from its address obtained without the usual locking. We can lock
66 * the structure to stabilize it and check it's still at the given address,
67 * only if we can be sure that the memory has not been meanwhile reused
68 * for some other kind of object (which our subsystem's lock might corrupt).
70 * rcu_read_lock before reading the address, then rcu_read_unlock after
71 * taking the spinlock within the structure expected at that address.
73 * Note that SLAB_TYPESAFE_BY_RCU was originally named SLAB_DESTROY_BY_RCU.
75 /* Defer freeing slabs to RCU */
76 #define SLAB_TYPESAFE_BY_RCU ((slab_flags_t __force)0x00080000UL)
77 /* Spread some memory over cpuset */
78 #define SLAB_MEM_SPREAD ((slab_flags_t __force)0x00100000UL)
79 /* Trace allocations and frees */
80 #define SLAB_TRACE ((slab_flags_t __force)0x00200000UL)
82 /* Flag to prevent checks on free */
83 #ifdef CONFIG_DEBUG_OBJECTS
84 # define SLAB_DEBUG_OBJECTS ((slab_flags_t __force)0x00400000UL)
86 # define SLAB_DEBUG_OBJECTS ((slab_flags_t __force)0x00000000UL)
89 /* Avoid kmemleak tracing */
90 #define SLAB_NOLEAKTRACE ((slab_flags_t __force)0x00800000UL)
92 /* Don't track use of uninitialized memory */
93 #ifdef CONFIG_KMEMCHECK
94 # define SLAB_NOTRACK ((slab_flags_t __force)0x01000000UL)
96 # define SLAB_NOTRACK ((slab_flags_t __force)0x00000000UL)
98 /* Fault injection mark */
99 #ifdef CONFIG_FAILSLAB
100 # define SLAB_FAILSLAB ((slab_flags_t __force)0x02000000UL)
102 # define SLAB_FAILSLAB ((slab_flags_t __force)0x00000000UL)
104 /* Account to memcg */
105 #if defined(CONFIG_MEMCG) && !defined(CONFIG_SLOB)
106 # define SLAB_ACCOUNT ((slab_flags_t __force)0x04000000UL)
108 # define SLAB_ACCOUNT ((slab_flags_t __force)0x00000000UL)
112 #define SLAB_KASAN ((slab_flags_t __force)0x08000000UL)
114 #define SLAB_KASAN ((slab_flags_t __force)0x00000000UL)
117 /* The following flags affect the page allocator grouping pages by mobility */
118 /* Objects are reclaimable */
119 #define SLAB_RECLAIM_ACCOUNT ((slab_flags_t __force)0x00020000UL)
120 #define SLAB_TEMPORARY SLAB_RECLAIM_ACCOUNT /* Objects are short-lived */
122 * ZERO_SIZE_PTR will be returned for zero sized kmalloc requests.
124 * Dereferencing ZERO_SIZE_PTR will lead to a distinct access fault.
126 * ZERO_SIZE_PTR can be passed to kfree though in the same way that NULL can.
127 * Both make kfree a no-op.
129 #define ZERO_SIZE_PTR ((void *)16)
131 #define ZERO_OR_NULL_PTR(x) ((unsigned long)(x) <= \
132 (unsigned long)ZERO_SIZE_PTR)
134 #include <linux/kmemleak.h>
135 #include <linux/kasan.h>
139 * struct kmem_cache related prototypes
141 void __init kmem_cache_init(void);
142 bool slab_is_available(void);
144 struct kmem_cache *kmem_cache_create(const char *, size_t, size_t,
147 void kmem_cache_destroy(struct kmem_cache *);
148 int kmem_cache_shrink(struct kmem_cache *);
150 void memcg_create_kmem_cache(struct mem_cgroup *, struct kmem_cache *);
151 void memcg_deactivate_kmem_caches(struct mem_cgroup *);
152 void memcg_destroy_kmem_caches(struct mem_cgroup *);
155 * Please use this macro to create slab caches. Simply specify the
156 * name of the structure and maybe some flags that are listed above.
158 * The alignment of the struct determines object alignment. If you
159 * f.e. add ____cacheline_aligned_in_smp to the struct declaration
160 * then the objects will be properly aligned in SMP configurations.
162 #define KMEM_CACHE(__struct, __flags) kmem_cache_create(#__struct,\
163 sizeof(struct __struct), __alignof__(struct __struct),\
167 * Common kmalloc functions provided by all allocators
169 void * __must_check __krealloc(const void *, size_t, gfp_t);
170 void * __must_check krealloc(const void *, size_t, gfp_t);
171 void kfree(const void *);
172 void kzfree(const void *);
173 size_t ksize(const void *);
175 #ifdef CONFIG_HAVE_HARDENED_USERCOPY_ALLOCATOR
176 const char *__check_heap_object(const void *ptr, unsigned long n,
179 static inline const char *__check_heap_object(const void *ptr,
188 * Some archs want to perform DMA into kmalloc caches and need a guaranteed
189 * alignment larger than the alignment of a 64-bit integer.
190 * Setting ARCH_KMALLOC_MINALIGN in arch headers allows that.
192 #if defined(ARCH_DMA_MINALIGN) && ARCH_DMA_MINALIGN > 8
193 #define ARCH_KMALLOC_MINALIGN ARCH_DMA_MINALIGN
194 #define KMALLOC_MIN_SIZE ARCH_DMA_MINALIGN
195 #define KMALLOC_SHIFT_LOW ilog2(ARCH_DMA_MINALIGN)
197 #define ARCH_KMALLOC_MINALIGN __alignof__(unsigned long long)
201 * Setting ARCH_SLAB_MINALIGN in arch headers allows a different alignment.
202 * Intended for arches that get misalignment faults even for 64 bit integer
205 #ifndef ARCH_SLAB_MINALIGN
206 #define ARCH_SLAB_MINALIGN __alignof__(unsigned long long)
210 * kmalloc and friends return ARCH_KMALLOC_MINALIGN aligned
211 * pointers. kmem_cache_alloc and friends return ARCH_SLAB_MINALIGN
214 #define __assume_kmalloc_alignment __assume_aligned(ARCH_KMALLOC_MINALIGN)
215 #define __assume_slab_alignment __assume_aligned(ARCH_SLAB_MINALIGN)
216 #define __assume_page_alignment __assume_aligned(PAGE_SIZE)
219 * Kmalloc array related definitions
224 * The largest kmalloc size supported by the SLAB allocators is
225 * 32 megabyte (2^25) or the maximum allocatable page order if that is
228 * WARNING: Its not easy to increase this value since the allocators have
229 * to do various tricks to work around compiler limitations in order to
230 * ensure proper constant folding.
232 #define KMALLOC_SHIFT_HIGH ((MAX_ORDER + PAGE_SHIFT - 1) <= 25 ? \
233 (MAX_ORDER + PAGE_SHIFT - 1) : 25)
234 #define KMALLOC_SHIFT_MAX KMALLOC_SHIFT_HIGH
235 #ifndef KMALLOC_SHIFT_LOW
236 #define KMALLOC_SHIFT_LOW 5
242 * SLUB directly allocates requests fitting in to an order-1 page
243 * (PAGE_SIZE*2). Larger requests are passed to the page allocator.
245 #define KMALLOC_SHIFT_HIGH (PAGE_SHIFT + 1)
246 #define KMALLOC_SHIFT_MAX (MAX_ORDER + PAGE_SHIFT - 1)
247 #ifndef KMALLOC_SHIFT_LOW
248 #define KMALLOC_SHIFT_LOW 3
254 * SLOB passes all requests larger than one page to the page allocator.
255 * No kmalloc array is necessary since objects of different sizes can
256 * be allocated from the same page.
258 #define KMALLOC_SHIFT_HIGH PAGE_SHIFT
259 #define KMALLOC_SHIFT_MAX (MAX_ORDER + PAGE_SHIFT - 1)
260 #ifndef KMALLOC_SHIFT_LOW
261 #define KMALLOC_SHIFT_LOW 3
265 /* Maximum allocatable size */
266 #define KMALLOC_MAX_SIZE (1UL << KMALLOC_SHIFT_MAX)
267 /* Maximum size for which we actually use a slab cache */
268 #define KMALLOC_MAX_CACHE_SIZE (1UL << KMALLOC_SHIFT_HIGH)
269 /* Maximum order allocatable via the slab allocagtor */
270 #define KMALLOC_MAX_ORDER (KMALLOC_SHIFT_MAX - PAGE_SHIFT)
275 #ifndef KMALLOC_MIN_SIZE
276 #define KMALLOC_MIN_SIZE (1 << KMALLOC_SHIFT_LOW)
280 * This restriction comes from byte sized index implementation.
281 * Page size is normally 2^12 bytes and, in this case, if we want to use
282 * byte sized index which can represent 2^8 entries, the size of the object
283 * should be equal or greater to 2^12 / 2^8 = 2^4 = 16.
284 * If minimum size of kmalloc is less than 16, we use it as minimum object
285 * size and give up to use byte sized index.
287 #define SLAB_OBJ_MIN_SIZE (KMALLOC_MIN_SIZE < 16 ? \
288 (KMALLOC_MIN_SIZE) : 16)
291 extern struct kmem_cache *kmalloc_caches[KMALLOC_SHIFT_HIGH + 1];
292 #ifdef CONFIG_ZONE_DMA
293 extern struct kmem_cache *kmalloc_dma_caches[KMALLOC_SHIFT_HIGH + 1];
297 * Figure out which kmalloc slab an allocation of a certain size
301 * 2 = 129 .. 192 bytes
302 * n = 2^(n-1)+1 .. 2^n
304 static __always_inline int kmalloc_index(size_t size)
309 if (size <= KMALLOC_MIN_SIZE)
310 return KMALLOC_SHIFT_LOW;
312 if (KMALLOC_MIN_SIZE <= 32 && size > 64 && size <= 96)
314 if (KMALLOC_MIN_SIZE <= 64 && size > 128 && size <= 192)
316 if (size <= 8) return 3;
317 if (size <= 16) return 4;
318 if (size <= 32) return 5;
319 if (size <= 64) return 6;
320 if (size <= 128) return 7;
321 if (size <= 256) return 8;
322 if (size <= 512) return 9;
323 if (size <= 1024) return 10;
324 if (size <= 2 * 1024) return 11;
325 if (size <= 4 * 1024) return 12;
326 if (size <= 8 * 1024) return 13;
327 if (size <= 16 * 1024) return 14;
328 if (size <= 32 * 1024) return 15;
329 if (size <= 64 * 1024) return 16;
330 if (size <= 128 * 1024) return 17;
331 if (size <= 256 * 1024) return 18;
332 if (size <= 512 * 1024) return 19;
333 if (size <= 1024 * 1024) return 20;
334 if (size <= 2 * 1024 * 1024) return 21;
335 if (size <= 4 * 1024 * 1024) return 22;
336 if (size <= 8 * 1024 * 1024) return 23;
337 if (size <= 16 * 1024 * 1024) return 24;
338 if (size <= 32 * 1024 * 1024) return 25;
339 if (size <= 64 * 1024 * 1024) return 26;
342 /* Will never be reached. Needed because the compiler may complain */
345 #endif /* !CONFIG_SLOB */
347 void *__kmalloc(size_t size, gfp_t flags) __assume_kmalloc_alignment __malloc;
348 void *kmem_cache_alloc(struct kmem_cache *, gfp_t flags) __assume_slab_alignment __malloc;
349 void kmem_cache_free(struct kmem_cache *, void *);
352 * Bulk allocation and freeing operations. These are accelerated in an
353 * allocator specific way to avoid taking locks repeatedly or building
354 * metadata structures unnecessarily.
356 * Note that interrupts must be enabled when calling these functions.
358 void kmem_cache_free_bulk(struct kmem_cache *, size_t, void **);
359 int kmem_cache_alloc_bulk(struct kmem_cache *, gfp_t, size_t, void **);
362 * Caller must not use kfree_bulk() on memory not originally allocated
363 * by kmalloc(), because the SLOB allocator cannot handle this.
365 static __always_inline void kfree_bulk(size_t size, void **p)
367 kmem_cache_free_bulk(NULL, size, p);
371 void *__kmalloc_node(size_t size, gfp_t flags, int node) __assume_kmalloc_alignment __malloc;
372 void *kmem_cache_alloc_node(struct kmem_cache *, gfp_t flags, int node) __assume_slab_alignment __malloc;
374 static __always_inline void *__kmalloc_node(size_t size, gfp_t flags, int node)
376 return __kmalloc(size, flags);
379 static __always_inline void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t flags, int node)
381 return kmem_cache_alloc(s, flags);
385 #ifdef CONFIG_TRACING
386 extern void *kmem_cache_alloc_trace(struct kmem_cache *, gfp_t, size_t) __assume_slab_alignment __malloc;
389 extern void *kmem_cache_alloc_node_trace(struct kmem_cache *s,
391 int node, size_t size) __assume_slab_alignment __malloc;
393 static __always_inline void *
394 kmem_cache_alloc_node_trace(struct kmem_cache *s,
396 int node, size_t size)
398 return kmem_cache_alloc_trace(s, gfpflags, size);
400 #endif /* CONFIG_NUMA */
402 #else /* CONFIG_TRACING */
403 static __always_inline void *kmem_cache_alloc_trace(struct kmem_cache *s,
404 gfp_t flags, size_t size)
406 void *ret = kmem_cache_alloc(s, flags);
408 kasan_kmalloc(s, ret, size, flags);
412 static __always_inline void *
413 kmem_cache_alloc_node_trace(struct kmem_cache *s,
415 int node, size_t size)
417 void *ret = kmem_cache_alloc_node(s, gfpflags, node);
419 kasan_kmalloc(s, ret, size, gfpflags);
422 #endif /* CONFIG_TRACING */
424 extern void *kmalloc_order(size_t size, gfp_t flags, unsigned int order) __assume_page_alignment __malloc;
426 #ifdef CONFIG_TRACING
427 extern void *kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order) __assume_page_alignment __malloc;
429 static __always_inline void *
430 kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order)
432 return kmalloc_order(size, flags, order);
436 static __always_inline void *kmalloc_large(size_t size, gfp_t flags)
438 unsigned int order = get_order(size);
439 return kmalloc_order_trace(size, flags, order);
443 * kmalloc - allocate memory
444 * @size: how many bytes of memory are required.
445 * @flags: the type of memory to allocate.
447 * kmalloc is the normal method of allocating memory
448 * for objects smaller than page size in the kernel.
450 * The @flags argument may be one of:
452 * %GFP_USER - Allocate memory on behalf of user. May sleep.
454 * %GFP_KERNEL - Allocate normal kernel ram. May sleep.
456 * %GFP_ATOMIC - Allocation will not sleep. May use emergency pools.
457 * For example, use this inside interrupt handlers.
459 * %GFP_HIGHUSER - Allocate pages from high memory.
461 * %GFP_NOIO - Do not do any I/O at all while trying to get memory.
463 * %GFP_NOFS - Do not make any fs calls while trying to get memory.
465 * %GFP_NOWAIT - Allocation will not sleep.
467 * %__GFP_THISNODE - Allocate node-local memory only.
469 * %GFP_DMA - Allocation suitable for DMA.
470 * Should only be used for kmalloc() caches. Otherwise, use a
471 * slab created with SLAB_DMA.
473 * Also it is possible to set different flags by OR'ing
474 * in one or more of the following additional @flags:
476 * %__GFP_COLD - Request cache-cold pages instead of
477 * trying to return cache-warm pages.
479 * %__GFP_HIGH - This allocation has high priority and may use emergency pools.
481 * %__GFP_NOFAIL - Indicate that this allocation is in no way allowed to fail
482 * (think twice before using).
484 * %__GFP_NORETRY - If memory is not immediately available,
485 * then give up at once.
487 * %__GFP_NOWARN - If allocation fails, don't issue any warnings.
489 * %__GFP_RETRY_MAYFAIL - Try really hard to succeed the allocation but fail
492 * There are other flags available as well, but these are not intended
493 * for general use, and so are not documented here. For a full list of
494 * potential flags, always refer to linux/gfp.h.
496 static __always_inline void *kmalloc(size_t size, gfp_t flags)
498 if (__builtin_constant_p(size)) {
499 if (size > KMALLOC_MAX_CACHE_SIZE)
500 return kmalloc_large(size, flags);
502 if (!(flags & GFP_DMA)) {
503 int index = kmalloc_index(size);
506 return ZERO_SIZE_PTR;
508 return kmem_cache_alloc_trace(kmalloc_caches[index],
513 return __kmalloc(size, flags);
517 * Determine size used for the nth kmalloc cache.
518 * return size or 0 if a kmalloc cache for that
519 * size does not exist
521 static __always_inline int kmalloc_size(int n)
527 if (n == 1 && KMALLOC_MIN_SIZE <= 32)
530 if (n == 2 && KMALLOC_MIN_SIZE <= 64)
536 static __always_inline void *kmalloc_node(size_t size, gfp_t flags, int node)
539 if (__builtin_constant_p(size) &&
540 size <= KMALLOC_MAX_CACHE_SIZE && !(flags & GFP_DMA)) {
541 int i = kmalloc_index(size);
544 return ZERO_SIZE_PTR;
546 return kmem_cache_alloc_node_trace(kmalloc_caches[i],
550 return __kmalloc_node(size, flags, node);
553 struct memcg_cache_array {
555 struct kmem_cache *entries[0];
559 * This is the main placeholder for memcg-related information in kmem caches.
560 * Both the root cache and the child caches will have it. For the root cache,
561 * this will hold a dynamically allocated array large enough to hold
562 * information about the currently limited memcgs in the system. To allow the
563 * array to be accessed without taking any locks, on relocation we free the old
564 * version only after a grace period.
566 * Root and child caches hold different metadata.
568 * @root_cache: Common to root and child caches. NULL for root, pointer to
569 * the root cache for children.
571 * The following fields are specific to root caches.
573 * @memcg_caches: kmemcg ID indexed table of child caches. This table is
574 * used to index child cachces during allocation and cleared
575 * early during shutdown.
577 * @root_caches_node: List node for slab_root_caches list.
579 * @children: List of all child caches. While the child caches are also
580 * reachable through @memcg_caches, a child cache remains on
581 * this list until it is actually destroyed.
583 * The following fields are specific to child caches.
585 * @memcg: Pointer to the memcg this cache belongs to.
587 * @children_node: List node for @root_cache->children list.
589 * @kmem_caches_node: List node for @memcg->kmem_caches list.
591 struct memcg_cache_params {
592 struct kmem_cache *root_cache;
595 struct memcg_cache_array __rcu *memcg_caches;
596 struct list_head __root_caches_node;
597 struct list_head children;
600 struct mem_cgroup *memcg;
601 struct list_head children_node;
602 struct list_head kmem_caches_node;
604 void (*deact_fn)(struct kmem_cache *);
606 struct rcu_head deact_rcu_head;
607 struct work_struct deact_work;
613 int memcg_update_all_caches(int num_memcgs);
616 * kmalloc_array - allocate memory for an array.
617 * @n: number of elements.
618 * @size: element size.
619 * @flags: the type of memory to allocate (see kmalloc).
621 static inline void *kmalloc_array(size_t n, size_t size, gfp_t flags)
623 if (size != 0 && n > SIZE_MAX / size)
625 if (__builtin_constant_p(n) && __builtin_constant_p(size))
626 return kmalloc(n * size, flags);
627 return __kmalloc(n * size, flags);
631 * kcalloc - allocate memory for an array. The memory is set to zero.
632 * @n: number of elements.
633 * @size: element size.
634 * @flags: the type of memory to allocate (see kmalloc).
636 static inline void *kcalloc(size_t n, size_t size, gfp_t flags)
638 return kmalloc_array(n, size, flags | __GFP_ZERO);
642 * kmalloc_track_caller is a special version of kmalloc that records the
643 * calling function of the routine calling it for slab leak tracking instead
644 * of just the calling function (confusing, eh?).
645 * It's useful when the call to kmalloc comes from a widely-used standard
646 * allocator where we care about the real place the memory allocation
647 * request comes from.
649 extern void *__kmalloc_track_caller(size_t, gfp_t, unsigned long);
650 #define kmalloc_track_caller(size, flags) \
651 __kmalloc_track_caller(size, flags, _RET_IP_)
654 extern void *__kmalloc_node_track_caller(size_t, gfp_t, int, unsigned long);
655 #define kmalloc_node_track_caller(size, flags, node) \
656 __kmalloc_node_track_caller(size, flags, node, \
659 #else /* CONFIG_NUMA */
661 #define kmalloc_node_track_caller(size, flags, node) \
662 kmalloc_track_caller(size, flags)
664 #endif /* CONFIG_NUMA */
669 static inline void *kmem_cache_zalloc(struct kmem_cache *k, gfp_t flags)
671 return kmem_cache_alloc(k, flags | __GFP_ZERO);
675 * kzalloc - allocate memory. The memory is set to zero.
676 * @size: how many bytes of memory are required.
677 * @flags: the type of memory to allocate (see kmalloc).
679 static inline void *kzalloc(size_t size, gfp_t flags)
681 return kmalloc(size, flags | __GFP_ZERO);
685 * kzalloc_node - allocate zeroed memory from a particular memory node.
686 * @size: how many bytes of memory are required.
687 * @flags: the type of memory to allocate (see kmalloc).
688 * @node: memory node from which to allocate
690 static inline void *kzalloc_node(size_t size, gfp_t flags, int node)
692 return kmalloc_node(size, flags | __GFP_ZERO, node);
695 unsigned int kmem_cache_size(struct kmem_cache *s);
696 void __init kmem_cache_init_late(void);
698 #if defined(CONFIG_SMP) && defined(CONFIG_SLAB)
699 int slab_prepare_cpu(unsigned int cpu);
700 int slab_dead_cpu(unsigned int cpu);
702 #define slab_prepare_cpu NULL
703 #define slab_dead_cpu NULL
706 #endif /* _LINUX_SLAB_H */