slab.c 107 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

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

Joonsoo Kim's avatar
Joonsoo Kim committed
195
196
197
198
199
struct alien_cache {
	spinlock_t lock;
	struct array_cache ac;
};

200
201
202
/*
 * Need this for bootstrapping a per node allocator.
 */
203
#define NUM_INIT_LISTS (2 * MAX_NUMNODES)
204
static struct kmem_cache_node __initdata init_kmem_cache_node[NUM_INIT_LISTS];
205
#define	CACHE_CACHE 0
206
#define	SIZE_NODE (MAX_NUMNODES)
207

208
static int drain_freelist(struct kmem_cache *cache,
209
			struct kmem_cache_node *n, int tofree);
210
static void free_block(struct kmem_cache *cachep, void **objpp, int len,
211
212
			int node, struct list_head *list);
static void slabs_destroy(struct kmem_cache *cachep, struct list_head *list);
213
static int enable_cpucache(struct kmem_cache *cachep, gfp_t gfp);
214
static void cache_reap(struct work_struct *unused);
215

216
217
static int slab_early_init = 1;

218
#define INDEX_NODE kmalloc_index(sizeof(struct kmem_cache_node))
Linus Torvalds's avatar
Linus Torvalds committed
219

220
static void kmem_cache_node_init(struct kmem_cache_node *parent)
221
222
223
224
225
226
{
	INIT_LIST_HEAD(&parent->slabs_full);
	INIT_LIST_HEAD(&parent->slabs_partial);
	INIT_LIST_HEAD(&parent->slabs_free);
	parent->shared = NULL;
	parent->alien = NULL;
227
	parent->colour_next = 0;
228
229
230
231
232
	spin_lock_init(&parent->list_lock);
	parent->free_objects = 0;
	parent->free_touched = 0;
}

Andrew Morton's avatar
Andrew Morton committed
233
234
235
#define MAKE_LIST(cachep, listp, slab, nodeid)				\
	do {								\
		INIT_LIST_HEAD(listp);					\
236
		list_splice(&get_node(cachep, nodeid)->slab, listp);	\
237
238
	} while (0)

Andrew Morton's avatar
Andrew Morton committed
239
240
#define	MAKE_ALL_LISTS(cachep, ptr, nodeid)				\
	do {								\
241
242
243
244
	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
245

246
#define CFLGS_OBJFREELIST_SLAB	(0x40000000UL)
Linus Torvalds's avatar
Linus Torvalds committed
247
#define CFLGS_OFF_SLAB		(0x80000000UL)
248
#define	OBJFREELIST_SLAB(x)	((x)->flags & CFLGS_OBJFREELIST_SLAB)
Linus Torvalds's avatar
Linus Torvalds committed
249
250
251
#define	OFF_SLAB(x)	((x)->flags & CFLGS_OFF_SLAB)

#define BATCHREFILL_LIMIT	16
Andrew Morton's avatar
Andrew Morton committed
252
253
254
/*
 * Optimization question: fewer reaps means less probability for unnessary
 * cpucache drain/refill cycles.
Linus Torvalds's avatar
Linus Torvalds committed
255
 *
Adrian Bunk's avatar
Adrian Bunk committed
256
 * OTOH the cpuarrays can contain lots of objects,
Linus Torvalds's avatar
Linus Torvalds committed
257
258
 * which could lock up otherwise freeable slabs.
 */
259
260
#define REAPTIMEOUT_AC		(2*HZ)
#define REAPTIMEOUT_NODE	(4*HZ)
Linus Torvalds's avatar
Linus Torvalds committed
261
262
263
264
265
266

#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++)
267
#define	STATS_ADD_REAPED(x,y)	((x)->reaped += (y))
Andrew Morton's avatar
Andrew Morton committed
268
269
270
271
272
#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
273
274
#define	STATS_INC_ERR(x)	((x)->errors++)
#define	STATS_INC_NODEALLOCS(x)	((x)->node_allocs++)
275
#define	STATS_INC_NODEFREES(x)	((x)->node_frees++)
276
#define STATS_INC_ACOVERFLOW(x)   ((x)->node_overflow++)
Andrew Morton's avatar
Andrew Morton committed
277
278
279
280
281
#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
282
283
284
285
286
287
288
289
290
#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)
291
#define	STATS_ADD_REAPED(x,y)	do { (void)(y); } while (0)
Linus Torvalds's avatar
Linus Torvalds committed
292
293
294
#define	STATS_SET_HIGH(x)	do { } while (0)
#define	STATS_INC_ERR(x)	do { } while (0)
#define	STATS_INC_NODEALLOCS(x)	do { } while (0)
295
#define	STATS_INC_NODEFREES(x)	do { } while (0)
296
#define STATS_INC_ACOVERFLOW(x)   do { } while (0)
Andrew Morton's avatar
Andrew Morton committed
297
#define	STATS_SET_FREEABLE(x, i) do { } while (0)
Linus Torvalds's avatar
Linus Torvalds committed
298
299
300
301
302
303
304
305
#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
306
307
/*
 * memory layout of objects:
Linus Torvalds's avatar
Linus Torvalds committed
308
 * 0		: objp
309
 * 0 .. cachep->obj_offset - BYTES_PER_WORD - 1: padding. This ensures that
Linus Torvalds's avatar
Linus Torvalds committed
310
311
 * 		the end of an object is aligned with the end of the real
 * 		allocation. Catches writes behind the end of the allocation.
312
 * cachep->obj_offset - BYTES_PER_WORD .. cachep->obj_offset - 1:
Linus Torvalds's avatar
Linus Torvalds committed
313
 * 		redzone word.
314
 * cachep->obj_offset: The real object.
315
316
 * 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
317
 *					[BYTES_PER_WORD long]
Linus Torvalds's avatar
Linus Torvalds committed
318
 */
319
static int obj_offset(struct kmem_cache *cachep)
Linus Torvalds's avatar
Linus Torvalds committed
320
{
321
	return cachep->obj_offset;
Linus Torvalds's avatar
Linus Torvalds committed
322
323
}

324
static unsigned long long *dbg_redzone1(struct kmem_cache *cachep, void *objp)
Linus Torvalds's avatar
Linus Torvalds committed
325
326
{
	BUG_ON(!(cachep->flags & SLAB_RED_ZONE));
327
328
	return (unsigned long long*) (objp + obj_offset(cachep) -
				      sizeof(unsigned long long));
Linus Torvalds's avatar
Linus Torvalds committed
329
330
}

331
static unsigned long long *dbg_redzone2(struct kmem_cache *cachep, void *objp)
Linus Torvalds's avatar
Linus Torvalds committed
332
333
334
{
	BUG_ON(!(cachep->flags & SLAB_RED_ZONE));
	if (cachep->flags & SLAB_STORE_USER)
335
		return (unsigned long long *)(objp + cachep->size -
336
					      sizeof(unsigned long long) -
David Woodhouse's avatar
David Woodhouse committed
337
					      REDZONE_ALIGN);
338
	return (unsigned long long *) (objp + cachep->size -
339
				       sizeof(unsigned long long));
Linus Torvalds's avatar
Linus Torvalds committed
340
341
}

342
static void **dbg_userword(struct kmem_cache *cachep, void *objp)
Linus Torvalds's avatar
Linus Torvalds committed
343
344
{
	BUG_ON(!(cachep->flags & SLAB_STORE_USER));
345
	return (void **)(objp + cachep->size - BYTES_PER_WORD);
Linus Torvalds's avatar
Linus Torvalds committed
346
347
348
349
}

#else

350
#define obj_offset(x)			0
351
352
#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
353
354
355
356
#define dbg_userword(cachep, objp)	({BUG(); (void **)NULL;})

#endif

357
358
#ifdef CONFIG_DEBUG_SLAB_LEAK

359
static inline bool is_store_user_clean(struct kmem_cache *cachep)
360
{
361
362
	return atomic_read(&cachep->store_user_clean) == 1;
}
363

364
365
366
367
static inline void set_store_user_clean(struct kmem_cache *cachep)
{
	atomic_set(&cachep->store_user_clean, 1);
}
368

369
370
371
372
static inline void set_store_user_dirty(struct kmem_cache *cachep)
{
	if (is_store_user_clean(cachep))
		atomic_set(&cachep->store_user_clean, 0);
373
374
375
}

#else
376
static inline void set_store_user_dirty(struct kmem_cache *cachep) {}
377
378
379

#endif

Linus Torvalds's avatar
Linus Torvalds committed
380
/*
381
382
 * 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
383
 */
384
385
386
#define	SLAB_MAX_ORDER_HI	1
#define	SLAB_MAX_ORDER_LO	0
static int slab_max_order = SLAB_MAX_ORDER_LO;
387
static bool slab_max_order_set __initdata;
Linus Torvalds's avatar
Linus Torvalds committed
388

389
390
static inline struct kmem_cache *virt_to_cache(const void *obj)
{
391
	struct page *page = virt_to_head_page(obj);
392
	return page->slab_cache;
393
394
}

395
static inline void *index_to_obj(struct kmem_cache *cache, struct page *page,
396
397
				 unsigned int idx)
{
398
	return page->s_mem + cache->size * idx;
399
400
}

401
/*
402
403
404
 * 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
405
406
407
 *   reciprocal_divide(offset, cache->reciprocal_buffer_size)
 */
static inline unsigned int obj_to_index(const struct kmem_cache *cache,
408
					const struct page *page, void *obj)
409
{
410
	u32 offset = (obj - page->s_mem);
411
	return reciprocal_divide(offset, cache->reciprocal_buffer_size);
412
413
}

414
#define BOOT_CPUCACHE_ENTRIES	1
Linus Torvalds's avatar
Linus Torvalds committed
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
static DEFINE_PER_CPU(struct delayed_work, slab_reap_work);
Linus Torvalds's avatar
Linus Torvalds committed
427

428
static inline struct array_cache *cpu_cache_get(struct kmem_cache *cachep)
Linus Torvalds's avatar
Linus Torvalds committed
429
{
430
	return this_cpu_ptr(cachep->cpu_cache);
Linus Torvalds's avatar
Linus Torvalds committed
431
432
}

Andrew Morton's avatar
Andrew Morton committed
433
434
435
/*
 * Calculate the number of objects and left-over bytes for a given buffer size.
 */
436
437
static unsigned int cache_estimate(unsigned long gfporder, size_t buffer_size,
		unsigned long flags, size_t *left_over)
438
{
439
	unsigned int num;
440
	size_t slab_size = PAGE_SIZE << gfporder;
Linus Torvalds's avatar
Linus Torvalds committed
441

442
443
444
445
446
447
	/*
	 * 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:
	 *
	 * - @buffer_size bytes for each object
448
449
450
451
452
	 * - One freelist_idx_t for each object
	 *
	 * We don't need to consider alignment of freelist because
	 * freelist will be at the end of slab page. The objects will be
	 * at the correct alignment.
453
454
455
456
457
458
	 *
	 * 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.
	 */
459
	if (flags & (CFLGS_OBJFREELIST_SLAB | CFLGS_OFF_SLAB)) {
460
		num = slab_size / buffer_size;
461
		*left_over = slab_size % buffer_size;
462
	} else {
463
		num = slab_size / (buffer_size + sizeof(freelist_idx_t));
464
465
		*left_over = slab_size %
			(buffer_size + sizeof(freelist_idx_t));
466
	}
467
468

	return num;
Linus Torvalds's avatar
Linus Torvalds committed
469
470
}

471
#if DEBUG
472
#define slab_error(cachep, msg) __slab_error(__func__, cachep, msg)
Linus Torvalds's avatar
Linus Torvalds committed
473

Andrew Morton's avatar
Andrew Morton committed
474
475
static void __slab_error(const char *function, struct kmem_cache *cachep,
			char *msg)
Linus Torvalds's avatar
Linus Torvalds committed
476
{
477
	pr_err("slab error in %s(): cache `%s': %s\n",
478
	       function, cachep->name, msg);
Linus Torvalds's avatar
Linus Torvalds committed
479
	dump_stack();
480
	add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
Linus Torvalds's avatar
Linus Torvalds committed
481
}
482
#endif
Linus Torvalds's avatar
Linus Torvalds committed
483

484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
/*
 * 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);

500
501
502
503
504
505
506
507
508
509
510
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);

511
512
513
514
515
516
517
#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.
 */
518
static DEFINE_PER_CPU(unsigned long, slab_reap_node);
519
520
521
522
523

static void init_reap_node(int cpu)
{
	int node;

524
	node = next_node(cpu_to_mem(cpu), node_online_map);
525
	if (node == MAX_NUMNODES)
526
		node = first_node(node_online_map);
527

528
	per_cpu(slab_reap_node, cpu) = node;
529
530
531
532
}

static void next_reap_node(void)
{
533
	int node = __this_cpu_read(slab_reap_node);
534
535
536
537

	node = next_node(node, node_online_map);
	if (unlikely(node >= MAX_NUMNODES))
		node = first_node(node_online_map);
538
	__this_cpu_write(slab_reap_node, node);
539
540
541
542
543
544
545
}

#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
546
547
548
549
550
551
552
/*
 * 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.
 */
553
static void start_cpu_timer(int cpu)
Linus Torvalds's avatar
Linus Torvalds committed
554
{
555
	struct delayed_work *reap_work = &per_cpu(slab_reap_work, cpu);
Linus Torvalds's avatar
Linus Torvalds committed
556
557
558
559
560
561

	/*
	 * When this gets called from do_initcalls via cpucache_init(),
	 * init_workqueues() has already run, so keventd will be setup
	 * at that time.
	 */
562
	if (keventd_up() && reap_work->work.func == NULL) {
563
		init_reap_node(cpu);
564
		INIT_DEFERRABLE_WORK(reap_work, cache_reap);
565
566
		schedule_delayed_work_on(cpu, reap_work,
					__round_jiffies_relative(HZ, cpu));
Linus Torvalds's avatar
Linus Torvalds committed
567
568
569
	}
}

570
static void init_arraycache(struct array_cache *ac, int limit, int batch)
Linus Torvalds's avatar
Linus Torvalds committed
571
{
572
573
	/*
	 * The array_cache structures contain pointers to free object.
Lucas De Marchi's avatar
Lucas De Marchi committed
574
	 * However, when such objects are allocated or transferred to another
575
576
577
578
	 * 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.
	 */
579
580
581
582
583
584
	kmemleak_no_scan(ac);
	if (ac) {
		ac->avail = 0;
		ac->limit = limit;
		ac->batchcount = batch;
		ac->touched = 0;
Linus Torvalds's avatar
Linus Torvalds committed
585
	}
586
587
588
589
590
}

static struct array_cache *alloc_arraycache(int node, int entries,
					    int batchcount, gfp_t gfp)
{
591
	size_t memsize = sizeof(void *) * entries + sizeof(struct array_cache);
592
593
594
595
596
	struct array_cache *ac = NULL;

	ac = kmalloc_node(memsize, gfp, node);
	init_arraycache(ac, entries, batchcount);
	return ac;
Linus Torvalds's avatar
Linus Torvalds committed
597
598
}

599
600
static noinline void cache_free_pfmemalloc(struct kmem_cache *cachep,
					struct page *page, void *objp)
601
{
602
603
604
	struct kmem_cache_node *n;
	int page_node;
	LIST_HEAD(list);
605

606
607
	page_node = page_to_nid(page);
	n = get_node(cachep, page_node);
608

609
610
611
	spin_lock(&n->list_lock);
	free_block(cachep, &objp, 1, page_node, &list);
	spin_unlock(&n->list_lock);
612

613
	slabs_destroy(cachep, &list);
614
615
}

616
617
618
619
620
621
622
623
624
625
/*
 * 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 */
626
	int nr = min3(from->avail, max, to->limit - to->avail);
627
628
629
630
631
632
633
634
635
636
637
638

	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;
}

639
640
641
#ifndef CONFIG_NUMA

#define drain_alien_cache(cachep, alien) do { } while (0)
642
#define reap_alien(cachep, n) do { } while (0)
643

Joonsoo Kim's avatar
Joonsoo Kim committed
644
645
static inline struct alien_cache **alloc_alien_cache(int node,
						int limit, gfp_t gfp)
646
{
647
	return (struct alien_cache **)BAD_ALIEN_MAGIC;
648
649
}

Joonsoo Kim's avatar
Joonsoo Kim committed
650
static inline void free_alien_cache(struct alien_cache **ac_ptr)
651
652
653
654
655
656
657
658
659
660
661
662
663
664
{
}

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;
}

665
static inline void *____cache_alloc_node(struct kmem_cache *cachep,
666
667
668
669
670
		 gfp_t flags, int nodeid)
{
	return NULL;
}

David Rientjes's avatar
David Rientjes committed
671
672
static inline gfp_t gfp_exact_node(gfp_t flags)
{
673
	return flags & ~__GFP_NOFAIL;
David Rientjes's avatar
David Rientjes committed
674
675
}

676
677
#else	/* CONFIG_NUMA */

678
static void *____cache_alloc_node(struct kmem_cache *, gfp_t, int);
679
static void *alternate_node_alloc(struct kmem_cache *, gfp_t);
680

Joonsoo Kim's avatar
Joonsoo Kim committed
681
682
683
static struct alien_cache *__alloc_alien_cache(int node, int entries,
						int batch, gfp_t gfp)
{
684
	size_t memsize = sizeof(void *) * entries + sizeof(struct alien_cache);
Joonsoo Kim's avatar
Joonsoo Kim committed
685
686
687
688
	struct alien_cache *alc = NULL;

	alc = kmalloc_node(memsize, gfp, node);
	init_arraycache(&alc->ac, entries, batch);
689
	spin_lock_init(&alc->lock);
Joonsoo Kim's avatar
Joonsoo Kim committed
690
691
692
693
	return alc;
}

static struct alien_cache **alloc_alien_cache(int node, int limit, gfp_t gfp)
694
{
Joonsoo Kim's avatar
Joonsoo Kim committed
695
	struct alien_cache **alc_ptr;
696
	size_t memsize = sizeof(void *) * nr_node_ids;
697
698
699
700
	int i;

	if (limit > 1)
		limit = 12;
Joonsoo Kim's avatar
Joonsoo Kim committed
701
702
703
704
705
706
707
708
709
710
711
712
713
	alc_ptr = kzalloc_node(memsize, gfp, node);
	if (!alc_ptr)
		return NULL;

	for_each_node(i) {
		if (i == node || !node_online(i))
			continue;
		alc_ptr[i] = __alloc_alien_cache(node, limit, 0xbaadf00d, gfp);
		if (!alc_ptr[i]) {
			for (i--; i >= 0; i--)
				kfree(alc_ptr[i]);
			kfree(alc_ptr);
			return NULL;
714
715
		}
	}
Joonsoo Kim's avatar
Joonsoo Kim committed
716
	return alc_ptr;
717
718
}

Joonsoo Kim's avatar
Joonsoo Kim committed
719
static void free_alien_cache(struct alien_cache **alc_ptr)
720
721
722
{
	int i;

Joonsoo Kim's avatar
Joonsoo Kim committed
723
	if (!alc_ptr)
724
725
		return;
	for_each_node(i)
Joonsoo Kim's avatar
Joonsoo Kim committed
726
727
	    kfree(alc_ptr[i]);
	kfree(alc_ptr);
728
729
}

730
static void __drain_alien_cache(struct kmem_cache *cachep,
731
732
				struct array_cache *ac, int node,
				struct list_head *list)
733
{
734
	struct kmem_cache_node *n = get_node(cachep, node);
735
736

	if (ac->avail) {
737
		spin_lock(&n->list_lock);
738
739
740
741
742
		/*
		 * 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.
		 */
743
744
		if (n->shared)
			transfer_objects(n->shared, ac, ac->limit);
745

746
		free_block(cachep, ac->entry, ac->avail, node, list);
747
		ac->avail = 0;
748
		spin_unlock(&n->list_lock);
749
750
751
	}
}

752
753
754
/*
 * Called from cache_reap() to regularly drain alien caches round robin.
 */
755
static void reap_alien(struct kmem_cache *cachep, struct kmem_cache_node *n)
756
{
757
	int node = __this_cpu_read(slab_reap_node);
758

759
	if (n->alien) {
Joonsoo Kim's avatar
Joonsoo Kim committed
760
761
762
763
764
		struct alien_cache *alc = n->alien[node];
		struct array_cache *ac;

		if (alc) {
			ac = &alc->ac;
765
			if (ac->avail && spin_trylock_irq(&alc->lock)) {
766
767
768
				LIST_HEAD(list);

				__drain_alien_cache(cachep, ac, node, &list);
769
				spin_unlock_irq(&alc->lock);
770
				slabs_destroy(cachep, &list);
Joonsoo Kim's avatar
Joonsoo Kim committed
771
			}
772
773
774
775
		}
	}
}

Andrew Morton's avatar
Andrew Morton committed
776
static void drain_alien_cache(struct kmem_cache *cachep,
Joonsoo Kim's avatar
Joonsoo Kim committed
777
				struct alien_cache **alien)
778
{
779
	int i = 0;
Joonsoo Kim's avatar
Joonsoo Kim committed
780
	struct alien_cache *alc;
781
782
783
784
	struct array_cache *ac;
	unsigned long flags;

	for_each_online_node(i) {
Joonsoo Kim's avatar
Joonsoo Kim committed
785
786
		alc = alien[i];
		if (alc) {
787
788
			LIST_HEAD(list);

Joonsoo Kim's avatar
Joonsoo Kim committed
789
			ac = &alc->ac;
790
			spin_lock_irqsave(&alc->lock, flags);
791
			__drain_alien_cache(cachep, ac, i, &list);
792
			spin_unlock_irqrestore(&alc->lock, flags);
793
			slabs_destroy(cachep, &list);
794
795
796
		}
	}
}
797

798
799
static int __cache_free_alien(struct kmem_cache *cachep, void *objp,
				int node, int page_node)
800
{
801
	struct kmem_cache_node *n;
Joonsoo Kim's avatar
Joonsoo Kim committed
802
803
	struct alien_cache *alien = NULL;
	struct array_cache *ac;
804
	LIST_HEAD(list);
805

806
	n = get_node(cachep, node);
807
	STATS_INC_NODEFREES(cachep);
808
809
	if (n->alien && n->alien[page_node]) {
		alien = n->alien[page_node];
Joonsoo Kim's avatar
Joonsoo Kim committed
810
		ac = &alien->ac;
811
		spin_lock(&alien->lock);
Joonsoo Kim's avatar
Joonsoo Kim committed
812
		if (unlikely(ac->avail == ac->limit)) {
813
			STATS_INC_ACOVERFLOW(cachep);
814
			__drain_alien_cache(cachep, ac, page_node, &list);
815
		}
816
		ac->entry[ac->avail++] = objp;
817
		spin_unlock(&alien->lock);
818
		slabs_destroy(cachep, &list);
819
	} else {
820
		n = get_node(cachep, page_node);
821
		spin_lock(&n->list_lock);
822
		free_block(cachep, &objp, 1, page_node, &list);
823
		spin_unlock(&n->list_lock);
824
		slabs_destroy(cachep, &list);
825
826
827
	}
	return 1;
}
828
829
830
831
832
833
834
835
836
837
838
839
840
841

static inline int cache_free_alien(struct kmem_cache *cachep, void *objp)
{
	int page_node = page_to_nid(virt_to_page(objp));
	int node = numa_mem_id();
	/*
	 * Make sure we are not freeing a object from another node to the array
	 * cache on this cpu.
	 */
	if (likely(node == page_node))
		return 0;

	return __cache_free_alien(cachep, objp, node, page_node);
}
David Rientjes's avatar
David Rientjes committed
842
843

/*
844
845
 * Construct gfp mask to allocate from a specific node but do not reclaim or
 * warn about failures.
David Rientjes's avatar
David Rientjes committed
846
847
848
 */
static inline gfp_t gfp_exact_node(gfp_t flags)
{
849
	return (flags | __GFP_THISNODE | __GFP_NOWARN) & ~(__GFP_RECLAIM|__GFP_NOFAIL);
David Rientjes's avatar
David Rientjes committed
850
}
851
852
#endif

853
/*
854
 * Allocates and initializes node for a node on each slab cache, used for
855
 * either memory or cpu hotplug.  If memory is being hot-added, the kmem_cache_node
856
 * will be allocated off-node since memory is not yet online for the new node.
857
 * When hotplugging memory or a cpu, existing node are not replaced if
858
859
 * already in use.
 *
860
 * Must hold slab_mutex.
861
 */
862
static int init_cache_node_node(int node)
863
864
{
	struct kmem_cache *cachep;
865
	struct kmem_cache_node *n;
866
	const size_t memsize = sizeof(struct kmem_cache_node);
867

868
	list_for_each_entry(cachep, &slab_caches, list) {
869
		/*
870
		 * Set up the kmem_cache_node for cpu before we can
871
872
873
		 * begin anything. Make sure some other cpu on this
		 * node has not already allocated this
		 */
874
875
		n = get_node(cachep, node);
		if (!n) {
876
877
			n = kmalloc_node(memsize, GFP_KERNEL, node);
			if (!n)
878
				return -ENOMEM;
879
			kmem_cache_node_init(n);
880
881
			n->next_reap = jiffies + REAPTIMEOUT_NODE +
			    ((unsigned long)cachep) % REAPTIMEOUT_NODE;
882
883

			/*
884
885
			 * The kmem_cache_nodes don't come and go as CPUs
			 * come and go.  slab_mutex is sufficient
886
887
			 * protection here.
			 */
888
			cachep->node[node] = n;
889
890
		}

891
892
		spin_lock_irq(&n->list_lock);
		n->free_limit =
893
894
			(1 + nr_cpus_node(node)) *
			cachep->batchcount + cachep->num;
895
		spin_unlock_irq(&n->list_lock);
896
897
898
899
	}
	return 0;
}

900
901
902
903
904
905
static inline int slabs_tofree(struct kmem_cache *cachep,
						struct kmem_cache_node *n)
{
	return (n->free_objects + cachep->num - 1) / cachep->num;
}

906
static void cpuup_canceled(long cpu)
907
908
{
	struct kmem_cache *cachep;
909
	struct kmem_cache_node *n = NULL;
910
	int node = cpu_to_mem(cpu);
911
	const struct cpumask *mask = cpumask_of_node(node);
912

913
	list_for_each_entry(cachep, &slab_caches, list) {
914
915
		struct array_cache *nc;
		struct array_cache *shared;
Joonsoo Kim's avatar
Joonsoo Kim committed
916
		struct alien_cache **alien;
917
		LIST_HEAD(list);
918

919
		n = get_node(cachep, node);
920
		if (!n)
921
			continue;
922

923
		spin_lock_irq(&n->list_lock);
924

925
926
		/* Free limit for this kmem_cache_node */
		n->free_limit -= cachep->batchcount;
927
928
929
930

		/* cpu is dead; no one can alloc from it. */
		nc = per_cpu_ptr(cachep->cpu_cache, cpu);
		if (nc) {
931
			free_block(cachep, nc->entry, nc->avail, node, &list);
932
933
			nc->avail = 0;
		}
934

935
		if (!cpumask_empty(mask)) {
936
			spin_unlock_irq(&n->list_lock);
937
			goto free_slab;
938
939
		}

940
		shared = n->shared;
941
942
		if (shared) {
			free_block(cachep, shared->entry,
943
				   shared->avail, node, &list);
944
			n->shared = NULL;
945
946
		}

947
948
		alien = n->alien;
		n->alien = NULL;
949

950
		spin_unlock_irq(&n->list_lock);
951
952
953
954
955
956

		kfree(shared);
		if (alien) {
			drain_alien_cache(cachep, alien);
			free_alien_cache(alien);
		}
957
958

free_slab:
959
		slabs_destroy(cachep, &list);
960
961
962
963
964
965
	}
	/*
	 * In the previous loop, all the objects were freed to
	 * the respective cache's slabs,  now we can go ahead and
	 * shrink each nodelist to its limit.
	 */
966
	list_for_each_entry(cachep, &slab_caches, list) {
967
		n = get_node(cachep, node);
968
		if (!n)
969
			continue;
970
		drain_freelist(cachep, n, slabs_tofree(cachep, n));
971
972
973
	}
}

974
static int cpuup_prepare(long cpu)
Linus Torvalds's avatar
Linus Torvalds committed
975
{
976
	struct kmem_cache *cachep;
977
	struct kmem_cache_node *n = NULL;
978
	int node = cpu_to_mem(cpu);
979
	int err;
Linus Torvalds's avatar
Linus Torvalds committed
980

981
982
983
984
	/*
	 * We need to do this right in the beginning since
	 * alloc_arraycache's are going to use this list.
	 * kmalloc_node allows us to add the slab to the right
985
	 * kmem_cache_node and not this cpu's kmem_cache_node
986
	 */
987
	err = init_cache_node_node(node);
988
989
	if (err < 0)
		goto bad;
990
991
992
993
994

	/*
	 * Now we can go ahead with allocating the shared arrays and
	 * array caches
	 */
995
	list_for_each_entry(cachep, &slab_caches, list) {
996
		struct array_cache *shared = NULL;
Joonsoo Kim's avatar
Joonsoo Kim committed
997
		struct alien_cache **alien = NULL;
998
999
1000
1001

		if (cachep->shared) {
			shared = alloc_arraycache(node,
				cachep->shared * cachep->batchcount,
1002
				0xbaadf00d, GFP_KERNEL);
1003
			if (!shared)
Linus Torvalds's avatar
Linus Torvalds committed
1004
				goto bad;