slab.c 111 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
218
219
220
static inline void fixup_objfreelist_debug(struct kmem_cache *cachep,
						void **list);
static inline void fixup_slab_list(struct kmem_cache *cachep,
				struct kmem_cache_node *n, struct page *page,
				void **list);
221
222
static int slab_early_init = 1;

223
#define INDEX_NODE kmalloc_index(sizeof(struct kmem_cache_node))
Linus Torvalds's avatar
Linus Torvalds committed
224

225
static void kmem_cache_node_init(struct kmem_cache_node *parent)
226
227
228
229
230
231
{
	INIT_LIST_HEAD(&parent->slabs_full);
	INIT_LIST_HEAD(&parent->slabs_partial);
	INIT_LIST_HEAD(&parent->slabs_free);
	parent->shared = NULL;
	parent->alien = NULL;
232
	parent->colour_next = 0;
233
234
235
236
237
	spin_lock_init(&parent->list_lock);
	parent->free_objects = 0;
	parent->free_touched = 0;
}

Andrew Morton's avatar
Andrew Morton committed
238
239
240
#define MAKE_LIST(cachep, listp, slab, nodeid)				\
	do {								\
		INIT_LIST_HEAD(listp);					\
241
		list_splice(&get_node(cachep, nodeid)->slab, listp);	\
242
243
	} while (0)

Andrew Morton's avatar
Andrew Morton committed
244
245
#define	MAKE_ALL_LISTS(cachep, ptr, nodeid)				\
	do {								\
246
247
248
249
	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
250

251
#define CFLGS_OBJFREELIST_SLAB	(0x40000000UL)
Linus Torvalds's avatar
Linus Torvalds committed
252
#define CFLGS_OFF_SLAB		(0x80000000UL)
253
#define	OBJFREELIST_SLAB(x)	((x)->flags & CFLGS_OBJFREELIST_SLAB)
Linus Torvalds's avatar
Linus Torvalds committed
254
255
256
#define	OFF_SLAB(x)	((x)->flags & CFLGS_OFF_SLAB)

#define BATCHREFILL_LIMIT	16
Andrew Morton's avatar
Andrew Morton committed
257
258
259
/*
 * Optimization question: fewer reaps means less probability for unnessary
 * cpucache drain/refill cycles.
Linus Torvalds's avatar
Linus Torvalds committed
260
 *
Adrian Bunk's avatar
Adrian Bunk committed
261
 * OTOH the cpuarrays can contain lots of objects,
Linus Torvalds's avatar
Linus Torvalds committed
262
263
 * which could lock up otherwise freeable slabs.
 */
264
265
#define REAPTIMEOUT_AC		(2*HZ)
#define REAPTIMEOUT_NODE	(4*HZ)
Linus Torvalds's avatar
Linus Torvalds committed
266
267
268
269
270
271

#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++)
272
#define	STATS_ADD_REAPED(x,y)	((x)->reaped += (y))
Andrew Morton's avatar
Andrew Morton committed
273
274
275
276
277
#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
278
279
#define	STATS_INC_ERR(x)	((x)->errors++)
#define	STATS_INC_NODEALLOCS(x)	((x)->node_allocs++)
280
#define	STATS_INC_NODEFREES(x)	((x)->node_frees++)
281
#define STATS_INC_ACOVERFLOW(x)   ((x)->node_overflow++)
Andrew Morton's avatar
Andrew Morton committed
282
283
284
285
286
#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
287
288
289
290
291
292
293
294
295
#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)
296
#define	STATS_ADD_REAPED(x,y)	do { (void)(y); } while (0)
Linus Torvalds's avatar
Linus Torvalds committed
297
298
299
#define	STATS_SET_HIGH(x)	do { } while (0)
#define	STATS_INC_ERR(x)	do { } while (0)
#define	STATS_INC_NODEALLOCS(x)	do { } while (0)
300
#define	STATS_INC_NODEFREES(x)	do { } while (0)
301
#define STATS_INC_ACOVERFLOW(x)   do { } while (0)
Andrew Morton's avatar
Andrew Morton committed
302
#define	STATS_SET_FREEABLE(x, i) do { } while (0)
Linus Torvalds's avatar
Linus Torvalds committed
303
304
305
306
307
308
309
310
#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
311
312
/*
 * memory layout of objects:
Linus Torvalds's avatar
Linus Torvalds committed
313
 * 0		: objp
314
 * 0 .. cachep->obj_offset - BYTES_PER_WORD - 1: padding. This ensures that
Linus Torvalds's avatar
Linus Torvalds committed
315
316
 * 		the end of an object is aligned with the end of the real
 * 		allocation. Catches writes behind the end of the allocation.
317
 * cachep->obj_offset - BYTES_PER_WORD .. cachep->obj_offset - 1:
Linus Torvalds's avatar
Linus Torvalds committed
318
 * 		redzone word.
319
 * cachep->obj_offset: The real object.
320
321
 * 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
322
 *					[BYTES_PER_WORD long]
Linus Torvalds's avatar
Linus Torvalds committed
323
 */
324
static int obj_offset(struct kmem_cache *cachep)
Linus Torvalds's avatar
Linus Torvalds committed
325
{
326
	return cachep->obj_offset;
Linus Torvalds's avatar
Linus Torvalds committed
327
328
}

329
static unsigned long long *dbg_redzone1(struct kmem_cache *cachep, void *objp)
Linus Torvalds's avatar
Linus Torvalds committed
330
331
{
	BUG_ON(!(cachep->flags & SLAB_RED_ZONE));
332
333
	return (unsigned long long*) (objp + obj_offset(cachep) -
				      sizeof(unsigned long long));
Linus Torvalds's avatar
Linus Torvalds committed
334
335
}

336
static unsigned long long *dbg_redzone2(struct kmem_cache *cachep, void *objp)
Linus Torvalds's avatar
Linus Torvalds committed
337
338
339
{
	BUG_ON(!(cachep->flags & SLAB_RED_ZONE));
	if (cachep->flags & SLAB_STORE_USER)
340
		return (unsigned long long *)(objp + cachep->size -
341
					      sizeof(unsigned long long) -
David Woodhouse's avatar
David Woodhouse committed
342
					      REDZONE_ALIGN);
343
	return (unsigned long long *) (objp + cachep->size -
344
				       sizeof(unsigned long long));
Linus Torvalds's avatar
Linus Torvalds committed
345
346
}

347
static void **dbg_userword(struct kmem_cache *cachep, void *objp)
Linus Torvalds's avatar
Linus Torvalds committed
348
349
{
	BUG_ON(!(cachep->flags & SLAB_STORE_USER));
350
	return (void **)(objp + cachep->size - BYTES_PER_WORD);
Linus Torvalds's avatar
Linus Torvalds committed
351
352
353
354
}

#else

355
#define obj_offset(x)			0
356
357
#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
358
359
360
361
#define dbg_userword(cachep, objp)	({BUG(); (void **)NULL;})

#endif

362
363
#ifdef CONFIG_DEBUG_SLAB_LEAK

364
static inline bool is_store_user_clean(struct kmem_cache *cachep)
365
{
366
367
	return atomic_read(&cachep->store_user_clean) == 1;
}
368

369
370
371
372
static inline void set_store_user_clean(struct kmem_cache *cachep)
{
	atomic_set(&cachep->store_user_clean, 1);
}
373

374
375
376
377
static inline void set_store_user_dirty(struct kmem_cache *cachep)
{
	if (is_store_user_clean(cachep))
		atomic_set(&cachep->store_user_clean, 0);
378
379
380
}

#else
381
static inline void set_store_user_dirty(struct kmem_cache *cachep) {}
382
383
384

#endif

Linus Torvalds's avatar
Linus Torvalds committed
385
/*
386
387
 * 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
388
 */
389
390
391
#define	SLAB_MAX_ORDER_HI	1
#define	SLAB_MAX_ORDER_LO	0
static int slab_max_order = SLAB_MAX_ORDER_LO;
392
static bool slab_max_order_set __initdata;
Linus Torvalds's avatar
Linus Torvalds committed
393

394
395
static inline struct kmem_cache *virt_to_cache(const void *obj)
{
396
	struct page *page = virt_to_head_page(obj);
397
	return page->slab_cache;
398
399
}

400
static inline void *index_to_obj(struct kmem_cache *cache, struct page *page,
401
402
				 unsigned int idx)
{
403
	return page->s_mem + cache->size * idx;
404
405
}

406
/*
407
408
409
 * 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
410
411
412
 *   reciprocal_divide(offset, cache->reciprocal_buffer_size)
 */
static inline unsigned int obj_to_index(const struct kmem_cache *cache,
413
					const struct page *page, void *obj)
414
{
415
	u32 offset = (obj - page->s_mem);
416
	return reciprocal_divide(offset, cache->reciprocal_buffer_size);
417
418
}

419
#define BOOT_CPUCACHE_ENTRIES	1
Linus Torvalds's avatar
Linus Torvalds committed
420
/* internal cache of cache description objs */
421
static struct kmem_cache kmem_cache_boot = {
422
423
424
	.batchcount = 1,
	.limit = BOOT_CPUCACHE_ENTRIES,
	.shared = 1,
425
	.size = sizeof(struct kmem_cache),
426
	.name = "kmem_cache",
Linus Torvalds's avatar
Linus Torvalds committed
427
428
};

429
static DEFINE_PER_CPU(struct delayed_work, slab_reap_work);
Linus Torvalds's avatar
Linus Torvalds committed
430

431
static inline struct array_cache *cpu_cache_get(struct kmem_cache *cachep)
Linus Torvalds's avatar
Linus Torvalds committed
432
{
433
	return this_cpu_ptr(cachep->cpu_cache);
Linus Torvalds's avatar
Linus Torvalds committed
434
435
}

Andrew Morton's avatar
Andrew Morton committed
436
437
438
/*
 * Calculate the number of objects and left-over bytes for a given buffer size.
 */
439
440
static unsigned int cache_estimate(unsigned long gfporder, size_t buffer_size,
		unsigned long flags, size_t *left_over)
441
{
442
	unsigned int num;
443
	size_t slab_size = PAGE_SIZE << gfporder;
Linus Torvalds's avatar
Linus Torvalds committed
444

445
446
447
448
449
450
	/*
	 * 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
451
452
453
454
455
	 * - 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.
456
457
458
459
460
461
	 *
	 * 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.
	 */
462
	if (flags & (CFLGS_OBJFREELIST_SLAB | CFLGS_OFF_SLAB)) {
463
		num = slab_size / buffer_size;
464
		*left_over = slab_size % buffer_size;
465
	} else {
466
		num = slab_size / (buffer_size + sizeof(freelist_idx_t));
467
468
		*left_over = slab_size %
			(buffer_size + sizeof(freelist_idx_t));
469
	}
470
471

	return num;
Linus Torvalds's avatar
Linus Torvalds committed
472
473
}

474
#if DEBUG
475
#define slab_error(cachep, msg) __slab_error(__func__, cachep, msg)
Linus Torvalds's avatar
Linus Torvalds committed
476

Andrew Morton's avatar
Andrew Morton committed
477
478
static void __slab_error(const char *function, struct kmem_cache *cachep,
			char *msg)
Linus Torvalds's avatar
Linus Torvalds committed
479
{
480
	pr_err("slab error in %s(): cache `%s': %s\n",
481
	       function, cachep->name, msg);
Linus Torvalds's avatar
Linus Torvalds committed
482
	dump_stack();
483
	add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
Linus Torvalds's avatar
Linus Torvalds committed
484
}
485
#endif
Linus Torvalds's avatar
Linus Torvalds committed
486

487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
/*
 * 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);

503
504
505
506
507
508
509
510
511
512
513
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);

514
515
516
517
518
519
520
#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.
 */
521
static DEFINE_PER_CPU(unsigned long, slab_reap_node);
522
523
524

static void init_reap_node(int cpu)
{
525
526
	per_cpu(slab_reap_node, cpu) = next_node_in(cpu_to_mem(cpu),
						    node_online_map);
527
528
529
530
}

static void next_reap_node(void)
{
531
	int node = __this_cpu_read(slab_reap_node);
532

533
	node = next_node_in(node, node_online_map);
534
	__this_cpu_write(slab_reap_node, node);
535
536
537
538
539
540
541
}

#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
542
543
544
545
546
547
548
/*
 * 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.
 */
549
static void start_cpu_timer(int cpu)
Linus Torvalds's avatar
Linus Torvalds committed
550
{
551
	struct delayed_work *reap_work = &per_cpu(slab_reap_work, cpu);
Linus Torvalds's avatar
Linus Torvalds committed
552
553
554
555
556
557

	/*
	 * When this gets called from do_initcalls via cpucache_init(),
	 * init_workqueues() has already run, so keventd will be setup
	 * at that time.
	 */
558
	if (keventd_up() && reap_work->work.func == NULL) {
559
		init_reap_node(cpu);
560
		INIT_DEFERRABLE_WORK(reap_work, cache_reap);
561
562
		schedule_delayed_work_on(cpu, reap_work,
					__round_jiffies_relative(HZ, cpu));
Linus Torvalds's avatar
Linus Torvalds committed
563
564
565
	}
}

566
static void init_arraycache(struct array_cache *ac, int limit, int batch)
Linus Torvalds's avatar
Linus Torvalds committed
567
{
568
569
	/*
	 * The array_cache structures contain pointers to free object.
Lucas De Marchi's avatar
Lucas De Marchi committed
570
	 * However, when such objects are allocated or transferred to another
571
572
573
574
	 * 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.
	 */
575
576
577
578
579
580
	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
581
	}
582
583
584
585
586
}

static struct array_cache *alloc_arraycache(int node, int entries,
					    int batchcount, gfp_t gfp)
{
587
	size_t memsize = sizeof(void *) * entries + sizeof(struct array_cache);
588
589
590
591
592
	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
593
594
}

595
596
static noinline void cache_free_pfmemalloc(struct kmem_cache *cachep,
					struct page *page, void *objp)
597
{
598
599
600
	struct kmem_cache_node *n;
	int page_node;
	LIST_HEAD(list);
601

602
603
	page_node = page_to_nid(page);
	n = get_node(cachep, page_node);
604

605
606
607
	spin_lock(&n->list_lock);
	free_block(cachep, &objp, 1, page_node, &list);
	spin_unlock(&n->list_lock);
608

609
	slabs_destroy(cachep, &list);
610
611
}

612
613
614
615
616
617
618
619
620
621
/*
 * 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 */
622
	int nr = min3(from->avail, max, to->limit - to->avail);
623
624
625
626
627
628
629
630
631
632
633
634

	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;
}

635
636
637
#ifndef CONFIG_NUMA

#define drain_alien_cache(cachep, alien) do { } while (0)
638
#define reap_alien(cachep, n) do { } while (0)
639

Joonsoo Kim's avatar
Joonsoo Kim committed
640
641
static inline struct alien_cache **alloc_alien_cache(int node,
						int limit, gfp_t gfp)
642
{
643
	return NULL;
644
645
}

Joonsoo Kim's avatar
Joonsoo Kim committed
646
static inline void free_alien_cache(struct alien_cache **ac_ptr)
647
648
649
650
651
652
653
654
655
656
657
658
659
660
{
}

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;
}

661
static inline void *____cache_alloc_node(struct kmem_cache *cachep,
662
663
664
665
666
		 gfp_t flags, int nodeid)
{
	return NULL;
}

David Rientjes's avatar
David Rientjes committed
667
668
static inline gfp_t gfp_exact_node(gfp_t flags)
{
669
	return flags & ~__GFP_NOFAIL;
David Rientjes's avatar
David Rientjes committed
670
671
}

672
673
#else	/* CONFIG_NUMA */

674
static void *____cache_alloc_node(struct kmem_cache *, gfp_t, int);
675
static void *alternate_node_alloc(struct kmem_cache *, gfp_t);
676

Joonsoo Kim's avatar
Joonsoo Kim committed
677
678
679
static struct alien_cache *__alloc_alien_cache(int node, int entries,
						int batch, gfp_t gfp)
{
680
	size_t memsize = sizeof(void *) * entries + sizeof(struct alien_cache);
Joonsoo Kim's avatar
Joonsoo Kim committed
681
682
683
684
	struct alien_cache *alc = NULL;

	alc = kmalloc_node(memsize, gfp, node);
	init_arraycache(&alc->ac, entries, batch);
685
	spin_lock_init(&alc->lock);
Joonsoo Kim's avatar
Joonsoo Kim committed
686
687
688
689
	return alc;
}

static struct alien_cache **alloc_alien_cache(int node, int limit, gfp_t gfp)
690
{
Joonsoo Kim's avatar
Joonsoo Kim committed
691
	struct alien_cache **alc_ptr;
692
	size_t memsize = sizeof(void *) * nr_node_ids;
693
694
695
696
	int i;

	if (limit > 1)
		limit = 12;
Joonsoo Kim's avatar
Joonsoo Kim committed
697
698
699
700
701
702
703
704
705
706
707
708
709
	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;
710
711
		}
	}
Joonsoo Kim's avatar
Joonsoo Kim committed
712
	return alc_ptr;
713
714
}

Joonsoo Kim's avatar
Joonsoo Kim committed
715
static void free_alien_cache(struct alien_cache **alc_ptr)
716
717
718
{
	int i;

Joonsoo Kim's avatar
Joonsoo Kim committed
719
	if (!alc_ptr)
720
721
		return;
	for_each_node(i)
Joonsoo Kim's avatar
Joonsoo Kim committed
722
723
	    kfree(alc_ptr[i]);
	kfree(alc_ptr);
724
725
}

726
static void __drain_alien_cache(struct kmem_cache *cachep,
727
728
				struct array_cache *ac, int node,
				struct list_head *list)
729
{
730
	struct kmem_cache_node *n = get_node(cachep, node);
731
732

	if (ac->avail) {
733
		spin_lock(&n->list_lock);
734
735
736
737
738
		/*
		 * 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.
		 */
739
740
		if (n->shared)
			transfer_objects(n->shared, ac, ac->limit);
741

742
		free_block(cachep, ac->entry, ac->avail, node, list);
743
		ac->avail = 0;
744
		spin_unlock(&n->list_lock);
745
746
747
	}
}

748
749
750
/*
 * Called from cache_reap() to regularly drain alien caches round robin.
 */
751
static void reap_alien(struct kmem_cache *cachep, struct kmem_cache_node *n)
752
{
753
	int node = __this_cpu_read(slab_reap_node);
754

755
	if (n->alien) {
Joonsoo Kim's avatar
Joonsoo Kim committed
756
757
758
759
760
		struct alien_cache *alc = n->alien[node];
		struct array_cache *ac;

		if (alc) {
			ac = &alc->ac;
761
			if (ac->avail && spin_trylock_irq(&alc->lock)) {
762
763
764
				LIST_HEAD(list);

				__drain_alien_cache(cachep, ac, node, &list);
765
				spin_unlock_irq(&alc->lock);
766
				slabs_destroy(cachep, &list);
Joonsoo Kim's avatar
Joonsoo Kim committed
767
			}
768
769
770
771
		}
	}
}

Andrew Morton's avatar
Andrew Morton committed
772
static void drain_alien_cache(struct kmem_cache *cachep,
Joonsoo Kim's avatar
Joonsoo Kim committed
773
				struct alien_cache **alien)
774
{
775
	int i = 0;
Joonsoo Kim's avatar
Joonsoo Kim committed
776
	struct alien_cache *alc;
777
778
779
780
	struct array_cache *ac;
	unsigned long flags;

	for_each_online_node(i) {
Joonsoo Kim's avatar
Joonsoo Kim committed
781
782
		alc = alien[i];
		if (alc) {
783
784
			LIST_HEAD(list);

Joonsoo Kim's avatar
Joonsoo Kim committed
785
			ac = &alc->ac;
786
			spin_lock_irqsave(&alc->lock, flags);
787
			__drain_alien_cache(cachep, ac, i, &list);
788
			spin_unlock_irqrestore(&alc->lock, flags);
789
			slabs_destroy(cachep, &list);
790
791
792
		}
	}
}
793

794
795
static int __cache_free_alien(struct kmem_cache *cachep, void *objp,
				int node, int page_node)
796
{
797
	struct kmem_cache_node *n;
Joonsoo Kim's avatar
Joonsoo Kim committed
798
799
	struct alien_cache *alien = NULL;
	struct array_cache *ac;
800
	LIST_HEAD(list);
801

802
	n = get_node(cachep, node);
803
	STATS_INC_NODEFREES(cachep);
804
805
	if (n->alien && n->alien[page_node]) {
		alien = n->alien[page_node];
Joonsoo Kim's avatar
Joonsoo Kim committed
806
		ac = &alien->ac;
807
		spin_lock(&alien->lock);
Joonsoo Kim's avatar
Joonsoo Kim committed
808
		if (unlikely(ac->avail == ac->limit)) {
809
			STATS_INC_ACOVERFLOW(cachep);
810
			__drain_alien_cache(cachep, ac, page_node, &list);
811
		}
812
		ac->entry[ac->avail++] = objp;
813
		spin_unlock(&alien->lock);
814
		slabs_destroy(cachep, &list);
815
	} else {
816
		n = get_node(cachep, page_node);
817
		spin_lock(&n->list_lock);
818
		free_block(cachep, &objp, 1, page_node, &list);
819
		spin_unlock(&n->list_lock);
820
		slabs_destroy(cachep, &list);
821
822
823
	}
	return 1;
}
824
825
826
827
828
829
830
831
832
833
834
835
836
837

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
838
839

/*
840
841
 * Construct gfp mask to allocate from a specific node but do not reclaim or
 * warn about failures.
David Rientjes's avatar
David Rientjes committed
842
843
844
 */
static inline gfp_t gfp_exact_node(gfp_t flags)
{
845
	return (flags | __GFP_THISNODE | __GFP_NOWARN) & ~(__GFP_RECLAIM|__GFP_NOFAIL);
David Rientjes's avatar
David Rientjes committed
846
}
847
848
#endif

849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
static int init_cache_node(struct kmem_cache *cachep, int node, gfp_t gfp)
{
	struct kmem_cache_node *n;

	/*
	 * Set up the kmem_cache_node for cpu before we can
	 * begin anything. Make sure some other cpu on this
	 * node has not already allocated this
	 */
	n = get_node(cachep, node);
	if (n) {
		spin_lock_irq(&n->list_lock);
		n->free_limit = (1 + nr_cpus_node(node)) * cachep->batchcount +
				cachep->num;
		spin_unlock_irq(&n->list_lock);

		return 0;
	}

	n = kmalloc_node(sizeof(struct kmem_cache_node), gfp, node);
	if (!n)
		return -ENOMEM;

	kmem_cache_node_init(n);
	n->next_reap = jiffies + REAPTIMEOUT_NODE +
		    ((unsigned long)cachep) % REAPTIMEOUT_NODE;

	n->free_limit =
		(1 + nr_cpus_node(node)) * cachep->batchcount + cachep->num;

	/*
	 * The kmem_cache_nodes don't come and go as CPUs
	 * come and go.  slab_mutex is sufficient
	 * protection here.
	 */
	cachep->node[node] = n;

	return 0;
}

889
/*
890
 * Allocates and initializes node for a node on each slab cache, used for
891
 * either memory or cpu hotplug.  If memory is being hot-added, the kmem_cache_node
892
 * will be allocated off-node since memory is not yet online for the new node.
893
 * When hotplugging memory or a cpu, existing node are not replaced if
894
895
 * already in use.
 *
896
 * Must hold slab_mutex.
897
 */
898
static int init_cache_node_node(int node)
899
{
900
	int ret;
901
902
	struct kmem_cache *cachep;

903
	list_for_each_entry(cachep, &slab_caches, list) {
904
905
906
		ret = init_cache_node(cachep, node, GFP_KERNEL);
		if (ret)
			return ret;
907
	}
908

909
910
911
	return 0;
}

912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
static int setup_kmem_cache_node(struct kmem_cache *cachep,
				int node, gfp_t gfp, bool force_change)
{
	int ret = -ENOMEM;
	struct kmem_cache_node *n;
	struct array_cache *old_shared = NULL;
	struct array_cache *new_shared = NULL;
	struct alien_cache **new_alien = NULL;
	LIST_HEAD(list);

	if (use_alien_caches) {
		new_alien = alloc_alien_cache(node, cachep->limit, gfp);
		if (!new_alien)
			goto fail;
	}

	if (cachep->shared) {
		new_shared = alloc_arraycache(node,
			cachep->shared * cachep->batchcount, 0xbaadf00d, gfp);
		if (!new_shared)
			goto fail;
	}

	ret = init_cache_node(cachep, node, gfp);
	if (ret)
		goto fail;

	n = get_node(cachep, node);
	spin_lock_irq(&n->list_lock);
	if (n->shared && force_change) {
		free_block(cachep, n->shared->entry,
				n->shared->avail, node, &list);
		n->shared->avail = 0;
	}

	if (!n->shared || force_change) {
		old_shared = n->shared;
		n->shared = new_shared;
		new_shared = NULL;
	}

	if (!n->alien) {
		n->alien = new_alien;
		new_alien = NULL;
	}

	spin_unlock_irq(&n->list_lock);
	slabs_destroy(cachep, &list);

961
962
963
964
965
966
967
968
969
	/*
	 * To protect lockless access to n->shared during irq disabled context.
	 * If n->shared isn't NULL in irq disabled context, accessing to it is
	 * guaranteed to be valid until irq is re-enabled, because it will be
	 * freed after synchronize_sched().
	 */
	if (force_change)
		synchronize_sched();

970
971
972
973
974
975
976
977
fail:
	kfree(old_shared);
	kfree(new_shared);
	free_alien_cache(new_alien);

	return ret;
}

978
static void cpuup_canceled(long cpu)
979
980
{
	struct kmem_cache *cachep;
981
	struct kmem_cache_node *n = NULL;
982
	int node = cpu_to_mem(cpu);
983
	const struct cpumask *mask = cpumask_of_node(node);
984

985
	list_for_each_entry(cachep, &slab_caches, list) {
986
987
		struct array_cache *nc;
		struct array_cache *shared;
Joonsoo Kim's avatar
Joonsoo Kim committed
988
		struct alien_cache **alien;
989
		LIST_HEAD(list);
990

991
		n = get_node(cachep, node);
992
		if (!n)
993
			continue;
994

995
		spin_lock_irq(&n->list_lock);
996

997
998
		/* Free limit for this kmem_cache_node */
		n->free_limit -= cachep->batchcount;
999
1000
1001
1002

		/* cpu is dead; no one can alloc from it. */
		nc = per_cpu_ptr(cachep->cpu_cache, cpu);
		if (nc) {
1003
			free_block(cachep, nc->entry, nc->avail, node, &list);
1004
1005
			nc->avail = 0;
		}
1006

1007
		if (!cpumask_empty(mask)) {
1008
			spin_unlock_irq(&n->list_lock);
1009
			goto free_slab;
1010
1011
		}

1012
		shared = n->shared;
1013
1014
		if (shared) {
			free_block(cachep, shared->entry,
1015
				   shared->avail, node, &list);
1016
			n->shared = NULL;
1017
1018
		}

1019
1020
		alien = n->alien;
		n->alien = NULL;
1021

1022
		spin_unlock_irq(&n->list_lock);
1023
1024
1025
1026
1027
1028

		kfree(shared);
		if (alien) {
			drain_alien_cache(cachep, alien);
			free_alien_cache(alien);
		}
1029
1030

free_slab:
1031
		slabs_destroy(cachep, &list);
1032
1033
1034
1035
1036
1037
	}
	/*
	 * 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.
	 */
1038
	list_for_each_entry(cachep, &slab_caches, list) {
1039
		n = get_node(cachep, node);
1040
		if (!n)
1041
			continue;
1042
		drain_freelist(cachep, n, INT_MAX);
1043
1044
1045
	}
}

1046
static int cpuup_prepare(long cpu)
Linus Torvalds's avatar
Linus Torvalds committed
1047
{
1048
	struct kmem_cache *cachep;
1049
	int node = cpu_to_mem(cpu);
1050
	int err;
Linus Torvalds's avatar
Linus Torvalds committed
1051

1052
1053
1054
1055
	/*
	 * 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
1056
	 * kmem_cache_node and not this cpu's kmem_cache_node
1057
	 */
1058
	err = init_cache_node_node(node);
1059
1060
	if (err < 0)
		goto bad;
1061
1062
1063
1064
1065

	/*
	 * Now we can go ahead with allocating the shared arrays and
	 * array caches
	 */
1066
	list_for_each_entry(cachep, &slab_caches, list) {
1067
1068
1069
		err = setup_kmem_cache_node(cachep, node, GFP_KERNEL, false);
		if (err)
			goto bad;
1070
	}
1071

1072
1073
	return 0;
bad:
1074
	cpuup_canceled(cpu);