blob: 9a139b637069730e85fe6903253baf16e211e7b0 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
Christoph Lameter2e892f42006-12-13 00:34:23 -08002 * Written by Mark Hemment, 1996 (markhe@nextd.demon.co.uk).
3 *
Christoph Lametercde53532008-07-04 09:59:22 -07004 * (C) SGI 2006, Christoph Lameter
Christoph Lameter2e892f42006-12-13 00:34:23 -08005 * Cleaned up and restructured to ease the addition of alternative
6 * implementations of SLAB allocators.
Christoph Lameterf1b6eb62013-09-04 16:35:34 +00007 * (C) Linux Foundation 2008-2013
8 * Unified interface for all slab allocators
Linus Torvalds1da177e2005-04-16 15:20:36 -07009 */
10
11#ifndef _LINUX_SLAB_H
12#define _LINUX_SLAB_H
13
Andrew Morton1b1cec42006-12-06 20:33:22 -080014#include <linux/gfp.h>
Andrew Morton1b1cec42006-12-06 20:33:22 -080015#include <linux/types.h>
Glauber Costa1f458cb2012-12-18 14:22:50 -080016#include <linux/workqueue.h>
17
Linus Torvalds1da177e2005-04-16 15:20:36 -070018
Christoph Lameter2e892f42006-12-13 00:34:23 -080019/*
20 * Flags to pass to kmem_cache_create().
21 * The ones marked DEBUG are only valid if CONFIG_SLAB_DEBUG is set.
Linus Torvalds1da177e2005-04-16 15:20:36 -070022 */
Christoph Lameter55935a32006-12-13 00:34:24 -080023#define SLAB_DEBUG_FREE 0x00000100UL /* DEBUG: Perform (expensive) checks on free */
Christoph Lameter55935a32006-12-13 00:34:24 -080024#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 */
Christoph Lameter2e892f42006-12-13 00:34:23 -080027#define SLAB_CACHE_DMA 0x00004000UL /* Use GFP_DMA memory */
Christoph Lameter2e892f42006-12-13 00:34:23 -080028#define SLAB_STORE_USER 0x00010000UL /* DEBUG: Store the last owner for bug hunting */
Christoph Lameter2e892f42006-12-13 00:34:23 -080029#define SLAB_PANIC 0x00040000UL /* Panic if kmem_cache_create() fails */
Peter Zijlstrad7de4c12008-11-13 20:40:12 +020030/*
31 * SLAB_DESTROY_BY_RCU - **WARNING** READ THIS!
32 *
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.
37 *
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:
41 *
42 * rcu_read_lock()
43 * again:
44 * obj = lockless_lookup(key);
45 * if (obj) {
46 * if (!try_get_ref(obj)) // might fail for free objects
47 * goto again;
48 *
49 * if (obj->key != key) { // not the object we expected
50 * put_ref(obj);
51 * goto again;
52 * }
53 * }
54 * rcu_read_unlock();
55 *
Joonsoo Kim68126702013-10-24 10:07:42 +090056 * 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).
61 *
62 * rcu_read_lock before reading the address, then rcu_read_unlock after
63 * taking the spinlock within the structure expected at that address.
Peter Zijlstrad7de4c12008-11-13 20:40:12 +020064 */
Christoph Lameter2e892f42006-12-13 00:34:23 -080065#define SLAB_DESTROY_BY_RCU 0x00080000UL /* Defer freeing slabs to RCU */
Paul Jackson101a5002006-03-24 03:16:07 -080066#define SLAB_MEM_SPREAD 0x00100000UL /* Spread some memory over cpuset */
Christoph Lameter81819f02007-05-06 14:49:36 -070067#define SLAB_TRACE 0x00200000UL /* Trace allocations and frees */
Linus Torvalds1da177e2005-04-16 15:20:36 -070068
Thomas Gleixner30327ac2008-04-30 00:54:59 -070069/* Flag to prevent checks on free */
70#ifdef CONFIG_DEBUG_OBJECTS
71# define SLAB_DEBUG_OBJECTS 0x00400000UL
72#else
73# define SLAB_DEBUG_OBJECTS 0x00000000UL
74#endif
75
Catalin Marinasd5cff632009-06-11 13:22:40 +010076#define SLAB_NOLEAKTRACE 0x00800000UL /* Avoid kmemleak tracing */
77
Vegard Nossum2dff4402008-05-31 15:56:17 +020078/* Don't track use of uninitialized memory */
79#ifdef CONFIG_KMEMCHECK
80# define SLAB_NOTRACK 0x01000000UL
81#else
82# define SLAB_NOTRACK 0x00000000UL
83#endif
Dmitry Monakhov4c13dd32010-02-26 09:36:12 +030084#ifdef CONFIG_FAILSLAB
85# define SLAB_FAILSLAB 0x02000000UL /* Fault injection mark */
86#else
87# define SLAB_FAILSLAB 0x00000000UL
88#endif
Vegard Nossum2dff4402008-05-31 15:56:17 +020089
Mel Gormane12ba742007-10-16 01:25:52 -070090/* 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 */
Christoph Lameter2e892f42006-12-13 00:34:23 -080093/*
Christoph Lameter6cb8f912007-07-17 04:03:22 -070094 * ZERO_SIZE_PTR will be returned for zero sized kmalloc requests.
95 *
96 * Dereferencing ZERO_SIZE_PTR will lead to a distinct access fault.
97 *
98 * ZERO_SIZE_PTR can be passed to kfree though in the same way that NULL can.
99 * Both make kfree a no-op.
100 */
101#define ZERO_SIZE_PTR ((void *)16)
102
Roland Dreier1d4ec7b2007-07-20 12:13:20 -0700103#define ZERO_OR_NULL_PTR(x) ((unsigned long)(x) <= \
Christoph Lameter6cb8f912007-07-17 04:03:22 -0700104 (unsigned long)ZERO_SIZE_PTR)
105
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000106#include <linux/kmemleak.h>
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500107
Glauber Costa2633d7a2012-12-18 14:22:34 -0800108struct mem_cgroup;
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500109/*
Christoph Lameter2e892f42006-12-13 00:34:23 -0800110 * struct kmem_cache related prototypes
111 */
112void __init kmem_cache_init(void);
Christoph Lameter81819f02007-05-06 14:49:36 -0700113int slab_is_available(void);
Matt Mackall10cef602006-01-08 01:01:45 -0800114
Christoph Lameter2e892f42006-12-13 00:34:23 -0800115struct kmem_cache *kmem_cache_create(const char *, size_t, size_t,
Christoph Lameterebe29732006-12-06 20:32:59 -0800116 unsigned long,
Alexey Dobriyan51cc5062008-07-25 19:45:34 -0700117 void (*)(void *));
Vladimir Davydov794b1242014-04-07 15:39:26 -0700118#ifdef CONFIG_MEMCG_KMEM
Vladimir Davydov776ed0f2014-06-04 16:10:02 -0700119struct kmem_cache *memcg_create_kmem_cache(struct mem_cgroup *,
Vladimir Davydov073ee1c2014-06-04 16:08:23 -0700120 struct kmem_cache *,
121 const char *);
Vladimir Davydov794b1242014-04-07 15:39:26 -0700122#endif
Christoph Lameter2e892f42006-12-13 00:34:23 -0800123void kmem_cache_destroy(struct kmem_cache *);
124int kmem_cache_shrink(struct kmem_cache *);
Christoph Lameter2e892f42006-12-13 00:34:23 -0800125void kmem_cache_free(struct kmem_cache *, void *);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700126
Christoph Lameter0a31bd52007-05-06 14:49:57 -0700127/*
128 * Please use this macro to create slab caches. Simply specify the
129 * name of the structure and maybe some flags that are listed above.
130 *
131 * The alignment of the struct determines object alignment. If you
132 * f.e. add ____cacheline_aligned_in_smp to the struct declaration
133 * then the objects will be properly aligned in SMP configurations.
134 */
135#define KMEM_CACHE(__struct, __flags) kmem_cache_create(#__struct,\
136 sizeof(struct __struct), __alignof__(struct __struct),\
Paul Mundt20c2df82007-07-20 10:11:58 +0900137 (__flags), NULL)
Christoph Lameter0a31bd52007-05-06 14:49:57 -0700138
Christoph Lameter2e892f42006-12-13 00:34:23 -0800139/*
Christoph Lameter34504662013-01-10 19:00:53 +0000140 * Common kmalloc functions provided by all allocators
141 */
142void * __must_check __krealloc(const void *, size_t, gfp_t);
143void * __must_check krealloc(const void *, size_t, gfp_t);
144void kfree(const void *);
145void kzfree(const void *);
146size_t ksize(const void *);
147
Christoph Lameterc601fd62013-02-05 16:36:47 +0000148/*
149 * Some archs want to perform DMA into kmalloc caches and need a guaranteed
150 * alignment larger than the alignment of a 64-bit integer.
151 * Setting ARCH_KMALLOC_MINALIGN in arch headers allows that.
152 */
153#if defined(ARCH_DMA_MINALIGN) && ARCH_DMA_MINALIGN > 8
154#define ARCH_KMALLOC_MINALIGN ARCH_DMA_MINALIGN
155#define KMALLOC_MIN_SIZE ARCH_DMA_MINALIGN
156#define KMALLOC_SHIFT_LOW ilog2(ARCH_DMA_MINALIGN)
157#else
158#define ARCH_KMALLOC_MINALIGN __alignof__(unsigned long long)
159#endif
160
Christoph Lameter34504662013-01-10 19:00:53 +0000161/*
Christoph Lameter95a05b42013-01-10 19:14:19 +0000162 * Kmalloc array related definitions
163 */
164
165#ifdef CONFIG_SLAB
166/*
167 * The largest kmalloc size supported by the SLAB allocators is
Christoph Lameter0aa817f2007-05-16 22:11:01 -0700168 * 32 megabyte (2^25) or the maximum allocatable page order if that is
169 * less than 32 MB.
170 *
171 * WARNING: Its not easy to increase this value since the allocators have
172 * to do various tricks to work around compiler limitations in order to
173 * ensure proper constant folding.
174 */
Christoph Lameterdebee072007-06-23 17:16:43 -0700175#define KMALLOC_SHIFT_HIGH ((MAX_ORDER + PAGE_SHIFT - 1) <= 25 ? \
176 (MAX_ORDER + PAGE_SHIFT - 1) : 25)
Christoph Lameter95a05b42013-01-10 19:14:19 +0000177#define KMALLOC_SHIFT_MAX KMALLOC_SHIFT_HIGH
Christoph Lameterc601fd62013-02-05 16:36:47 +0000178#ifndef KMALLOC_SHIFT_LOW
Christoph Lameter95a05b42013-01-10 19:14:19 +0000179#define KMALLOC_SHIFT_LOW 5
Christoph Lameterc601fd62013-02-05 16:36:47 +0000180#endif
Christoph Lameter069e2b352013-06-14 19:55:13 +0000181#endif
182
183#ifdef CONFIG_SLUB
Christoph Lameter95a05b42013-01-10 19:14:19 +0000184/*
Dave Hansen433a91f2014-01-28 14:24:50 -0800185 * SLUB directly allocates requests fitting in to an order-1 page
186 * (PAGE_SIZE*2). Larger requests are passed to the page allocator.
Christoph Lameter95a05b42013-01-10 19:14:19 +0000187 */
188#define KMALLOC_SHIFT_HIGH (PAGE_SHIFT + 1)
189#define KMALLOC_SHIFT_MAX (MAX_ORDER + PAGE_SHIFT)
Christoph Lameterc601fd62013-02-05 16:36:47 +0000190#ifndef KMALLOC_SHIFT_LOW
Christoph Lameter95a05b42013-01-10 19:14:19 +0000191#define KMALLOC_SHIFT_LOW 3
192#endif
Christoph Lameterc601fd62013-02-05 16:36:47 +0000193#endif
Christoph Lameter0aa817f2007-05-16 22:11:01 -0700194
Christoph Lameter069e2b352013-06-14 19:55:13 +0000195#ifdef CONFIG_SLOB
196/*
Dave Hansen433a91f2014-01-28 14:24:50 -0800197 * SLOB passes all requests larger than one page to the page allocator.
Christoph Lameter069e2b352013-06-14 19:55:13 +0000198 * No kmalloc array is necessary since objects of different sizes can
199 * be allocated from the same page.
200 */
Christoph Lameter069e2b352013-06-14 19:55:13 +0000201#define KMALLOC_SHIFT_HIGH PAGE_SHIFT
Dave Hansen433a91f2014-01-28 14:24:50 -0800202#define KMALLOC_SHIFT_MAX 30
Christoph Lameter069e2b352013-06-14 19:55:13 +0000203#ifndef KMALLOC_SHIFT_LOW
204#define KMALLOC_SHIFT_LOW 3
205#endif
206#endif
207
Christoph Lameter95a05b42013-01-10 19:14:19 +0000208/* Maximum allocatable size */
209#define KMALLOC_MAX_SIZE (1UL << KMALLOC_SHIFT_MAX)
210/* Maximum size for which we actually use a slab cache */
211#define KMALLOC_MAX_CACHE_SIZE (1UL << KMALLOC_SHIFT_HIGH)
212/* Maximum order allocatable via the slab allocagtor */
213#define KMALLOC_MAX_ORDER (KMALLOC_SHIFT_MAX - PAGE_SHIFT)
Christoph Lameter0aa817f2007-05-16 22:11:01 -0700214
Christoph Lameter90810642011-06-23 09:36:12 -0500215/*
Christoph Lameterce6a5022013-01-10 19:14:19 +0000216 * Kmalloc subsystem.
217 */
Christoph Lameterc601fd62013-02-05 16:36:47 +0000218#ifndef KMALLOC_MIN_SIZE
Christoph Lameter95a05b42013-01-10 19:14:19 +0000219#define KMALLOC_MIN_SIZE (1 << KMALLOC_SHIFT_LOW)
Christoph Lameterce6a5022013-01-10 19:14:19 +0000220#endif
Christoph Lameterce6a5022013-01-10 19:14:19 +0000221
Joonsoo Kim24f870d2014-03-12 17:06:19 +0900222/*
223 * This restriction comes from byte sized index implementation.
224 * Page size is normally 2^12 bytes and, in this case, if we want to use
225 * byte sized index which can represent 2^8 entries, the size of the object
226 * should be equal or greater to 2^12 / 2^8 = 2^4 = 16.
227 * If minimum size of kmalloc is less than 16, we use it as minimum object
228 * size and give up to use byte sized index.
229 */
230#define SLAB_OBJ_MIN_SIZE (KMALLOC_MIN_SIZE < 16 ? \
231 (KMALLOC_MIN_SIZE) : 16)
232
Christoph Lameter069e2b352013-06-14 19:55:13 +0000233#ifndef CONFIG_SLOB
Christoph Lameter9425c582013-01-10 19:12:17 +0000234extern struct kmem_cache *kmalloc_caches[KMALLOC_SHIFT_HIGH + 1];
235#ifdef CONFIG_ZONE_DMA
236extern struct kmem_cache *kmalloc_dma_caches[KMALLOC_SHIFT_HIGH + 1];
237#endif
238
Christoph Lameterce6a5022013-01-10 19:14:19 +0000239/*
240 * Figure out which kmalloc slab an allocation of a certain size
241 * belongs to.
242 * 0 = zero alloc
243 * 1 = 65 .. 96 bytes
244 * 2 = 120 .. 192 bytes
245 * n = 2^(n-1) .. 2^n -1
246 */
247static __always_inline int kmalloc_index(size_t size)
248{
249 if (!size)
250 return 0;
251
252 if (size <= KMALLOC_MIN_SIZE)
253 return KMALLOC_SHIFT_LOW;
254
255 if (KMALLOC_MIN_SIZE <= 32 && size > 64 && size <= 96)
256 return 1;
257 if (KMALLOC_MIN_SIZE <= 64 && size > 128 && size <= 192)
258 return 2;
259 if (size <= 8) return 3;
260 if (size <= 16) return 4;
261 if (size <= 32) return 5;
262 if (size <= 64) return 6;
263 if (size <= 128) return 7;
264 if (size <= 256) return 8;
265 if (size <= 512) return 9;
266 if (size <= 1024) return 10;
267 if (size <= 2 * 1024) return 11;
268 if (size <= 4 * 1024) return 12;
269 if (size <= 8 * 1024) return 13;
270 if (size <= 16 * 1024) return 14;
271 if (size <= 32 * 1024) return 15;
272 if (size <= 64 * 1024) return 16;
273 if (size <= 128 * 1024) return 17;
274 if (size <= 256 * 1024) return 18;
275 if (size <= 512 * 1024) return 19;
276 if (size <= 1024 * 1024) return 20;
277 if (size <= 2 * 1024 * 1024) return 21;
278 if (size <= 4 * 1024 * 1024) return 22;
279 if (size <= 8 * 1024 * 1024) return 23;
280 if (size <= 16 * 1024 * 1024) return 24;
281 if (size <= 32 * 1024 * 1024) return 25;
282 if (size <= 64 * 1024 * 1024) return 26;
283 BUG();
284
285 /* Will never be reached. Needed because the compiler may complain */
286 return -1;
287}
Christoph Lameter069e2b352013-06-14 19:55:13 +0000288#endif /* !CONFIG_SLOB */
Christoph Lameterce6a5022013-01-10 19:14:19 +0000289
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000290void *__kmalloc(size_t size, gfp_t flags);
291void *kmem_cache_alloc(struct kmem_cache *, gfp_t flags);
292
293#ifdef CONFIG_NUMA
294void *__kmalloc_node(size_t size, gfp_t flags, int node);
295void *kmem_cache_alloc_node(struct kmem_cache *, gfp_t flags, int node);
296#else
297static __always_inline void *__kmalloc_node(size_t size, gfp_t flags, int node)
298{
299 return __kmalloc(size, flags);
300}
301
302static __always_inline void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t flags, int node)
303{
304 return kmem_cache_alloc(s, flags);
305}
306#endif
307
308#ifdef CONFIG_TRACING
309extern void *kmem_cache_alloc_trace(struct kmem_cache *, gfp_t, size_t);
310
311#ifdef CONFIG_NUMA
312extern void *kmem_cache_alloc_node_trace(struct kmem_cache *s,
313 gfp_t gfpflags,
314 int node, size_t size);
315#else
316static __always_inline void *
317kmem_cache_alloc_node_trace(struct kmem_cache *s,
318 gfp_t gfpflags,
319 int node, size_t size)
320{
321 return kmem_cache_alloc_trace(s, gfpflags, size);
322}
323#endif /* CONFIG_NUMA */
324
325#else /* CONFIG_TRACING */
326static __always_inline void *kmem_cache_alloc_trace(struct kmem_cache *s,
327 gfp_t flags, size_t size)
328{
329 return kmem_cache_alloc(s, flags);
330}
331
332static __always_inline void *
333kmem_cache_alloc_node_trace(struct kmem_cache *s,
334 gfp_t gfpflags,
335 int node, size_t size)
336{
337 return kmem_cache_alloc_node(s, gfpflags, node);
338}
339#endif /* CONFIG_TRACING */
340
Vladimir Davydov52383432014-06-04 16:06:39 -0700341extern void *kmalloc_order(size_t size, gfp_t flags, unsigned int order);
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000342
343#ifdef CONFIG_TRACING
344extern void *kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order);
345#else
346static __always_inline void *
347kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order)
348{
349 return kmalloc_order(size, flags, order);
350}
Christoph Lameterce6a5022013-01-10 19:14:19 +0000351#endif
352
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000353static __always_inline void *kmalloc_large(size_t size, gfp_t flags)
354{
355 unsigned int order = get_order(size);
356 return kmalloc_order_trace(size, flags, order);
357}
358
359/**
360 * kmalloc - allocate memory
361 * @size: how many bytes of memory are required.
Randy Dunlap7e3528c2013-11-22 18:14:38 -0800362 * @flags: the type of memory to allocate.
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000363 *
364 * kmalloc is the normal method of allocating memory
365 * for objects smaller than page size in the kernel.
Randy Dunlap7e3528c2013-11-22 18:14:38 -0800366 *
367 * The @flags argument may be one of:
368 *
369 * %GFP_USER - Allocate memory on behalf of user. May sleep.
370 *
371 * %GFP_KERNEL - Allocate normal kernel ram. May sleep.
372 *
373 * %GFP_ATOMIC - Allocation will not sleep. May use emergency pools.
374 * For example, use this inside interrupt handlers.
375 *
376 * %GFP_HIGHUSER - Allocate pages from high memory.
377 *
378 * %GFP_NOIO - Do not do any I/O at all while trying to get memory.
379 *
380 * %GFP_NOFS - Do not make any fs calls while trying to get memory.
381 *
382 * %GFP_NOWAIT - Allocation will not sleep.
383 *
Johannes Weinere97ca8e2014-03-10 15:49:43 -0700384 * %__GFP_THISNODE - Allocate node-local memory only.
Randy Dunlap7e3528c2013-11-22 18:14:38 -0800385 *
386 * %GFP_DMA - Allocation suitable for DMA.
387 * Should only be used for kmalloc() caches. Otherwise, use a
388 * slab created with SLAB_DMA.
389 *
390 * Also it is possible to set different flags by OR'ing
391 * in one or more of the following additional @flags:
392 *
393 * %__GFP_COLD - Request cache-cold pages instead of
394 * trying to return cache-warm pages.
395 *
396 * %__GFP_HIGH - This allocation has high priority and may use emergency pools.
397 *
398 * %__GFP_NOFAIL - Indicate that this allocation is in no way allowed to fail
399 * (think twice before using).
400 *
401 * %__GFP_NORETRY - If memory is not immediately available,
402 * then give up at once.
403 *
404 * %__GFP_NOWARN - If allocation fails, don't issue any warnings.
405 *
406 * %__GFP_REPEAT - If allocation fails initially, try once more before failing.
407 *
408 * There are other flags available as well, but these are not intended
409 * for general use, and so are not documented here. For a full list of
410 * potential flags, always refer to linux/gfp.h.
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000411 */
412static __always_inline void *kmalloc(size_t size, gfp_t flags)
413{
414 if (__builtin_constant_p(size)) {
415 if (size > KMALLOC_MAX_CACHE_SIZE)
416 return kmalloc_large(size, flags);
417#ifndef CONFIG_SLOB
418 if (!(flags & GFP_DMA)) {
419 int index = kmalloc_index(size);
420
421 if (!index)
422 return ZERO_SIZE_PTR;
423
424 return kmem_cache_alloc_trace(kmalloc_caches[index],
425 flags, size);
426 }
427#endif
428 }
429 return __kmalloc(size, flags);
430}
431
Christoph Lameterce6a5022013-01-10 19:14:19 +0000432/*
433 * Determine size used for the nth kmalloc cache.
434 * return size or 0 if a kmalloc cache for that
435 * size does not exist
436 */
437static __always_inline int kmalloc_size(int n)
438{
Christoph Lameter069e2b352013-06-14 19:55:13 +0000439#ifndef CONFIG_SLOB
Christoph Lameterce6a5022013-01-10 19:14:19 +0000440 if (n > 2)
441 return 1 << n;
442
443 if (n == 1 && KMALLOC_MIN_SIZE <= 32)
444 return 96;
445
446 if (n == 2 && KMALLOC_MIN_SIZE <= 64)
447 return 192;
Christoph Lameter069e2b352013-06-14 19:55:13 +0000448#endif
Christoph Lameterce6a5022013-01-10 19:14:19 +0000449 return 0;
450}
Christoph Lameterce6a5022013-01-10 19:14:19 +0000451
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000452static __always_inline void *kmalloc_node(size_t size, gfp_t flags, int node)
453{
454#ifndef CONFIG_SLOB
455 if (__builtin_constant_p(size) &&
Christoph Lameter23774a22013-09-04 19:58:08 +0000456 size <= KMALLOC_MAX_CACHE_SIZE && !(flags & GFP_DMA)) {
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000457 int i = kmalloc_index(size);
458
459 if (!i)
460 return ZERO_SIZE_PTR;
461
462 return kmem_cache_alloc_node_trace(kmalloc_caches[i],
463 flags, node, size);
464 }
465#endif
466 return __kmalloc_node(size, flags, node);
467}
468
Christoph Lameterce6a5022013-01-10 19:14:19 +0000469/*
Christoph Lameter90810642011-06-23 09:36:12 -0500470 * Setting ARCH_SLAB_MINALIGN in arch headers allows a different alignment.
471 * Intended for arches that get misalignment faults even for 64 bit integer
472 * aligned buffers.
473 */
Christoph Lameter3192b922011-06-14 16:16:36 -0500474#ifndef ARCH_SLAB_MINALIGN
475#define ARCH_SLAB_MINALIGN __alignof__(unsigned long long)
476#endif
Christoph Lameter0aa817f2007-05-16 22:11:01 -0700477/*
Glauber Costaba6c4962012-12-18 14:22:27 -0800478 * This is the main placeholder for memcg-related information in kmem caches.
479 * struct kmem_cache will hold a pointer to it, so the memory cost while
480 * disabled is 1 pointer. The runtime cost while enabled, gets bigger than it
481 * would otherwise be if that would be bundled in kmem_cache: we'll need an
482 * extra pointer chase. But the trade off clearly lays in favor of not
483 * penalizing non-users.
484 *
485 * Both the root cache and the child caches will have it. For the root cache,
486 * this will hold a dynamically allocated array large enough to hold
Vladimir Davydovf8570262014-01-23 15:53:06 -0800487 * information about the currently limited memcgs in the system. To allow the
488 * array to be accessed without taking any locks, on relocation we free the old
489 * version only after a grace period.
Glauber Costaba6c4962012-12-18 14:22:27 -0800490 *
491 * Child caches will hold extra metadata needed for its operation. Fields are:
492 *
493 * @memcg: pointer to the memcg this cache belongs to
Glauber Costa2633d7a2012-12-18 14:22:34 -0800494 * @list: list_head for the list of all caches in this memcg
495 * @root_cache: pointer to the global, root cache, this cache was derived from
Glauber Costaba6c4962012-12-18 14:22:27 -0800496 */
497struct memcg_cache_params {
498 bool is_root_cache;
499 union {
Vladimir Davydovf8570262014-01-23 15:53:06 -0800500 struct {
501 struct rcu_head rcu_head;
502 struct kmem_cache *memcg_caches[0];
503 };
Glauber Costa2633d7a2012-12-18 14:22:34 -0800504 struct {
505 struct mem_cgroup *memcg;
506 struct list_head list;
507 struct kmem_cache *root_cache;
508 };
Glauber Costaba6c4962012-12-18 14:22:27 -0800509 };
510};
511
Glauber Costa2633d7a2012-12-18 14:22:34 -0800512int memcg_update_all_caches(int num_memcgs);
513
Christoph Lameter2e892f42006-12-13 00:34:23 -0800514/**
Michael Opdenackere7efa612013-06-25 18:16:55 +0200515 * kmalloc_array - allocate memory for an array.
516 * @n: number of elements.
517 * @size: element size.
518 * @flags: the type of memory to allocate (see kmalloc).
Paul Drynoff800590f2006-06-23 02:03:48 -0700519 */
Xi Wanga8203722012-03-05 15:14:41 -0800520static inline void *kmalloc_array(size_t n, size_t size, gfp_t flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700521{
Xi Wanga3860c12012-05-31 16:26:04 -0700522 if (size != 0 && n > SIZE_MAX / size)
Paul Mundt6193a2f2007-07-15 23:38:22 -0700523 return NULL;
Xi Wanga8203722012-03-05 15:14:41 -0800524 return __kmalloc(n * size, flags);
525}
526
527/**
528 * kcalloc - allocate memory for an array. The memory is set to zero.
529 * @n: number of elements.
530 * @size: element size.
531 * @flags: the type of memory to allocate (see kmalloc).
532 */
533static inline void *kcalloc(size_t n, size_t size, gfp_t flags)
534{
535 return kmalloc_array(n, size, flags | __GFP_ZERO);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700536}
537
Christoph Hellwig1d2c8ee2006-10-04 02:15:25 -0700538/*
539 * kmalloc_track_caller is a special version of kmalloc that records the
540 * calling function of the routine calling it for slab leak tracking instead
541 * of just the calling function (confusing, eh?).
542 * It's useful when the call to kmalloc comes from a widely-used standard
543 * allocator where we care about the real place the memory allocation
544 * request comes from.
545 */
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300546extern void *__kmalloc_track_caller(size_t, gfp_t, unsigned long);
Christoph Hellwig1d2c8ee2006-10-04 02:15:25 -0700547#define kmalloc_track_caller(size, flags) \
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300548 __kmalloc_track_caller(size, flags, _RET_IP_)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700549
Manfred Spraul97e2bde2005-05-01 08:58:38 -0700550#ifdef CONFIG_NUMA
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300551extern void *__kmalloc_node_track_caller(size_t, gfp_t, int, unsigned long);
Christoph Hellwig8b98c162006-12-06 20:32:30 -0800552#define kmalloc_node_track_caller(size, flags, node) \
553 __kmalloc_node_track_caller(size, flags, node, \
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300554 _RET_IP_)
Christoph Lameter2e892f42006-12-13 00:34:23 -0800555
Christoph Hellwig8b98c162006-12-06 20:32:30 -0800556#else /* CONFIG_NUMA */
Christoph Lameter2e892f42006-12-13 00:34:23 -0800557
558#define kmalloc_node_track_caller(size, flags, node) \
559 kmalloc_track_caller(size, flags)
560
Pascal Terjandfcd3612008-11-25 15:08:19 +0100561#endif /* CONFIG_NUMA */
Christoph Hellwig8b98c162006-12-06 20:32:30 -0800562
Christoph Lameter81cda662007-07-17 04:03:29 -0700563/*
564 * Shortcuts
565 */
566static inline void *kmem_cache_zalloc(struct kmem_cache *k, gfp_t flags)
567{
568 return kmem_cache_alloc(k, flags | __GFP_ZERO);
569}
570
571/**
572 * kzalloc - allocate memory. The memory is set to zero.
573 * @size: how many bytes of memory are required.
574 * @flags: the type of memory to allocate (see kmalloc).
575 */
576static inline void *kzalloc(size_t size, gfp_t flags)
577{
578 return kmalloc(size, flags | __GFP_ZERO);
579}
580
Jeff Layton979b0fe2008-06-05 22:47:00 -0700581/**
582 * kzalloc_node - allocate zeroed memory from a particular memory node.
583 * @size: how many bytes of memory are required.
584 * @flags: the type of memory to allocate (see kmalloc).
585 * @node: memory node from which to allocate
586 */
587static inline void *kzalloc_node(size_t size, gfp_t flags, int node)
588{
589 return kmalloc_node(size, flags | __GFP_ZERO, node);
590}
591
Joonsoo Kim07f361b2014-10-09 15:26:00 -0700592unsigned int kmem_cache_size(struct kmem_cache *s);
Pekka Enberg7e85ee02009-06-12 14:03:06 +0300593void __init kmem_cache_init_late(void);
594
Linus Torvalds1da177e2005-04-16 15:20:36 -0700595#endif /* _LINUX_SLAB_H */