slab.c 110 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
/*
 * 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
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
/*
 * 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;
183
	spinlock_t lock;
184
	void *entry[];	/*
Andrew Morton's avatar
Andrew Morton committed
185
186
187
			 * Must have this definition in here for the proper
			 * alignment of array_cache. Also simplifies accessing
			 * the entries.
188
189
190
191
			 *
			 * 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
192
			 */
Linus Torvalds's avatar
Linus Torvalds committed
193
194
};

195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
#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
212
213
214
/*
 * 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
215
216
217
218
 */
#define BOOT_CPUCACHE_ENTRIES	1
struct arraycache_init {
	struct array_cache cache;
219
	void *entries[BOOT_CPUCACHE_ENTRIES];
Linus Torvalds's avatar
Linus Torvalds committed
220
221
};

222
223
224
/*
 * Need this for bootstrapping a per node allocator.
 */
225
#define NUM_INIT_LISTS (3 * MAX_NUMNODES)
226
static struct kmem_cache_node __initdata init_kmem_cache_node[NUM_INIT_LISTS];
227
#define	CACHE_CACHE 0
228
#define	SIZE_AC MAX_NUMNODES
229
#define	SIZE_NODE (2 * MAX_NUMNODES)
230

231
static int drain_freelist(struct kmem_cache *cache,
232
			struct kmem_cache_node *n, int tofree);
233
234
static void free_block(struct kmem_cache *cachep, void **objpp, int len,
			int node);
235
static int enable_cpucache(struct kmem_cache *cachep, gfp_t gfp);
236
static void cache_reap(struct work_struct *unused);
237

238
239
static int slab_early_init = 1;

240
#define INDEX_AC kmalloc_index(sizeof(struct arraycache_init))
241
#define INDEX_NODE kmalloc_index(sizeof(struct kmem_cache_node))
Linus Torvalds's avatar
Linus Torvalds committed
242

243
static void kmem_cache_node_init(struct kmem_cache_node *parent)
244
245
246
247
248
249
{
	INIT_LIST_HEAD(&parent->slabs_full);
	INIT_LIST_HEAD(&parent->slabs_partial);
	INIT_LIST_HEAD(&parent->slabs_free);
	parent->shared = NULL;
	parent->alien = NULL;
250
	parent->colour_next = 0;
251
252
253
254
255
	spin_lock_init(&parent->list_lock);
	parent->free_objects = 0;
	parent->free_touched = 0;
}

Andrew Morton's avatar
Andrew Morton committed
256
257
258
#define MAKE_LIST(cachep, listp, slab, nodeid)				\
	do {								\
		INIT_LIST_HEAD(listp);					\
259
		list_splice(&(cachep->node[nodeid]->slab), listp);	\
260
261
	} while (0)

Andrew Morton's avatar
Andrew Morton committed
262
263
#define	MAKE_ALL_LISTS(cachep, ptr, nodeid)				\
	do {								\
264
265
266
267
	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
268
269
270
271
272

#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
273
274
275
/*
 * Optimization question: fewer reaps means less probability for unnessary
 * cpucache drain/refill cycles.
Linus Torvalds's avatar
Linus Torvalds committed
276
 *
Adrian Bunk's avatar
Adrian Bunk committed
277
 * OTOH the cpuarrays can contain lots of objects,
Linus Torvalds's avatar
Linus Torvalds committed
278
279
280
281
282
283
284
285
286
287
 * which could lock up otherwise freeable slabs.
 */
#define REAPTIMEOUT_CPUC	(2*HZ)
#define REAPTIMEOUT_LIST3	(4*HZ)

#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++)
288
#define	STATS_ADD_REAPED(x,y)	((x)->reaped += (y))
Andrew Morton's avatar
Andrew Morton committed
289
290
291
292
293
#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
294
295
#define	STATS_INC_ERR(x)	((x)->errors++)
#define	STATS_INC_NODEALLOCS(x)	((x)->node_allocs++)
296
#define	STATS_INC_NODEFREES(x)	((x)->node_frees++)
297
#define STATS_INC_ACOVERFLOW(x)   ((x)->node_overflow++)
Andrew Morton's avatar
Andrew Morton committed
298
299
300
301
302
#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
303
304
305
306
307
308
309
310
311
#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)
312
#define	STATS_ADD_REAPED(x,y)	do { (void)(y); } while (0)
Linus Torvalds's avatar
Linus Torvalds committed
313
314
315
#define	STATS_SET_HIGH(x)	do { } while (0)
#define	STATS_INC_ERR(x)	do { } while (0)
#define	STATS_INC_NODEALLOCS(x)	do { } while (0)
316
#define	STATS_INC_NODEFREES(x)	do { } while (0)
317
#define STATS_INC_ACOVERFLOW(x)   do { } while (0)
Andrew Morton's avatar
Andrew Morton committed
318
#define	STATS_SET_FREEABLE(x, i) do { } while (0)
Linus Torvalds's avatar
Linus Torvalds committed
319
320
321
322
323
324
325
326
#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
327
328
/*
 * memory layout of objects:
Linus Torvalds's avatar
Linus Torvalds committed
329
 * 0		: objp
330
 * 0 .. cachep->obj_offset - BYTES_PER_WORD - 1: padding. This ensures that
Linus Torvalds's avatar
Linus Torvalds committed
331
332
 * 		the end of an object is aligned with the end of the real
 * 		allocation. Catches writes behind the end of the allocation.
333
 * cachep->obj_offset - BYTES_PER_WORD .. cachep->obj_offset - 1:
Linus Torvalds's avatar
Linus Torvalds committed
334
 * 		redzone word.
335
 * cachep->obj_offset: The real object.
336
337
 * 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
338
 *					[BYTES_PER_WORD long]
Linus Torvalds's avatar
Linus Torvalds committed
339
 */
340
static int obj_offset(struct kmem_cache *cachep)
Linus Torvalds's avatar
Linus Torvalds committed
341
{
342
	return cachep->obj_offset;
Linus Torvalds's avatar
Linus Torvalds committed
343
344
}

345
static unsigned long long *dbg_redzone1(struct kmem_cache *cachep, void *objp)
Linus Torvalds's avatar
Linus Torvalds committed
346
347
{
	BUG_ON(!(cachep->flags & SLAB_RED_ZONE));
348
349
	return (unsigned long long*) (objp + obj_offset(cachep) -
				      sizeof(unsigned long long));
Linus Torvalds's avatar
Linus Torvalds committed
350
351
}

352
static unsigned long long *dbg_redzone2(struct kmem_cache *cachep, void *objp)
Linus Torvalds's avatar
Linus Torvalds committed
353
354
355
{
	BUG_ON(!(cachep->flags & SLAB_RED_ZONE));
	if (cachep->flags & SLAB_STORE_USER)
356
		return (unsigned long long *)(objp + cachep->size -
357
					      sizeof(unsigned long long) -
David Woodhouse's avatar
David Woodhouse committed
358
					      REDZONE_ALIGN);
359
	return (unsigned long long *) (objp + cachep->size -
360
				       sizeof(unsigned long long));
Linus Torvalds's avatar
Linus Torvalds committed
361
362
}

363
static void **dbg_userword(struct kmem_cache *cachep, void *objp)
Linus Torvalds's avatar
Linus Torvalds committed
364
365
{
	BUG_ON(!(cachep->flags & SLAB_STORE_USER));
366
	return (void **)(objp + cachep->size - BYTES_PER_WORD);
Linus Torvalds's avatar
Linus Torvalds committed
367
368
369
370
}

#else

371
#define obj_offset(x)			0
372
373
#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
374
375
376
377
378
#define dbg_userword(cachep, objp)	({BUG(); (void **)NULL;})

#endif

/*
379
380
 * 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
381
 */
382
383
384
#define	SLAB_MAX_ORDER_HI	1
#define	SLAB_MAX_ORDER_LO	0
static int slab_max_order = SLAB_MAX_ORDER_LO;
385
static bool slab_max_order_set __initdata;
Linus Torvalds's avatar
Linus Torvalds committed
386

387
388
static inline struct kmem_cache *virt_to_cache(const void *obj)
{
389
	struct page *page = virt_to_head_page(obj);
390
	return page->slab_cache;
391
392
}

393
static inline void *index_to_obj(struct kmem_cache *cache, struct page *page,
394
395
				 unsigned int idx)
{
396
	return page->s_mem + cache->size * idx;
397
398
}

399
/*
400
401
402
 * 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
403
404
405
 *   reciprocal_divide(offset, cache->reciprocal_buffer_size)
 */
static inline unsigned int obj_to_index(const struct kmem_cache *cache,
406
					const struct page *page, void *obj)
407
{
408
	u32 offset = (obj - page->s_mem);
409
	return reciprocal_divide(offset, cache->reciprocal_buffer_size);
410
411
}

Linus Torvalds's avatar
Linus Torvalds committed
412
static struct arraycache_init initarray_generic =
413
    { {0, BOOT_CPUCACHE_ENTRIES, 1, 0} };
Linus Torvalds's avatar
Linus Torvalds committed
414
415

/* internal cache of cache description objs */
416
static struct kmem_cache kmem_cache_boot = {
417
418
419
	.batchcount = 1,
	.limit = BOOT_CPUCACHE_ENTRIES,
	.shared = 1,
420
	.size = sizeof(struct kmem_cache),
421
	.name = "kmem_cache",
Linus Torvalds's avatar
Linus Torvalds committed
422
423
};

424
425
#define BAD_ALIEN_MAGIC 0x01020304ul

426
427
428
429
430
431
432
433
#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.
434
435
436
437
 *
 * 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
438
 */
439
440
441
static struct lock_class_key on_slab_l3_key;
static struct lock_class_key on_slab_alc_key;

442
443
444
445
446
447
448
449
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;
450
	struct kmem_cache_node *n;
451
452
	int r;

453
454
	n = cachep->node[q];
	if (!n)
455
456
		return;

457
458
	lockdep_set_class(&n->list_lock, l3_key);
	alc = n->alien;
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
	/*
	 * 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);
}

487
static void init_node_lock_keys(int q)
488
{
489
	int i;
490

491
	if (slab_state < UP)
492
493
		return;

Christoph Lameter's avatar
Christoph Lameter committed
494
	for (i = 1; i <= KMALLOC_SHIFT_HIGH; i++) {
495
		struct kmem_cache_node *n;
496
497
498
499
		struct kmem_cache *cache = kmalloc_caches[i];

		if (!cache)
			continue;
500

501
502
		n = cache->node[q];
		if (!n || OFF_SLAB(cache))
503
			continue;
504

505
		slab_set_lock_classes(cache, &on_slab_l3_key,
506
				&on_slab_alc_key, q);
507
508
	}
}
509

510
511
static void on_slab_lock_classes_node(struct kmem_cache *cachep, int q)
{
512
	if (!cachep->node[q])
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
		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);
}

528
529
530
531
532
533
534
static inline void init_lock_keys(void)
{
	int node;

	for_each_node(node)
		init_node_lock_keys(node);
}
535
#else
536
537
538
539
static void init_node_lock_keys(int q)
{
}

540
static inline void init_lock_keys(void)
541
542
{
}
543

544
545
546
547
548
549
550
551
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)
{
}

552
553
554
555
556
557
558
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)
{
}
559
560
#endif

561
static DEFINE_PER_CPU(struct delayed_work, slab_reap_work);
Linus Torvalds's avatar
Linus Torvalds committed
562

563
static inline struct array_cache *cpu_cache_get(struct kmem_cache *cachep)
Linus Torvalds's avatar
Linus Torvalds committed
564
565
566
567
{
	return cachep->array[smp_processor_id()];
}

568
569
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
570
{
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
	int nr_objs;
	size_t freelist_size;

	/*
	 * 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.
	 */
	nr_objs = slab_size / (buffer_size + idx_size);

	/*
	 * This calculated number will be either the right
	 * amount, or one greater than what we want.
	 */
	freelist_size = slab_size - nr_objs * buffer_size;
	if (freelist_size < ALIGN(nr_objs * idx_size, align))
		nr_objs--;

	return nr_objs;
593
}
Linus Torvalds's avatar
Linus Torvalds committed
594

Andrew Morton's avatar
Andrew Morton committed
595
596
597
/*
 * Calculate the number of objects and left-over bytes for a given buffer size.
 */
598
599
600
601
602
603
604
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
605

606
607
608
609
610
	/*
	 * 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
611
	 * - One unsigned int for each object
612
613
614
615
616
617
618
619
620
621
622
623
624
	 * - 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 {
625
626
627
		nr_objs = calculate_nr_objs(slab_size, buffer_size,
					sizeof(unsigned int), align);
		mgmt_size = ALIGN(nr_objs * sizeof(unsigned int), align);
628
629
630
	}
	*num = nr_objs;
	*left_over = slab_size - nr_objs*buffer_size - mgmt_size;
Linus Torvalds's avatar
Linus Torvalds committed
631
632
}

633
#if DEBUG
634
#define slab_error(cachep, msg) __slab_error(__func__, cachep, msg)
Linus Torvalds's avatar
Linus Torvalds committed
635

Andrew Morton's avatar
Andrew Morton committed
636
637
static void __slab_error(const char *function, struct kmem_cache *cachep,
			char *msg)
Linus Torvalds's avatar
Linus Torvalds committed
638
639
{
	printk(KERN_ERR "slab error in %s(): cache `%s': %s\n",
640
	       function, cachep->name, msg);
Linus Torvalds's avatar
Linus Torvalds committed
641
	dump_stack();
642
	add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
Linus Torvalds's avatar
Linus Torvalds committed
643
}
644
#endif
Linus Torvalds's avatar
Linus Torvalds committed
645

646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
/*
 * 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);

662
663
664
665
666
667
668
669
670
671
672
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);

673
674
675
676
677
678
679
#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.
 */
680
static DEFINE_PER_CPU(unsigned long, slab_reap_node);
681
682
683
684
685

static void init_reap_node(int cpu)
{
	int node;

686
	node = next_node(cpu_to_mem(cpu), node_online_map);
687
	if (node == MAX_NUMNODES)
688
		node = first_node(node_online_map);
689

690
	per_cpu(slab_reap_node, cpu) = node;
691
692
693
694
}

static void next_reap_node(void)
{
695
	int node = __this_cpu_read(slab_reap_node);
696
697
698
699

	node = next_node(node, node_online_map);
	if (unlikely(node >= MAX_NUMNODES))
		node = first_node(node_online_map);
700
	__this_cpu_write(slab_reap_node, node);
701
702
703
704
705
706
707
}

#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
708
709
710
711
712
713
714
/*
 * 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.
 */
715
static void start_cpu_timer(int cpu)
Linus Torvalds's avatar
Linus Torvalds committed
716
{
717
	struct delayed_work *reap_work = &per_cpu(slab_reap_work, cpu);
Linus Torvalds's avatar
Linus Torvalds committed
718
719
720
721
722
723

	/*
	 * When this gets called from do_initcalls via cpucache_init(),
	 * init_workqueues() has already run, so keventd will be setup
	 * at that time.
	 */
724
	if (keventd_up() && reap_work->work.func == NULL) {
725
		init_reap_node(cpu);
726
		INIT_DEFERRABLE_WORK(reap_work, cache_reap);
727
728
		schedule_delayed_work_on(cpu, reap_work,
					__round_jiffies_relative(HZ, cpu));
Linus Torvalds's avatar
Linus Torvalds committed
729
730
731
	}
}

732
static struct array_cache *alloc_arraycache(int node, int entries,
733
					    int batchcount, gfp_t gfp)
Linus Torvalds's avatar
Linus Torvalds committed
734
{
735
	int memsize = sizeof(void *) * entries + sizeof(struct array_cache);
Linus Torvalds's avatar
Linus Torvalds committed
736
737
	struct array_cache *nc = NULL;

738
	nc = kmalloc_node(memsize, gfp, node);
739
740
	/*
	 * The array_cache structures contain pointers to free object.
Lucas De Marchi's avatar
Lucas De Marchi committed
741
	 * However, when such objects are allocated or transferred to another
742
743
744
745
746
	 * 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
747
748
749
750
751
	if (nc) {
		nc->avail = 0;
		nc->limit = entries;
		nc->batchcount = batchcount;
		nc->touched = 0;
752
		spin_lock_init(&nc->lock);
Linus Torvalds's avatar
Linus Torvalds committed
753
754
755
756
	}
	return nc;
}

757
static inline bool is_slab_pfmemalloc(struct page *page)
758
759
760
761
762
763
764
765
{
	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)
{
766
	struct kmem_cache_node *n = cachep->node[numa_mem_id()];
767
	struct page *page;
768
769
770
771
772
	unsigned long flags;

	if (!pfmemalloc_active)
		return;

773
	spin_lock_irqsave(&n->list_lock, flags);
774
775
	list_for_each_entry(page, &n->slabs_full, lru)
		if (is_slab_pfmemalloc(page))
776
777
			goto out;

778
779
	list_for_each_entry(page, &n->slabs_partial, lru)
		if (is_slab_pfmemalloc(page))
780
781
			goto out;

782
783
	list_for_each_entry(page, &n->slabs_free, lru)
		if (is_slab_pfmemalloc(page))
784
785
786
787
			goto out;

	pfmemalloc_active = false;
out:
788
	spin_unlock_irqrestore(&n->list_lock, flags);
789
790
}

791
static void *__ac_get_obj(struct kmem_cache *cachep, struct array_cache *ac,
792
793
794
795
796
797
798
						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))) {
799
		struct kmem_cache_node *n;
800
801
802
803
804
805
806

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

		/* The caller cannot use PFMEMALLOC objects, find another one */
807
		for (i = 0; i < ac->avail; i++) {
808
809
810
811
812
813
814
815
816
817
818
819
820
			/* 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.
		 */
821
822
		n = cachep->node[numa_mem_id()];
		if (!list_empty(&n->slabs_free) && force_refill) {
823
			struct page *page = virt_to_head_page(objp);
824
			ClearPageSlabPfmemalloc(page);
825
826
827
828
829
830
831
832
833
834
835
836
837
			clear_obj_pfmemalloc(&objp);
			recheck_pfmemalloc_active(cachep, ac);
			return objp;
		}

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

	return objp;
}

838
839
840
841
842
843
844
845
846
847
848
849
850
851
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,
852
853
854
855
								void *objp)
{
	if (unlikely(pfmemalloc_active)) {
		/* Some pfmemalloc slabs exist, check if this is one */
856
		struct page *page = virt_to_head_page(objp);
857
858
859
860
		if (PageSlabPfmemalloc(page))
			set_obj_pfmemalloc(&objp);
	}

861
862
863
864
865
866
867
868
869
	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);

870
871
872
	ac->entry[ac->avail++] = objp;
}

873
874
875
876
877
878
879
880
881
882
/*
 * 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 */
883
	int nr = min3(from->avail, max, to->limit - to->avail);
884
885
886
887
888
889
890
891
892
893
894
895

	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;
}

896
897
898
#ifndef CONFIG_NUMA

#define drain_alien_cache(cachep, alien) do { } while (0)
899
#define reap_alien(cachep, n) do { } while (0)
900

901
static inline struct array_cache **alloc_alien_cache(int node, int limit, gfp_t gfp)
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
{
	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;
}

921
static inline void *____cache_alloc_node(struct kmem_cache *cachep,
922
923
924
925
926
927
928
		 gfp_t flags, int nodeid)
{
	return NULL;
}

#else	/* CONFIG_NUMA */

929
static void *____cache_alloc_node(struct kmem_cache *, gfp_t, int);
930
static void *alternate_node_alloc(struct kmem_cache *, gfp_t);
931

932
static struct array_cache **alloc_alien_cache(int node, int limit, gfp_t gfp)
933
934
{
	struct array_cache **ac_ptr;
935
	int memsize = sizeof(void *) * nr_node_ids;
936
937
938
939
	int i;

	if (limit > 1)
		limit = 12;
940
	ac_ptr = kzalloc_node(memsize, gfp, node);
941
942
	if (ac_ptr) {
		for_each_node(i) {
943
			if (i == node || !node_online(i))
944
				continue;
945
			ac_ptr[i] = alloc_arraycache(node, limit, 0xbaadf00d, gfp);
946
			if (!ac_ptr[i]) {
947
				for (i--; i >= 0; i--)
948
949
950
951
952
953
954
955
956
					kfree(ac_ptr[i]);
				kfree(ac_ptr);
				return NULL;
			}
		}
	}
	return ac_ptr;
}

Pekka Enberg's avatar
Pekka Enberg committed
957
static void free_alien_cache(struct array_cache **ac_ptr)
958
959
960
961
962
963
{
	int i;

	if (!ac_ptr)
		return;
	for_each_node(i)
964
	    kfree(ac_ptr[i]);
965
966
967
	kfree(ac_ptr);
}

968
static void __drain_alien_cache(struct kmem_cache *cachep,
Pekka Enberg's avatar
Pekka Enberg committed
969
				struct array_cache *ac, int node)
970
{
971
	struct kmem_cache_node *n = cachep->node[node];
972
973

	if (ac->avail) {
974
		spin_lock(&n->list_lock);
975
976
977
978
979
		/*
		 * 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.
		 */
980
981
		if (n->shared)
			transfer_objects(n->shared, ac, ac->limit);
982

983
		free_block(cachep, ac->entry, ac->avail, node);
984
		ac->avail = 0;
985
		spin_unlock(&n->list_lock);
986
987
988
	}
}

989
990
991
/*
 * Called from cache_reap() to regularly drain alien caches round robin.
 */
992
static void reap_alien(struct kmem_cache *cachep, struct kmem_cache_node *n)
993
{
994
	int node = __this_cpu_read(slab_reap_node);
995

996
997
	if (n->alien) {
		struct array_cache *ac = n->alien[node];
998
999

		if (ac && ac->avail && spin_trylock_irq(&ac->lock)) {
1000
1001
1002
1003
1004
1005
			__drain_alien_cache(cachep, ac, node);
			spin_unlock_irq(&ac->lock);
		}
	}
}

Andrew Morton's avatar
Andrew Morton committed
1006
1007
static void drain_alien_cache(struct kmem_cache *cachep,
				struct array_cache **alien)
1008
{
1009
	int i = 0;
1010
1011
1012
1013
	struct array_cache *ac;
	unsigned long flags;

	for_each_online_node(i) {
1014
		ac = alien[i];
1015
1016
1017
1018
1019
1020
1021
		if (ac) {
			spin_lock_irqsave(&ac->lock, flags);
			__drain_alien_cache(cachep, ac, i);
			spin_unlock_irqrestore(&ac->lock, flags);
		}
	}
}
1022

1023
static inline int cache_free_alien(struct kmem_cache *cachep, void *objp)
1024
{
1025
	int nodeid = page_to_nid(virt_to_page(objp));
1026
	struct kmem_cache_node *n;
1027
	struct array_cache *alien = NULL;
1028
1029
	int node;

1030
	node = numa_mem_id();
1031
1032
1033
1034
1035

	/*
	 * Make sure we are not freeing a object from another node to the array
	 * cache on this cpu.
	 */
1036
	if (likely(nodeid == node))
1037
1038
		return 0;

1039
	n = cachep->node[node];
1040
	STATS_INC_NODEFREES(cachep);
1041
1042
	if (n->alien && n->alien[nodeid]) {
		alien = n->alien[nodeid];
1043
		spin_lock(&alien->lock);
1044
1045
1046
1047
		if (unlikely(alien->avail == alien->limit)) {
			STATS_INC_ACOVERFLOW(cachep);
			__drain_alien_cache(cachep, alien, nodeid);
		}
1048
		ac_put_obj(cachep, alien, objp);
1049
1050
		spin_unlock(&alien->lock);
	} else {
1051
		spin_lock(&(cachep->node[nodeid])->list_lock);
1052
		free_block(cachep, &objp, 1, nodeid);
1053
		spin_unlock(&(cachep->node[nodeid])->list_lock);
1054
1055
1056
	}
	return 1;
}
1057
1058
#endif

1059
/*
1060
 * Allocates and initializes node for a node on each slab cache, used for
1061
 * either memory or cpu hotplug.  If memory is being hot-added, the kmem_cache_node
1062
 * will be allocated off-node since memory is not yet online for the new node.
1063
 * When hotplugging memory or a cpu, existing node are not replaced if
1064
1065
 * already in use.
 *
1066
 * Must hold slab_mutex.
1067
 */
1068
static int init_cache_node_node(int node)
1069
1070
{
	struct kmem_cache *cachep;
1071
	struct kmem_cache_node *n;
1072
	const int memsize = sizeof(struct kmem_cache_node);
1073

1074
	list_for_each_entry(cachep, &slab_caches, list) {
1075
1076
1077
1078
1079
		/*
		 * Set up the size64 kmemlist for cpu before we can
		 * begin anything. Make sure some other cpu on this
		 * node has not already allocated this
		 */
1080
		if (!cachep->node[node]) {
1081
1082
			n = kmalloc_node(memsize, GFP_KERNEL, node);
			if (!n)
1083
				return -ENOMEM;
1084
1085
			kmem_cache_node_init(n);
			n->next_reap = jiffies + REAPTIMEOUT_LIST3 +
1086
1087
1088
1089
			    ((unsigned long)cachep) % REAPTIMEOUT_LIST3;

			/*
			 * The l3s don't come and go as CPUs come and
1090
			 * go.  slab_mutex is sufficient
1091
1092
			 * protection here.
			 */
1093
			cachep->node[node] = n;
1094
1095
		}

1096
1097
		spin_lock_irq(&cachep->node[node]->list_lock);
		cachep->node[node]->free_limit =
1098
1099
			(1 + nr_cpus_node(node)) *
			cachep->batchcount + cachep->num;
1100
		spin_unlock_irq(&cachep->node[node]->list_lock);
1101
1102
1103
1104
	}
	return 0;
}

1105
1106
1107
1108
1109
1110
static inline int slabs_tofree(struct kmem_cache *cachep,
						struct kmem_cache_node *n)
{
	return (n->free_objects + cachep->num - 1) / cachep->num;
}

1111
static void cpuup_canceled(long cpu)
1112
1113
{
	struct kmem_cache *cachep;
1114
	struct kmem_cache_node *n = NULL;
1115
	int node = cpu_to_mem(cpu);
1116
	const struct cpumask *mask = cpumask_of_node(node);
1117

1118
	list_for_each_entry(cachep, &slab_caches, list) {
1119
1120
1121
1122
1123
1124
1125
		struct array_cache *nc;
		struct array_cache *shared;
		struct array_cache **alien;

		/* cpu is dead; no one can alloc from it. */
		nc = cachep->array[cpu];
		cachep->array[cpu] = NULL;
1126
		n = cachep->node[node];
1127

1128
		if (!n)
1129
1130
			goto free_array_cache;

1131
		spin_lock_irq(&n->list_lock);
1132

1133
1134
		/* Free limit for this kmem_cache_node */
		n->free_limit -= cachep->batchcount;
1135
1136
1137
		if (nc)
			free_block(cachep, nc->entry, nc->avail, node);

1138
		if (!cpumask_empty(mask)) {
1139
			spin_unlock_irq(&n->list_lock);
1140
1141
1142
			goto free_array_cache;
		}

1143
		shared = n->shared;
1144
1145
1146
		if (shared) {
			free_block(cachep, shared->entry,
				   shared->avail, node);
1147
			n->shared = NULL;
1148
1149
		}

1150
1151
		alien = n->alien;
		n->alien = NULL;
1152

1153
		spin_unlock_irq(&n->list_lock);