slab.c 112 KB
Newer Older
Linus Torvalds's avatar
Linus Torvalds committed
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
/*
 * linux/mm/slab.c
 * Written by Mark Hemment, 1996/97.
 * (markhe@nextd.demon.co.uk)
 *
 * kmem_cache_destroy() + some cleanup - 1999 Andrea Arcangeli
 *
 * Major cleanup, different bufctl logic, per-cpu arrays
 *	(c) 2000 Manfred Spraul
 *
 * Cleanup, make the head arrays unconditional, preparation for NUMA
 * 	(c) 2002 Manfred Spraul
 *
 * An implementation of the Slab Allocator as described in outline in;
 *	UNIX Internals: The New Frontiers by Uresh Vahalia
 *	Pub: Prentice Hall	ISBN 0-13-101908-2
 * or with a little more detail in;
 *	The Slab Allocator: An Object-Caching Kernel Memory Allocator
 *	Jeff Bonwick (Sun Microsystems).
 *	Presented at: USENIX Summer 1994 Technical Conference
 *
 * The memory is organized in caches, one cache for each object type.
 * (e.g. inode_cache, dentry_cache, buffer_head, vm_area_struct)
 * Each cache consists out of many slabs (they are small (usually one
 * page long) and always contiguous), and each slab contains multiple
 * initialized objects.
 *
 * This means, that your constructor is used only for newly allocated
Simon Arlott's avatar
Simon Arlott committed
29
 * slabs and you must pass objects with the same initializations to
Linus Torvalds's avatar
Linus Torvalds committed
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
 * kmem_cache_free.
 *
 * Each cache can only support one memory type (GFP_DMA, GFP_HIGHMEM,
 * normal). If you need a special memory type, then must create a new
 * cache for that memory type.
 *
 * In order to reduce fragmentation, the slabs are sorted in 3 groups:
 *   full slabs with 0 free objects
 *   partial slabs
 *   empty slabs with no allocated objects
 *
 * If partial slabs exist, then new allocations come from these slabs,
 * otherwise from empty slabs or new slabs are allocated.
 *
 * kmem_cache_destroy() CAN CRASH if you try to allocate from the cache
 * during kmem_cache_destroy(). The caller must prevent concurrent allocs.
 *
 * Each cache has a short per-cpu head array, most allocs
 * and frees go into that array, and if that array overflows, then 1/2
 * of the entries in the array are given back into the global cache.
 * The head array is strictly LIFO and should improve the cache hit rates.
 * On SMP, it additionally reduces the spinlock operations.
 *
Andrew Morton's avatar
Andrew Morton committed
53
 * The c_cpuarray may not be read with enabled local interrupts -
Linus Torvalds's avatar
Linus Torvalds committed
54
55
56
57
 * it's changed with a smp_call_function().
 *
 * SMP synchronization:
 *  constructors and destructors are called without any locking.
58
 *  Several members in struct kmem_cache and struct slab never change, they
Linus Torvalds's avatar
Linus Torvalds committed
59
60
61
62
63
64
65
66
67
68
69
70
 *	are accessed without any locking.
 *  The per-cpu arrays are never accessed from the wrong cpu, no locking,
 *  	and local interrupts are disabled so slab code is preempt-safe.
 *  The non-constant members are protected with a per-cache irq spinlock.
 *
 * Many thanks to Mark Hemment, who wrote another per-cpu slab patch
 * in 2000 - many ideas in the current implementation are derived from
 * his patch.
 *
 * Further notes from the original documentation:
 *
 * 11 April '97.  Started multi-threading - markhe
71
 *	The global cache-chain is protected by the mutex 'slab_mutex'.
Linus Torvalds's avatar
Linus Torvalds committed
72
73
74
75
76
77
 *	The sem is only needed when accessing/extending the cache-chain, which
 *	can never happen inside an interrupt (kmem_cache_create(),
 *	kmem_cache_shrink() and kmem_cache_reap()).
 *
 *	At present, each engine can be growing a cache.  This should be blocked.
 *
78
79
80
81
82
83
84
85
86
 * 15 March 2005. NUMA slab allocator.
 *	Shai Fultheim <shai@scalex86.org>.
 *	Shobhit Dayal <shobhit@calsoftinc.com>
 *	Alok N Kataria <alokk@calsoftinc.com>
 *	Christoph Lameter <christoph@lameter.com>
 *
 *	Modified the slab allocator to be node aware on NUMA systems.
 *	Each node has its own list of partial, free and full slabs.
 *	All object allocations for a node occur from node specific slab lists.
Linus Torvalds's avatar
Linus Torvalds committed
87
88
89
90
 */

#include	<linux/slab.h>
#include	<linux/mm.h>
91
#include	<linux/poison.h>
Linus Torvalds's avatar
Linus Torvalds committed
92
93
94
95
96
#include	<linux/swap.h>
#include	<linux/cache.h>
#include	<linux/interrupt.h>
#include	<linux/init.h>
#include	<linux/compiler.h>
97
#include	<linux/cpuset.h>
98
#include	<linux/proc_fs.h>
Linus Torvalds's avatar
Linus Torvalds committed
99
100
101
102
103
104
105
#include	<linux/seq_file.h>
#include	<linux/notifier.h>
#include	<linux/kallsyms.h>
#include	<linux/cpu.h>
#include	<linux/sysctl.h>
#include	<linux/module.h>
#include	<linux/rcupdate.h>
106
#include	<linux/string.h>
107
#include	<linux/uaccess.h>
108
#include	<linux/nodemask.h>
109
#include	<linux/kmemleak.h>
110
#include	<linux/mempolicy.h>
Ingo Molnar's avatar
Ingo Molnar committed
111
#include	<linux/mutex.h>
112
#include	<linux/fault-inject.h>
Ingo Molnar's avatar
Ingo Molnar committed
113
#include	<linux/rtmutex.h>
114
#include	<linux/reciprocal_div.h>
115
#include	<linux/debugobjects.h>
Pekka Enberg's avatar
Pekka Enberg committed
116
#include	<linux/kmemcheck.h>
117
#include	<linux/memory.h>
118
#include	<linux/prefetch.h>
Linus Torvalds's avatar
Linus Torvalds committed
119

120
121
#include	<net/sock.h>

Linus Torvalds's avatar
Linus Torvalds committed
122
123
124
125
#include	<asm/cacheflush.h>
#include	<asm/tlbflush.h>
#include	<asm/page.h>

126
127
#include <trace/events/kmem.h>

128
129
#include	"internal.h"

130
131
#include	"slab.h"

Linus Torvalds's avatar
Linus Torvalds committed
132
/*
133
 * DEBUG	- 1 for kmem_cache_create() to honour; SLAB_RED_ZONE & SLAB_POISON.
Linus Torvalds's avatar
Linus Torvalds committed
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
 *		  0 for faster, smaller code (especially in the critical paths).
 *
 * STATS	- 1 to collect stats for /proc/slabinfo.
 *		  0 for faster, smaller code (especially in the critical paths).
 *
 * FORCED_DEBUG	- 1 enables SLAB_RED_ZONE and SLAB_POISON (if possible)
 */

#ifdef CONFIG_DEBUG_SLAB
#define	DEBUG		1
#define	STATS		1
#define	FORCED_DEBUG	1
#else
#define	DEBUG		0
#define	STATS		0
#define	FORCED_DEBUG	0
#endif

/* Shouldn't this be in a header file somewhere? */
#define	BYTES_PER_WORD		sizeof(void *)
David Woodhouse's avatar
David Woodhouse committed
154
#define	REDZONE_ALIGN		max(BYTES_PER_WORD, __alignof__(unsigned long long))
Linus Torvalds's avatar
Linus Torvalds committed
155
156
157
158
159

#ifndef ARCH_KMALLOC_FLAGS
#define ARCH_KMALLOC_FLAGS SLAB_HWCACHE_ALIGN
#endif

160
161
162
163
164
165
166
167
168
#define FREELIST_BYTE_INDEX (((PAGE_SIZE >> BITS_PER_BYTE) \
				<= SLAB_OBJ_MIN_SIZE) ? 1 : 0)

#if FREELIST_BYTE_INDEX
typedef unsigned char freelist_idx_t;
#else
typedef unsigned short freelist_idx_t;
#endif

169
#define SLAB_OBJ_MAX_NUM ((1 << sizeof(freelist_idx_t) * BITS_PER_BYTE) - 1)
170

171
172
173
174
175
176
/*
 * true if a page was allocated from pfmemalloc reserves for network-based
 * swap
 */
static bool pfmemalloc_active __read_mostly;

Linus Torvalds's avatar
Linus Torvalds committed
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
/*
 * struct array_cache
 *
 * Purpose:
 * - LIFO ordering, to hand out cache-warm objects from _alloc
 * - reduce the number of linked list operations
 * - reduce spinlock operations
 *
 * The limit is stored in the per-cpu structure to reduce the data cache
 * footprint.
 *
 */
struct array_cache {
	unsigned int avail;
	unsigned int limit;
	unsigned int batchcount;
	unsigned int touched;
194
	spinlock_t lock;
195
	void *entry[];	/*
Andrew Morton's avatar
Andrew Morton committed
196
197
198
			 * Must have this definition in here for the proper
			 * alignment of array_cache. Also simplifies accessing
			 * the entries.
199
200
201
202
			 *
			 * Entries should not be directly dereferenced as
			 * entries belonging to slabs marked pfmemalloc will
			 * have the lower bits set SLAB_OBJ_PFMEMALLOC
Andrew Morton's avatar
Andrew Morton committed
203
			 */
Linus Torvalds's avatar
Linus Torvalds committed
204
205
};

206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
#define SLAB_OBJ_PFMEMALLOC	1
static inline bool is_obj_pfmemalloc(void *objp)
{
	return (unsigned long)objp & SLAB_OBJ_PFMEMALLOC;
}

static inline void set_obj_pfmemalloc(void **objp)
{
	*objp = (void *)((unsigned long)*objp | SLAB_OBJ_PFMEMALLOC);
	return;
}

static inline void clear_obj_pfmemalloc(void **objp)
{
	*objp = (void *)((unsigned long)*objp & ~SLAB_OBJ_PFMEMALLOC);
}

Andrew Morton's avatar
Andrew Morton committed
223
224
225
/*
 * bootstrap: The caches do not work without cpuarrays anymore, but the
 * cpuarrays are allocated from the generic caches...
Linus Torvalds's avatar
Linus Torvalds committed
226
227
228
229
 */
#define BOOT_CPUCACHE_ENTRIES	1
struct arraycache_init {
	struct array_cache cache;
230
	void *entries[BOOT_CPUCACHE_ENTRIES];
Linus Torvalds's avatar
Linus Torvalds committed
231
232
};

233
234
235
/*
 * Need this for bootstrapping a per node allocator.
 */
236
#define NUM_INIT_LISTS (3 * MAX_NUMNODES)
237
static struct kmem_cache_node __initdata init_kmem_cache_node[NUM_INIT_LISTS];
238
#define	CACHE_CACHE 0
239
#define	SIZE_AC MAX_NUMNODES
240
#define	SIZE_NODE (2 * MAX_NUMNODES)
241

242
static int drain_freelist(struct kmem_cache *cache,
243
			struct kmem_cache_node *n, int tofree);
244
245
static void free_block(struct kmem_cache *cachep, void **objpp, int len,
			int node);
246
static int enable_cpucache(struct kmem_cache *cachep, gfp_t gfp);
247
static void cache_reap(struct work_struct *unused);
248

249
250
static int slab_early_init = 1;

251
#define INDEX_AC kmalloc_index(sizeof(struct arraycache_init))
252
#define INDEX_NODE kmalloc_index(sizeof(struct kmem_cache_node))
Linus Torvalds's avatar
Linus Torvalds committed
253

254
static void kmem_cache_node_init(struct kmem_cache_node *parent)
255
256
257
258
259
260
{
	INIT_LIST_HEAD(&parent->slabs_full);
	INIT_LIST_HEAD(&parent->slabs_partial);
	INIT_LIST_HEAD(&parent->slabs_free);
	parent->shared = NULL;
	parent->alien = NULL;
261
	parent->colour_next = 0;
262
263
264
265
266
	spin_lock_init(&parent->list_lock);
	parent->free_objects = 0;
	parent->free_touched = 0;
}

Andrew Morton's avatar
Andrew Morton committed
267
268
269
#define MAKE_LIST(cachep, listp, slab, nodeid)				\
	do {								\
		INIT_LIST_HEAD(listp);					\
270
		list_splice(&(cachep->node[nodeid]->slab), listp);	\
271
272
	} while (0)

Andrew Morton's avatar
Andrew Morton committed
273
274
#define	MAKE_ALL_LISTS(cachep, ptr, nodeid)				\
	do {								\
275
276
277
278
	MAKE_LIST((cachep), (&(ptr)->slabs_full), slabs_full, nodeid);	\
	MAKE_LIST((cachep), (&(ptr)->slabs_partial), slabs_partial, nodeid); \
	MAKE_LIST((cachep), (&(ptr)->slabs_free), slabs_free, nodeid);	\
	} while (0)
Linus Torvalds's avatar
Linus Torvalds committed
279
280
281
282
283

#define CFLGS_OFF_SLAB		(0x80000000UL)
#define	OFF_SLAB(x)	((x)->flags & CFLGS_OFF_SLAB)

#define BATCHREFILL_LIMIT	16
Andrew Morton's avatar
Andrew Morton committed
284
285
286
/*
 * Optimization question: fewer reaps means less probability for unnessary
 * cpucache drain/refill cycles.
Linus Torvalds's avatar
Linus Torvalds committed
287
 *
Adrian Bunk's avatar
Adrian Bunk committed
288
 * OTOH the cpuarrays can contain lots of objects,
Linus Torvalds's avatar
Linus Torvalds committed
289
290
 * which could lock up otherwise freeable slabs.
 */
291
292
#define REAPTIMEOUT_AC		(2*HZ)
#define REAPTIMEOUT_NODE	(4*HZ)
Linus Torvalds's avatar
Linus Torvalds committed
293
294
295
296
297
298

#if STATS
#define	STATS_INC_ACTIVE(x)	((x)->num_active++)
#define	STATS_DEC_ACTIVE(x)	((x)->num_active--)
#define	STATS_INC_ALLOCED(x)	((x)->num_allocations++)
#define	STATS_INC_GROWN(x)	((x)->grown++)
299
#define	STATS_ADD_REAPED(x,y)	((x)->reaped += (y))
Andrew Morton's avatar
Andrew Morton committed
300
301
302
303
304
#define	STATS_SET_HIGH(x)						\
	do {								\
		if ((x)->num_active > (x)->high_mark)			\
			(x)->high_mark = (x)->num_active;		\
	} while (0)
Linus Torvalds's avatar
Linus Torvalds committed
305
306
#define	STATS_INC_ERR(x)	((x)->errors++)
#define	STATS_INC_NODEALLOCS(x)	((x)->node_allocs++)
307
#define	STATS_INC_NODEFREES(x)	((x)->node_frees++)
308
#define STATS_INC_ACOVERFLOW(x)   ((x)->node_overflow++)
Andrew Morton's avatar
Andrew Morton committed
309
310
311
312
313
#define	STATS_SET_FREEABLE(x, i)					\
	do {								\
		if ((x)->max_freeable < i)				\
			(x)->max_freeable = i;				\
	} while (0)
Linus Torvalds's avatar
Linus Torvalds committed
314
315
316
317
318
319
320
321
322
#define STATS_INC_ALLOCHIT(x)	atomic_inc(&(x)->allochit)
#define STATS_INC_ALLOCMISS(x)	atomic_inc(&(x)->allocmiss)
#define STATS_INC_FREEHIT(x)	atomic_inc(&(x)->freehit)
#define STATS_INC_FREEMISS(x)	atomic_inc(&(x)->freemiss)
#else
#define	STATS_INC_ACTIVE(x)	do { } while (0)
#define	STATS_DEC_ACTIVE(x)	do { } while (0)
#define	STATS_INC_ALLOCED(x)	do { } while (0)
#define	STATS_INC_GROWN(x)	do { } while (0)
323
#define	STATS_ADD_REAPED(x,y)	do { (void)(y); } while (0)
Linus Torvalds's avatar
Linus Torvalds committed
324
325
326
#define	STATS_SET_HIGH(x)	do { } while (0)
#define	STATS_INC_ERR(x)	do { } while (0)
#define	STATS_INC_NODEALLOCS(x)	do { } while (0)
327
#define	STATS_INC_NODEFREES(x)	do { } while (0)
328
#define STATS_INC_ACOVERFLOW(x)   do { } while (0)
Andrew Morton's avatar
Andrew Morton committed
329
#define	STATS_SET_FREEABLE(x, i) do { } while (0)
Linus Torvalds's avatar
Linus Torvalds committed
330
331
332
333
334
335
336
337
#define STATS_INC_ALLOCHIT(x)	do { } while (0)
#define STATS_INC_ALLOCMISS(x)	do { } while (0)
#define STATS_INC_FREEHIT(x)	do { } while (0)
#define STATS_INC_FREEMISS(x)	do { } while (0)
#endif

#if DEBUG

Andrew Morton's avatar
Andrew Morton committed
338
339
/*
 * memory layout of objects:
Linus Torvalds's avatar
Linus Torvalds committed
340
 * 0		: objp
341
 * 0 .. cachep->obj_offset - BYTES_PER_WORD - 1: padding. This ensures that
Linus Torvalds's avatar
Linus Torvalds committed
342
343
 * 		the end of an object is aligned with the end of the real
 * 		allocation. Catches writes behind the end of the allocation.
344
 * cachep->obj_offset - BYTES_PER_WORD .. cachep->obj_offset - 1:
Linus Torvalds's avatar
Linus Torvalds committed
345
 * 		redzone word.
346
 * cachep->obj_offset: The real object.
347
348
 * cachep->size - 2* BYTES_PER_WORD: redzone word [BYTES_PER_WORD long]
 * cachep->size - 1* BYTES_PER_WORD: last caller address
Andrew Morton's avatar
Andrew Morton committed
349
 *					[BYTES_PER_WORD long]
Linus Torvalds's avatar
Linus Torvalds committed
350
 */
351
static int obj_offset(struct kmem_cache *cachep)
Linus Torvalds's avatar
Linus Torvalds committed
352
{
353
	return cachep->obj_offset;
Linus Torvalds's avatar
Linus Torvalds committed
354
355
}

356
static unsigned long long *dbg_redzone1(struct kmem_cache *cachep, void *objp)
Linus Torvalds's avatar
Linus Torvalds committed
357
358
{
	BUG_ON(!(cachep->flags & SLAB_RED_ZONE));
359
360
	return (unsigned long long*) (objp + obj_offset(cachep) -
				      sizeof(unsigned long long));
Linus Torvalds's avatar
Linus Torvalds committed
361
362
}

363
static unsigned long long *dbg_redzone2(struct kmem_cache *cachep, void *objp)
Linus Torvalds's avatar
Linus Torvalds committed
364
365
366
{
	BUG_ON(!(cachep->flags & SLAB_RED_ZONE));
	if (cachep->flags & SLAB_STORE_USER)
367
		return (unsigned long long *)(objp + cachep->size -
368
					      sizeof(unsigned long long) -
David Woodhouse's avatar
David Woodhouse committed
369
					      REDZONE_ALIGN);
370
	return (unsigned long long *) (objp + cachep->size -
371
				       sizeof(unsigned long long));
Linus Torvalds's avatar
Linus Torvalds committed
372
373
}

374
static void **dbg_userword(struct kmem_cache *cachep, void *objp)
Linus Torvalds's avatar
Linus Torvalds committed
375
376
{
	BUG_ON(!(cachep->flags & SLAB_STORE_USER));
377
	return (void **)(objp + cachep->size - BYTES_PER_WORD);
Linus Torvalds's avatar
Linus Torvalds committed
378
379
380
381
}

#else

382
#define obj_offset(x)			0
383
384
#define dbg_redzone1(cachep, objp)	({BUG(); (unsigned long long *)NULL;})
#define dbg_redzone2(cachep, objp)	({BUG(); (unsigned long long *)NULL;})
Linus Torvalds's avatar
Linus Torvalds committed
385
386
387
388
#define dbg_userword(cachep, objp)	({BUG(); (void **)NULL;})

#endif

389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
#define OBJECT_FREE (0)
#define OBJECT_ACTIVE (1)

#ifdef CONFIG_DEBUG_SLAB_LEAK

static void set_obj_status(struct page *page, int idx, int val)
{
	int freelist_size;
	char *status;
	struct kmem_cache *cachep = page->slab_cache;

	freelist_size = cachep->num * sizeof(freelist_idx_t);
	status = (char *)page->freelist + freelist_size;
	status[idx] = val;
}

static inline unsigned int get_obj_status(struct page *page, int idx)
{
	int freelist_size;
	char *status;
	struct kmem_cache *cachep = page->slab_cache;

	freelist_size = cachep->num * sizeof(freelist_idx_t);
	status = (char *)page->freelist + freelist_size;

	return status[idx];
}

#else
static inline void set_obj_status(struct page *page, int idx, int val) {}

#endif

Linus Torvalds's avatar
Linus Torvalds committed
422
/*
423
424
 * Do not go above this order unless 0 objects fit into the slab or
 * overridden on the command line.
Linus Torvalds's avatar
Linus Torvalds committed
425
 */
426
427
428
#define	SLAB_MAX_ORDER_HI	1
#define	SLAB_MAX_ORDER_LO	0
static int slab_max_order = SLAB_MAX_ORDER_LO;
429
static bool slab_max_order_set __initdata;
Linus Torvalds's avatar
Linus Torvalds committed
430

431
432
static inline struct kmem_cache *virt_to_cache(const void *obj)
{
433
	struct page *page = virt_to_head_page(obj);
434
	return page->slab_cache;
435
436
}

437
static inline void *index_to_obj(struct kmem_cache *cache, struct page *page,
438
439
				 unsigned int idx)
{
440
	return page->s_mem + cache->size * idx;
441
442
}

443
/*
444
445
446
 * We want to avoid an expensive divide : (offset / cache->size)
 *   Using the fact that size is a constant for a particular cache,
 *   we can replace (offset / cache->size) by
447
448
449
 *   reciprocal_divide(offset, cache->reciprocal_buffer_size)
 */
static inline unsigned int obj_to_index(const struct kmem_cache *cache,
450
					const struct page *page, void *obj)
451
{
452
	u32 offset = (obj - page->s_mem);
453
	return reciprocal_divide(offset, cache->reciprocal_buffer_size);
454
455
}

Linus Torvalds's avatar
Linus Torvalds committed
456
static struct arraycache_init initarray_generic =
457
    { {0, BOOT_CPUCACHE_ENTRIES, 1, 0} };
Linus Torvalds's avatar
Linus Torvalds committed
458
459

/* internal cache of cache description objs */
460
static struct kmem_cache kmem_cache_boot = {
461
462
463
	.batchcount = 1,
	.limit = BOOT_CPUCACHE_ENTRIES,
	.shared = 1,
464
	.size = sizeof(struct kmem_cache),
465
	.name = "kmem_cache",
Linus Torvalds's avatar
Linus Torvalds committed
466
467
};

468
469
#define BAD_ALIEN_MAGIC 0x01020304ul

470
471
472
473
474
475
476
477
#ifdef CONFIG_LOCKDEP

/*
 * Slab sometimes uses the kmalloc slabs to store the slab headers
 * for other slabs "off slab".
 * The locking for this is tricky in that it nests within the locks
 * of all other slabs in a few places; to deal with this special
 * locking we put on-slab caches into a separate lock-class.
478
479
480
481
 *
 * We set lock class for alien array caches which are up during init.
 * The lock annotation will be lost if all cpus of a node goes down and
 * then comes back up during hotplug
482
 */
483
484
485
static struct lock_class_key on_slab_l3_key;
static struct lock_class_key on_slab_alc_key;

486
487
488
489
490
491
492
493
static struct lock_class_key debugobj_l3_key;
static struct lock_class_key debugobj_alc_key;

static void slab_set_lock_classes(struct kmem_cache *cachep,
		struct lock_class_key *l3_key, struct lock_class_key *alc_key,
		int q)
{
	struct array_cache **alc;
494
	struct kmem_cache_node *n;
495
496
	int r;

497
498
	n = cachep->node[q];
	if (!n)
499
500
		return;

501
502
	lockdep_set_class(&n->list_lock, l3_key);
	alc = n->alien;
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
	/*
	 * FIXME: This check for BAD_ALIEN_MAGIC
	 * should go away when common slab code is taught to
	 * work even without alien caches.
	 * Currently, non NUMA code returns BAD_ALIEN_MAGIC
	 * for alloc_alien_cache,
	 */
	if (!alc || (unsigned long)alc == BAD_ALIEN_MAGIC)
		return;
	for_each_node(r) {
		if (alc[r])
			lockdep_set_class(&alc[r]->lock, alc_key);
	}
}

static void slab_set_debugobj_lock_classes_node(struct kmem_cache *cachep, int node)
{
	slab_set_lock_classes(cachep, &debugobj_l3_key, &debugobj_alc_key, node);
}

static void slab_set_debugobj_lock_classes(struct kmem_cache *cachep)
{
	int node;

	for_each_online_node(node)
		slab_set_debugobj_lock_classes_node(cachep, node);
}

531
static void init_node_lock_keys(int q)
532
{
533
	int i;
534

535
	if (slab_state < UP)
536
537
		return;

Christoph Lameter's avatar
Christoph Lameter committed
538
	for (i = 1; i <= KMALLOC_SHIFT_HIGH; i++) {
539
		struct kmem_cache_node *n;
540
541
542
543
		struct kmem_cache *cache = kmalloc_caches[i];

		if (!cache)
			continue;
544

545
546
		n = cache->node[q];
		if (!n || OFF_SLAB(cache))
547
			continue;
548

549
		slab_set_lock_classes(cache, &on_slab_l3_key,
550
				&on_slab_alc_key, q);
551
552
	}
}
553

554
555
static void on_slab_lock_classes_node(struct kmem_cache *cachep, int q)
{
556
	if (!cachep->node[q])
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
		return;

	slab_set_lock_classes(cachep, &on_slab_l3_key,
			&on_slab_alc_key, q);
}

static inline void on_slab_lock_classes(struct kmem_cache *cachep)
{
	int node;

	VM_BUG_ON(OFF_SLAB(cachep));
	for_each_node(node)
		on_slab_lock_classes_node(cachep, node);
}

572
static inline void __init init_lock_keys(void)
573
574
575
576
577
578
{
	int node;

	for_each_node(node)
		init_node_lock_keys(node);
}
579
#else
580
static void __init init_node_lock_keys(int q)
581
582
583
{
}

584
static inline void init_lock_keys(void)
585
586
{
}
587

588
589
590
591
592
593
594
595
static inline void on_slab_lock_classes(struct kmem_cache *cachep)
{
}

static inline void on_slab_lock_classes_node(struct kmem_cache *cachep, int node)
{
}

596
597
598
599
600
601
602
static void slab_set_debugobj_lock_classes_node(struct kmem_cache *cachep, int node)
{
}

static void slab_set_debugobj_lock_classes(struct kmem_cache *cachep)
{
}
603
604
#endif

605
static DEFINE_PER_CPU(struct delayed_work, slab_reap_work);
Linus Torvalds's avatar
Linus Torvalds committed
606

607
static inline struct array_cache *cpu_cache_get(struct kmem_cache *cachep)
Linus Torvalds's avatar
Linus Torvalds committed
608
609
610
611
{
	return cachep->array[smp_processor_id()];
}

612
613
614
615
616
617
618
619
620
621
622
623
624
625
static size_t calculate_freelist_size(int nr_objs, size_t align)
{
	size_t freelist_size;

	freelist_size = nr_objs * sizeof(freelist_idx_t);
	if (IS_ENABLED(CONFIG_DEBUG_SLAB_LEAK))
		freelist_size += nr_objs * sizeof(char);

	if (align)
		freelist_size = ALIGN(freelist_size, align);

	return freelist_size;
}

626
627
static int calculate_nr_objs(size_t slab_size, size_t buffer_size,
				size_t idx_size, size_t align)
Linus Torvalds's avatar
Linus Torvalds committed
628
{
629
	int nr_objs;
630
	size_t remained_size;
631
	size_t freelist_size;
632
	int extra_space = 0;
633

634
635
	if (IS_ENABLED(CONFIG_DEBUG_SLAB_LEAK))
		extra_space = sizeof(char);
636
637
638
639
640
641
642
643
	/*
	 * Ignore padding for the initial guess. The padding
	 * is at most @align-1 bytes, and @buffer_size is at
	 * least @align. In the worst case, this result will
	 * be one greater than the number of objects that fit
	 * into the memory allocation when taking the padding
	 * into account.
	 */
644
	nr_objs = slab_size / (buffer_size + idx_size + extra_space);
645
646
647
648
649

	/*
	 * This calculated number will be either the right
	 * amount, or one greater than what we want.
	 */
650
651
652
	remained_size = slab_size - nr_objs * buffer_size;
	freelist_size = calculate_freelist_size(nr_objs, align);
	if (remained_size < freelist_size)
653
654
655
		nr_objs--;

	return nr_objs;
656
}
Linus Torvalds's avatar
Linus Torvalds committed
657

Andrew Morton's avatar
Andrew Morton committed
658
659
660
/*
 * Calculate the number of objects and left-over bytes for a given buffer size.
 */
661
662
663
664
665
666
667
static void cache_estimate(unsigned long gfporder, size_t buffer_size,
			   size_t align, int flags, size_t *left_over,
			   unsigned int *num)
{
	int nr_objs;
	size_t mgmt_size;
	size_t slab_size = PAGE_SIZE << gfporder;
Linus Torvalds's avatar
Linus Torvalds committed
668

669
670
671
672
673
	/*
	 * The slab management structure can be either off the slab or
	 * on it. For the latter case, the memory allocated for a
	 * slab is used for:
	 *
Joonsoo Kim's avatar
Joonsoo Kim committed
674
	 * - One unsigned int for each object
675
676
677
678
679
680
681
682
683
684
685
686
687
	 * - Padding to respect alignment of @align
	 * - @buffer_size bytes for each object
	 *
	 * If the slab management structure is off the slab, then the
	 * alignment will already be calculated into the size. Because
	 * the slabs are all pages aligned, the objects will be at the
	 * correct alignment when allocated.
	 */
	if (flags & CFLGS_OFF_SLAB) {
		mgmt_size = 0;
		nr_objs = slab_size / buffer_size;

	} else {
688
		nr_objs = calculate_nr_objs(slab_size, buffer_size,
689
					sizeof(freelist_idx_t), align);
690
		mgmt_size = calculate_freelist_size(nr_objs, align);
691
692
693
	}
	*num = nr_objs;
	*left_over = slab_size - nr_objs*buffer_size - mgmt_size;
Linus Torvalds's avatar
Linus Torvalds committed
694
695
}

696
#if DEBUG
697
#define slab_error(cachep, msg) __slab_error(__func__, cachep, msg)
Linus Torvalds's avatar
Linus Torvalds committed
698

Andrew Morton's avatar
Andrew Morton committed
699
700
static void __slab_error(const char *function, struct kmem_cache *cachep,
			char *msg)
Linus Torvalds's avatar
Linus Torvalds committed
701
702
{
	printk(KERN_ERR "slab error in %s(): cache `%s': %s\n",
703
	       function, cachep->name, msg);
Linus Torvalds's avatar
Linus Torvalds committed
704
	dump_stack();
705
	add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
Linus Torvalds's avatar
Linus Torvalds committed
706
}
707
#endif
Linus Torvalds's avatar
Linus Torvalds committed
708

709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
/*
 * By default on NUMA we use alien caches to stage the freeing of
 * objects allocated from other nodes. This causes massive memory
 * inefficiencies when using fake NUMA setup to split memory into a
 * large number of small nodes, so it can be disabled on the command
 * line
  */

static int use_alien_caches __read_mostly = 1;
static int __init noaliencache_setup(char *s)
{
	use_alien_caches = 0;
	return 1;
}
__setup("noaliencache", noaliencache_setup);

725
726
727
728
729
730
731
732
733
734
735
static int __init slab_max_order_setup(char *str)
{
	get_option(&str, &slab_max_order);
	slab_max_order = slab_max_order < 0 ? 0 :
				min(slab_max_order, MAX_ORDER - 1);
	slab_max_order_set = true;

	return 1;
}
__setup("slab_max_order=", slab_max_order_setup);

736
737
738
739
740
741
742
#ifdef CONFIG_NUMA
/*
 * Special reaping functions for NUMA systems called from cache_reap().
 * These take care of doing round robin flushing of alien caches (containing
 * objects freed on different nodes from which they were allocated) and the
 * flushing of remote pcps by calling drain_node_pages.
 */
743
static DEFINE_PER_CPU(unsigned long, slab_reap_node);
744
745
746
747
748

static void init_reap_node(int cpu)
{
	int node;

749
	node = next_node(cpu_to_mem(cpu), node_online_map);
750
	if (node == MAX_NUMNODES)
751
		node = first_node(node_online_map);
752

753
	per_cpu(slab_reap_node, cpu) = node;
754
755
756
757
}

static void next_reap_node(void)
{
758
	int node = __this_cpu_read(slab_reap_node);
759
760
761
762

	node = next_node(node, node_online_map);
	if (unlikely(node >= MAX_NUMNODES))
		node = first_node(node_online_map);
763
	__this_cpu_write(slab_reap_node, node);
764
765
766
767
768
769
770
}

#else
#define init_reap_node(cpu) do { } while (0)
#define next_reap_node(void) do { } while (0)
#endif

Linus Torvalds's avatar
Linus Torvalds committed
771
772
773
774
775
776
777
/*
 * Initiate the reap timer running on the target CPU.  We run at around 1 to 2Hz
 * via the workqueue/eventd.
 * Add the CPU number into the expiration time to minimize the possibility of
 * the CPUs getting into lockstep and contending for the global cache chain
 * lock.
 */
778
static void start_cpu_timer(int cpu)
Linus Torvalds's avatar
Linus Torvalds committed
779
{
780
	struct delayed_work *reap_work = &per_cpu(slab_reap_work, cpu);
Linus Torvalds's avatar
Linus Torvalds committed
781
782
783
784
785
786

	/*
	 * When this gets called from do_initcalls via cpucache_init(),
	 * init_workqueues() has already run, so keventd will be setup
	 * at that time.
	 */
787
	if (keventd_up() && reap_work->work.func == NULL) {
788
		init_reap_node(cpu);
789
		INIT_DEFERRABLE_WORK(reap_work, cache_reap);
790
791
		schedule_delayed_work_on(cpu, reap_work,
					__round_jiffies_relative(HZ, cpu));
Linus Torvalds's avatar
Linus Torvalds committed
792
793
794
	}
}

795
static struct array_cache *alloc_arraycache(int node, int entries,
796
					    int batchcount, gfp_t gfp)
Linus Torvalds's avatar
Linus Torvalds committed
797
{
798
	int memsize = sizeof(void *) * entries + sizeof(struct array_cache);
Linus Torvalds's avatar
Linus Torvalds committed
799
800
	struct array_cache *nc = NULL;

801
	nc = kmalloc_node(memsize, gfp, node);
802
803
	/*
	 * The array_cache structures contain pointers to free object.
Lucas De Marchi's avatar
Lucas De Marchi committed
804
	 * However, when such objects are allocated or transferred to another
805
806
807
808
809
	 * cache the pointers are not cleared and they could be counted as
	 * valid references during a kmemleak scan. Therefore, kmemleak must
	 * not scan such objects.
	 */
	kmemleak_no_scan(nc);
Linus Torvalds's avatar
Linus Torvalds committed
810
811
812
813
814
	if (nc) {
		nc->avail = 0;
		nc->limit = entries;
		nc->batchcount = batchcount;
		nc->touched = 0;
815
		spin_lock_init(&nc->lock);
Linus Torvalds's avatar
Linus Torvalds committed
816
817
818
819
	}
	return nc;
}

820
static inline bool is_slab_pfmemalloc(struct page *page)
821
822
823
824
825
826
827
828
{
	return PageSlabPfmemalloc(page);
}

/* Clears pfmemalloc_active if no slabs have pfmalloc set */
static void recheck_pfmemalloc_active(struct kmem_cache *cachep,
						struct array_cache *ac)
{
829
	struct kmem_cache_node *n = cachep->node[numa_mem_id()];
830
	struct page *page;
831
832
833
834
835
	unsigned long flags;

	if (!pfmemalloc_active)
		return;

836
	spin_lock_irqsave(&n->list_lock, flags);
837
838
	list_for_each_entry(page, &n->slabs_full, lru)
		if (is_slab_pfmemalloc(page))
839
840
			goto out;

841
842
	list_for_each_entry(page, &n->slabs_partial, lru)
		if (is_slab_pfmemalloc(page))
843
844
			goto out;

845
846
	list_for_each_entry(page, &n->slabs_free, lru)
		if (is_slab_pfmemalloc(page))
847
848
849
850
			goto out;

	pfmemalloc_active = false;
out:
851
	spin_unlock_irqrestore(&n->list_lock, flags);
852
853
}

854
static void *__ac_get_obj(struct kmem_cache *cachep, struct array_cache *ac,
855
856
857
858
859
860
861
						gfp_t flags, bool force_refill)
{
	int i;
	void *objp = ac->entry[--ac->avail];

	/* Ensure the caller is allowed to use objects from PFMEMALLOC slab */
	if (unlikely(is_obj_pfmemalloc(objp))) {
862
		struct kmem_cache_node *n;
863
864
865
866
867
868
869

		if (gfp_pfmemalloc_allowed(flags)) {
			clear_obj_pfmemalloc(&objp);
			return objp;
		}

		/* The caller cannot use PFMEMALLOC objects, find another one */
870
		for (i = 0; i < ac->avail; i++) {
871
872
873
874
875
876
877
878
879
880
881
882
883
			/* If a !PFMEMALLOC object is found, swap them */
			if (!is_obj_pfmemalloc(ac->entry[i])) {
				objp = ac->entry[i];
				ac->entry[i] = ac->entry[ac->avail];
				ac->entry[ac->avail] = objp;
				return objp;
			}
		}

		/*
		 * If there are empty slabs on the slabs_free list and we are
		 * being forced to refill the cache, mark this one !pfmemalloc.
		 */
884
885
		n = cachep->node[numa_mem_id()];
		if (!list_empty(&n->slabs_free) && force_refill) {
886
			struct page *page = virt_to_head_page(objp);
887
			ClearPageSlabPfmemalloc(page);
888
889
890
891
892
893
894
895
896
897
898
899
900
			clear_obj_pfmemalloc(&objp);
			recheck_pfmemalloc_active(cachep, ac);
			return objp;
		}

		/* No !PFMEMALLOC objects available */
		ac->avail++;
		objp = NULL;
	}

	return objp;
}

901
902
903
904
905
906
907
908
909
910
911
912
913
914
static inline void *ac_get_obj(struct kmem_cache *cachep,
			struct array_cache *ac, gfp_t flags, bool force_refill)
{
	void *objp;

	if (unlikely(sk_memalloc_socks()))
		objp = __ac_get_obj(cachep, ac, flags, force_refill);
	else
		objp = ac->entry[--ac->avail];

	return objp;
}

static void *__ac_put_obj(struct kmem_cache *cachep, struct array_cache *ac,
915
916
917
918
								void *objp)
{
	if (unlikely(pfmemalloc_active)) {
		/* Some pfmemalloc slabs exist, check if this is one */
919
		struct page *page = virt_to_head_page(objp);
920
921
922
923
		if (PageSlabPfmemalloc(page))
			set_obj_pfmemalloc(&objp);
	}

924
925
926
927
928
929
930
931
932
	return objp;
}

static inline void ac_put_obj(struct kmem_cache *cachep, struct array_cache *ac,
								void *objp)
{
	if (unlikely(sk_memalloc_socks()))
		objp = __ac_put_obj(cachep, ac, objp);

933
934
935
	ac->entry[ac->avail++] = objp;
}

936
937
938
939
940
941
942
943
944
945
/*
 * Transfer objects in one arraycache to another.
 * Locking must be handled by the caller.
 *
 * Return the number of entries transferred.
 */
static int transfer_objects(struct array_cache *to,
		struct array_cache *from, unsigned int max)
{
	/* Figure out how many entries to transfer */
946
	int nr = min3(from->avail, max, to->limit - to->avail);
947
948
949
950
951
952
953
954
955
956
957
958

	if (!nr)
		return 0;

	memcpy(to->entry + to->avail, from->entry + from->avail -nr,
			sizeof(void *) *nr);

	from->avail -= nr;
	to->avail += nr;
	return nr;
}

959
960
961
#ifndef CONFIG_NUMA

#define drain_alien_cache(cachep, alien) do { } while (0)
962
#define reap_alien(cachep, n) do { } while (0)
963

964
static inline struct array_cache **alloc_alien_cache(int node, int limit, gfp_t gfp)
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
{
	return (struct array_cache **)BAD_ALIEN_MAGIC;
}

static inline void free_alien_cache(struct array_cache **ac_ptr)
{
}

static inline int cache_free_alien(struct kmem_cache *cachep, void *objp)
{
	return 0;
}

static inline void *alternate_node_alloc(struct kmem_cache *cachep,
		gfp_t flags)
{
	return NULL;
}

984
static inline void *____cache_alloc_node(struct kmem_cache *cachep,
985
986
987
988
989
990
991
		 gfp_t flags, int nodeid)
{
	return NULL;
}

#else	/* CONFIG_NUMA */

992
static void *____cache_alloc_node(struct kmem_cache *, gfp_t, int);
993
static void *alternate_node_alloc(struct kmem_cache *, gfp_t);
994

995
static struct array_cache **alloc_alien_cache(int node, int limit, gfp_t gfp)
996
997
{
	struct array_cache **ac_ptr;
998
	int memsize = sizeof(void *) * nr_node_ids;
999
1000
1001
1002
	int i;

	if (limit > 1)
		limit = 12;
1003
	ac_ptr = kzalloc_node(memsize, gfp, node);
1004
1005
	if (ac_ptr) {
		for_each_node(i) {
1006
			if (i == node || !node_online(i))
1007
				continue;
1008
			ac_ptr[i] = alloc_arraycache(node, limit, 0xbaadf00d, gfp);
1009
			if (!ac_ptr[i]) {
1010
				for (i--; i >= 0; i--)
1011
1012
1013
1014
1015
1016
1017
1018
1019
					kfree(ac_ptr[i]);
				kfree(ac_ptr);
				return NULL;
			}
		}
	}
	return ac_ptr;
}

Pekka Enberg's avatar
Pekka Enberg committed
1020
static void free_alien_cache(struct array_cache **ac_ptr)
1021
1022
1023
1024
1025
1026
{
	int i;

	if (!ac_ptr)
		return;
	for_each_node(i)
1027
	    kfree(ac_ptr[i]);
1028
1029
1030
	kfree(ac_ptr);
}

1031
static void __drain_alien_cache(struct kmem_cache *cachep,
Pekka Enberg's avatar
Pekka Enberg committed
1032
				struct array_cache *ac, int node)
1033
{
1034
	struct kmem_cache_node *n = cachep->node[node];
1035
1036

	if (ac->avail) {
1037
		spin_lock(&n->list_lock);
1038
1039
1040
1041
1042
		/*
		 * Stuff objects into the remote nodes shared array first.
		 * That way we could avoid the overhead of putting the objects
		 * into the free lists and getting them back later.
		 */
1043
1044
		if (n->shared)
			transfer_objects(n->shared, ac, ac->limit);
1045

1046
		free_block(cachep, ac->entry, ac->avail, node);
1047
		ac->avail = 0;
1048
		spin_unlock(&n->list_lock);
1049
1050
1051
	}
}

1052
1053
1054
/*
 * Called from cache_reap() to regularly drain alien caches round robin.
 */
1055
static void reap_alien(struct kmem_cache *cachep, struct kmem_cache_node *n)
1056
{
1057
	int node = __this_cpu_read(slab_reap_node);
1058

1059
1060
	if (n->alien) {
		struct array_cache *ac = n->alien[node];
1061
1062

		if (ac && ac->avail && spin_trylock_irq(&ac->lock)) {
1063
1064
1065
1066
1067
1068
			__drain_alien_cache(cachep, ac, node);
			spin_unlock_irq(&ac->lock);
		}
	}
}

Andrew Morton's avatar
Andrew Morton committed
1069
1070
static void drain_alien_cache(struct kmem_cache *cachep,
				struct array_cache **alien)
1071
{
1072
	int i = 0;
1073
1074
1075
1076
	struct array_cache *ac;
	unsigned long flags;

	for_each_online_node(i) {
1077
		ac = alien[i];
1078
1079
1080
1081
1082
1083
1084
		if (ac) {
			spin_lock_irqsave(&ac->lock, flags);
			__drain_alien_cache(cachep, ac, i);
			spin_unlock_irqrestore(&ac->lock, flags);
		}
	}
}
1085

1086
static inline int cache_free_alien(struct kmem_cache *cachep, void *objp)
1087
{
1088
	int nodeid = page_to_nid(virt_to_page(objp));
1089
	struct kmem_cache_node *n;
1090
	struct array_cache *alien = NULL;
1091
1092
	int node;

1093
	node = numa_mem_id();
1094
1095
1096
1097
1098

	/*
	 * Make sure we are not freeing a object from another node to the array
	 * cache on this cpu.
	 */
1099
	if (likely(nodeid == node))
1100
1101
		return 0;

1102
	n = cachep->node[node];
1103
	STATS_INC_NODEFREES(cachep);
1104
1105
	if (n->alien && n->alien[nodeid]) {
		alien = n->alien[nodeid];
1106
		spin_lock(&alien->lock);
1107
1108
1109
1110
		if (unlikely(alien->avail == alien->limit)) {
			STATS_INC_ACOVERFLOW(cachep);
			__drain_alien_cache(cachep, alien, nodeid);
		}
1111
		ac_put_obj(cachep, alien, objp);
1112
1113
		spin_unlock(&alien->lock);
	} else {
1114
		spin_lock(&(cachep->node[nodeid])->list_lock);
1115
		free_block(cachep, &objp, 1, nodeid);
1116
		spin_unlock(&(cachep->node[nodeid])->list_lock);
1117
1118
1119
	}
	return 1;
}
1120
1121
#endif

1122
/*
1123
 * Allocates and initializes node for a node on each slab cache, used for
1124
 * either memory or cpu hotplug.  If memory is being hot-added, the kmem_cache_node
1125
 * will be allocated off-node since memory is not yet online for the new node.
1126
 * When hotplugging memory or a cpu, existing node are not replaced if
1127
1128
 * already in use.
 *
1129
 * Must hold slab_mutex.
1130
 */
1131
static int init_cache_node_node(int node)
1132
1133
{
	struct kmem_cache *cachep;
1134
	struct kmem_cache_node *n;
1135
	const int memsize = sizeof(struct kmem_cache_node);
1136

1137
	list_for_each_entry(cachep, &slab_caches, list) {
1138
		/*
1139
		 * Set up the kmem_cache_node for cpu before we can
1140
1141
1142
		 * begin anything. Make sure some other cpu on this
		 * node has not already allocated this
		 */
1143
		if (!cachep->node[node]) {
1144
1145
			n = kmalloc_node(memsize, GFP_KERNEL, node);
			if (!n)
1146
				return -ENOMEM;
1147
			kmem_cache_node_init(n);
1148
1149
			n->next_reap = jiffies + REAPTIMEOUT_NODE +
			    ((unsigned long)cachep) % REAPTIMEOUT_NODE;
1150
1151

			/*
1152
1153
			 * The kmem_cache_nodes don't come and go as CPUs
			 * come and go.  slab_mutex is sufficient
1154
1155
			 * protection here.
			 */
1156
			cachep->node[node] = n;
1157
1158
		}

1159
1160
		spin_lock_irq(&cachep->node[node]->list_lock);
		cachep->node[node]->free_limit =
1161
1162
			(1 + nr_cpus_node(node)) *
			cachep->batchcount + cachep->num;
1163
		spin_unlock_irq(&cachep->node[node]->list_lock);
1164
1165
1166
1167
	}
	return 0;
}

Wanpeng Li's avatar