2 * Slab allocator functions that are independent of the allocator strategy
4 * (C) 2012 Christoph Lameter <cl@linux.com>
6 #include <linux/slab.h>
9 #include <linux/poison.h>
10 #include <linux/interrupt.h>
11 #include <linux/memory.h>
12 #include <linux/compiler.h>
13 #include <linux/module.h>
14 #include <linux/cpu.h>
15 #include <linux/uaccess.h>
16 #include <linux/seq_file.h>
17 #include <linux/proc_fs.h>
18 #include <asm/cacheflush.h>
19 #include <asm/tlbflush.h>
21 #include <linux/memcontrol.h>
22 #include <trace/events/kmem.h>
26 enum slab_state slab_state
;
27 LIST_HEAD(slab_caches
);
28 DEFINE_MUTEX(slab_mutex
);
29 struct kmem_cache
*kmem_cache
;
31 #ifdef CONFIG_DEBUG_VM
32 static int kmem_cache_sanity_check(const char *name
, size_t size
)
34 struct kmem_cache
*s
= NULL
;
36 if (!name
|| in_interrupt() || size
< sizeof(void *) ||
37 size
> KMALLOC_MAX_SIZE
) {
38 pr_err("kmem_cache_create(%s) integrity check failed\n", name
);
42 list_for_each_entry(s
, &slab_caches
, list
) {
47 * This happens when the module gets unloaded and doesn't
48 * destroy its slab cache and no-one else reuses the vmalloc
49 * area of the module. Print a warning.
51 res
= probe_kernel_address(s
->name
, tmp
);
53 pr_err("Slab cache with size %d has lost its name\n",
58 #if !defined(CONFIG_SLUB) || !defined(CONFIG_SLUB_DEBUG_ON)
59 if (!strcmp(s
->name
, name
)) {
60 pr_err("%s (%s): Cache name already exists.\n",
69 WARN_ON(strchr(name
, ' ')); /* It confuses parsers */
73 static inline int kmem_cache_sanity_check(const char *name
, size_t size
)
79 #ifdef CONFIG_MEMCG_KMEM
80 int memcg_update_all_caches(int num_memcgs
)
84 mutex_lock(&slab_mutex
);
86 list_for_each_entry(s
, &slab_caches
, list
) {
87 if (!is_root_cache(s
))
90 ret
= memcg_update_cache_size(s
, num_memcgs
);
92 * See comment in memcontrol.c, memcg_update_cache_size:
93 * Instead of freeing the memory, we'll just leave the caches
94 * up to this point in an updated state.
100 memcg_update_array_size(num_memcgs
);
102 mutex_unlock(&slab_mutex
);
108 * Figure out what the alignment of the objects will be given a set of
109 * flags, a user specified alignment and the size of the objects.
111 unsigned long calculate_alignment(unsigned long flags
,
112 unsigned long align
, unsigned long size
)
115 * If the user wants hardware cache aligned objects then follow that
116 * suggestion if the object is sufficiently large.
118 * The hardware cache alignment cannot override the specified
119 * alignment though. If that is greater then use it.
121 if (flags
& SLAB_HWCACHE_ALIGN
) {
122 unsigned long ralign
= cache_line_size();
123 while (size
<= ralign
/ 2)
125 align
= max(align
, ralign
);
128 if (align
< ARCH_SLAB_MINALIGN
)
129 align
= ARCH_SLAB_MINALIGN
;
131 return ALIGN(align
, sizeof(void *));
134 static struct kmem_cache
*
135 do_kmem_cache_create(char *name
, size_t object_size
, size_t size
, size_t align
,
136 unsigned long flags
, void (*ctor
)(void *),
137 struct mem_cgroup
*memcg
, struct kmem_cache
*root_cache
)
139 struct kmem_cache
*s
;
143 s
= kmem_cache_zalloc(kmem_cache
, GFP_KERNEL
);
148 s
->object_size
= object_size
;
153 err
= memcg_alloc_cache_params(memcg
, s
, root_cache
);
157 err
= __kmem_cache_create(s
, flags
);
162 list_add(&s
->list
, &slab_caches
);
163 memcg_register_cache(s
);
170 memcg_free_cache_params(s
);
176 * kmem_cache_create - Create a cache.
177 * @name: A string which is used in /proc/slabinfo to identify this cache.
178 * @size: The size of objects to be created in this cache.
179 * @align: The required alignment for the objects.
181 * @ctor: A constructor for the objects.
183 * Returns a ptr to the cache on success, NULL on failure.
184 * Cannot be called within a interrupt, but can be interrupted.
185 * The @ctor is run when new pages are allocated by the cache.
189 * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5)
190 * to catch references to uninitialised memory.
192 * %SLAB_RED_ZONE - Insert `Red' zones around the allocated memory to check
193 * for buffer overruns.
195 * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware
196 * cacheline. This can be beneficial if you're counting cycles as closely
200 kmem_cache_create(const char *name
, size_t size
, size_t align
,
201 unsigned long flags
, void (*ctor
)(void *))
203 struct kmem_cache
*s
;
208 mutex_lock(&slab_mutex
);
210 err
= kmem_cache_sanity_check(name
, size
);
215 * Some allocators will constraint the set of valid flags to a subset
216 * of all flags. We expect them to define CACHE_CREATE_MASK in this
217 * case, and we'll just provide them with a sanitized version of the
220 flags
&= CACHE_CREATE_MASK
;
222 s
= __kmem_cache_alias(name
, size
, align
, flags
, ctor
);
226 cache_name
= kstrdup(name
, GFP_KERNEL
);
232 s
= do_kmem_cache_create(cache_name
, size
, size
,
233 calculate_alignment(flags
, align
, size
),
234 flags
, ctor
, NULL
, NULL
);
241 mutex_unlock(&slab_mutex
);
245 if (flags
& SLAB_PANIC
)
246 panic("kmem_cache_create: Failed to create slab '%s'. Error %d\n",
249 printk(KERN_WARNING
"kmem_cache_create(%s) failed with error %d",
257 EXPORT_SYMBOL(kmem_cache_create
);
259 #ifdef CONFIG_MEMCG_KMEM
261 * kmem_cache_create_memcg - Create a cache for a memory cgroup.
262 * @memcg: The memory cgroup the new cache is for.
263 * @root_cache: The parent of the new cache.
265 * This function attempts to create a kmem cache that will serve allocation
266 * requests going from @memcg to @root_cache. The new cache inherits properties
269 void kmem_cache_create_memcg(struct mem_cgroup
*memcg
, struct kmem_cache
*root_cache
)
271 struct kmem_cache
*s
;
275 mutex_lock(&slab_mutex
);
278 * Since per-memcg caches are created asynchronously on first
279 * allocation (see memcg_kmem_get_cache()), several threads can try to
280 * create the same cache, but only one of them may succeed.
282 if (cache_from_memcg_idx(root_cache
, memcg_cache_id(memcg
)))
285 cache_name
= memcg_create_cache_name(memcg
, root_cache
);
289 s
= do_kmem_cache_create(cache_name
, root_cache
->object_size
,
290 root_cache
->size
, root_cache
->align
,
291 root_cache
->flags
, root_cache
->ctor
,
298 s
->allocflags
|= __GFP_KMEMCG
;
301 mutex_unlock(&slab_mutex
);
304 #endif /* CONFIG_MEMCG_KMEM */
306 void kmem_cache_destroy(struct kmem_cache
*s
)
308 /* Destroy all the children caches if we aren't a memcg cache */
309 kmem_cache_destroy_memcg_children(s
);
312 mutex_lock(&slab_mutex
);
316 memcg_unregister_cache(s
);
318 if (!__kmem_cache_shutdown(s
)) {
319 mutex_unlock(&slab_mutex
);
320 if (s
->flags
& SLAB_DESTROY_BY_RCU
)
323 memcg_free_cache_params(s
);
325 kmem_cache_free(kmem_cache
, s
);
327 list_add(&s
->list
, &slab_caches
);
328 memcg_register_cache(s
);
329 mutex_unlock(&slab_mutex
);
330 printk(KERN_ERR
"kmem_cache_destroy %s: Slab cache still has objects\n",
335 mutex_unlock(&slab_mutex
);
339 EXPORT_SYMBOL(kmem_cache_destroy
);
341 int slab_is_available(void)
343 return slab_state
>= UP
;
347 /* Create a cache during boot when no slab services are available yet */
348 void __init
create_boot_cache(struct kmem_cache
*s
, const char *name
, size_t size
,
354 s
->size
= s
->object_size
= size
;
355 s
->align
= calculate_alignment(flags
, ARCH_KMALLOC_MINALIGN
, size
);
356 err
= __kmem_cache_create(s
, flags
);
359 panic("Creation of kmalloc slab %s size=%zu failed. Reason %d\n",
362 s
->refcount
= -1; /* Exempt from merging for now */
365 struct kmem_cache
*__init
create_kmalloc_cache(const char *name
, size_t size
,
368 struct kmem_cache
*s
= kmem_cache_zalloc(kmem_cache
, GFP_NOWAIT
);
371 panic("Out of memory when creating slab %s\n", name
);
373 create_boot_cache(s
, name
, size
, flags
);
374 list_add(&s
->list
, &slab_caches
);
379 struct kmem_cache
*kmalloc_caches
[KMALLOC_SHIFT_HIGH
+ 1];
380 EXPORT_SYMBOL(kmalloc_caches
);
382 #ifdef CONFIG_ZONE_DMA
383 struct kmem_cache
*kmalloc_dma_caches
[KMALLOC_SHIFT_HIGH
+ 1];
384 EXPORT_SYMBOL(kmalloc_dma_caches
);
388 * Conversion table for small slabs sizes / 8 to the index in the
389 * kmalloc array. This is necessary for slabs < 192 since we have non power
390 * of two cache sizes there. The size of larger slabs can be determined using
393 static s8 size_index
[24] = {
420 static inline int size_index_elem(size_t bytes
)
422 return (bytes
- 1) / 8;
426 * Find the kmem_cache structure that serves a given size of
429 struct kmem_cache
*kmalloc_slab(size_t size
, gfp_t flags
)
433 if (unlikely(size
> KMALLOC_MAX_SIZE
)) {
434 WARN_ON_ONCE(!(flags
& __GFP_NOWARN
));
440 return ZERO_SIZE_PTR
;
442 index
= size_index
[size_index_elem(size
)];
444 index
= fls(size
- 1);
446 #ifdef CONFIG_ZONE_DMA
447 if (unlikely((flags
& GFP_DMA
)))
448 return kmalloc_dma_caches
[index
];
451 return kmalloc_caches
[index
];
455 * Create the kmalloc array. Some of the regular kmalloc arrays
456 * may already have been created because they were needed to
457 * enable allocations for slab creation.
459 void __init
create_kmalloc_caches(unsigned long flags
)
464 * Patch up the size_index table if we have strange large alignment
465 * requirements for the kmalloc array. This is only the case for
466 * MIPS it seems. The standard arches will not generate any code here.
468 * Largest permitted alignment is 256 bytes due to the way we
469 * handle the index determination for the smaller caches.
471 * Make sure that nothing crazy happens if someone starts tinkering
472 * around with ARCH_KMALLOC_MINALIGN
474 BUILD_BUG_ON(KMALLOC_MIN_SIZE
> 256 ||
475 (KMALLOC_MIN_SIZE
& (KMALLOC_MIN_SIZE
- 1)));
477 for (i
= 8; i
< KMALLOC_MIN_SIZE
; i
+= 8) {
478 int elem
= size_index_elem(i
);
480 if (elem
>= ARRAY_SIZE(size_index
))
482 size_index
[elem
] = KMALLOC_SHIFT_LOW
;
485 if (KMALLOC_MIN_SIZE
>= 64) {
487 * The 96 byte size cache is not used if the alignment
490 for (i
= 64 + 8; i
<= 96; i
+= 8)
491 size_index
[size_index_elem(i
)] = 7;
495 if (KMALLOC_MIN_SIZE
>= 128) {
497 * The 192 byte sized cache is not used if the alignment
498 * is 128 byte. Redirect kmalloc to use the 256 byte cache
501 for (i
= 128 + 8; i
<= 192; i
+= 8)
502 size_index
[size_index_elem(i
)] = 8;
504 for (i
= KMALLOC_SHIFT_LOW
; i
<= KMALLOC_SHIFT_HIGH
; i
++) {
505 if (!kmalloc_caches
[i
]) {
506 kmalloc_caches
[i
] = create_kmalloc_cache(NULL
,
511 * Caches that are not of the two-to-the-power-of size.
512 * These have to be created immediately after the
513 * earlier power of two caches
515 if (KMALLOC_MIN_SIZE
<= 32 && !kmalloc_caches
[1] && i
== 6)
516 kmalloc_caches
[1] = create_kmalloc_cache(NULL
, 96, flags
);
518 if (KMALLOC_MIN_SIZE
<= 64 && !kmalloc_caches
[2] && i
== 7)
519 kmalloc_caches
[2] = create_kmalloc_cache(NULL
, 192, flags
);
522 /* Kmalloc array is now usable */
525 for (i
= 0; i
<= KMALLOC_SHIFT_HIGH
; i
++) {
526 struct kmem_cache
*s
= kmalloc_caches
[i
];
530 n
= kasprintf(GFP_NOWAIT
, "kmalloc-%d", kmalloc_size(i
));
537 #ifdef CONFIG_ZONE_DMA
538 for (i
= 0; i
<= KMALLOC_SHIFT_HIGH
; i
++) {
539 struct kmem_cache
*s
= kmalloc_caches
[i
];
542 int size
= kmalloc_size(i
);
543 char *n
= kasprintf(GFP_NOWAIT
,
544 "dma-kmalloc-%d", size
);
547 kmalloc_dma_caches
[i
] = create_kmalloc_cache(n
,
548 size
, SLAB_CACHE_DMA
| flags
);
553 #endif /* !CONFIG_SLOB */
555 #ifdef CONFIG_TRACING
556 void *kmalloc_order_trace(size_t size
, gfp_t flags
, unsigned int order
)
558 void *ret
= kmalloc_order(size
, flags
, order
);
559 trace_kmalloc(_RET_IP_
, ret
, size
, PAGE_SIZE
<< order
, flags
);
562 EXPORT_SYMBOL(kmalloc_order_trace
);
565 #ifdef CONFIG_SLABINFO
568 #define SLABINFO_RIGHTS (S_IWUSR | S_IRUSR)
570 #define SLABINFO_RIGHTS S_IRUSR
573 void print_slabinfo_header(struct seq_file
*m
)
576 * Output format version, so at least we can change it
577 * without _too_ many complaints.
579 #ifdef CONFIG_DEBUG_SLAB
580 seq_puts(m
, "slabinfo - version: 2.1 (statistics)\n");
582 seq_puts(m
, "slabinfo - version: 2.1\n");
584 seq_puts(m
, "# name <active_objs> <num_objs> <objsize> "
585 "<objperslab> <pagesperslab>");
586 seq_puts(m
, " : tunables <limit> <batchcount> <sharedfactor>");
587 seq_puts(m
, " : slabdata <active_slabs> <num_slabs> <sharedavail>");
588 #ifdef CONFIG_DEBUG_SLAB
589 seq_puts(m
, " : globalstat <listallocs> <maxobjs> <grown> <reaped> "
590 "<error> <maxfreeable> <nodeallocs> <remotefrees> <alienoverflow>");
591 seq_puts(m
, " : cpustat <allochit> <allocmiss> <freehit> <freemiss>");
596 static void *s_start(struct seq_file
*m
, loff_t
*pos
)
600 mutex_lock(&slab_mutex
);
602 print_slabinfo_header(m
);
604 return seq_list_start(&slab_caches
, *pos
);
607 void *slab_next(struct seq_file
*m
, void *p
, loff_t
*pos
)
609 return seq_list_next(p
, &slab_caches
, pos
);
612 void slab_stop(struct seq_file
*m
, void *p
)
614 mutex_unlock(&slab_mutex
);
618 memcg_accumulate_slabinfo(struct kmem_cache
*s
, struct slabinfo
*info
)
620 struct kmem_cache
*c
;
621 struct slabinfo sinfo
;
624 if (!is_root_cache(s
))
627 for_each_memcg_cache_index(i
) {
628 c
= cache_from_memcg_idx(s
, i
);
632 memset(&sinfo
, 0, sizeof(sinfo
));
633 get_slabinfo(c
, &sinfo
);
635 info
->active_slabs
+= sinfo
.active_slabs
;
636 info
->num_slabs
+= sinfo
.num_slabs
;
637 info
->shared_avail
+= sinfo
.shared_avail
;
638 info
->active_objs
+= sinfo
.active_objs
;
639 info
->num_objs
+= sinfo
.num_objs
;
643 int cache_show(struct kmem_cache
*s
, struct seq_file
*m
)
645 struct slabinfo sinfo
;
647 memset(&sinfo
, 0, sizeof(sinfo
));
648 get_slabinfo(s
, &sinfo
);
650 memcg_accumulate_slabinfo(s
, &sinfo
);
652 seq_printf(m
, "%-17s %6lu %6lu %6u %4u %4d",
653 cache_name(s
), sinfo
.active_objs
, sinfo
.num_objs
, s
->size
,
654 sinfo
.objects_per_slab
, (1 << sinfo
.cache_order
));
656 seq_printf(m
, " : tunables %4u %4u %4u",
657 sinfo
.limit
, sinfo
.batchcount
, sinfo
.shared
);
658 seq_printf(m
, " : slabdata %6lu %6lu %6lu",
659 sinfo
.active_slabs
, sinfo
.num_slabs
, sinfo
.shared_avail
);
660 slabinfo_show_stats(m
, s
);
665 static int s_show(struct seq_file
*m
, void *p
)
667 struct kmem_cache
*s
= list_entry(p
, struct kmem_cache
, list
);
669 if (!is_root_cache(s
))
671 return cache_show(s
, m
);
675 * slabinfo_op - iterator that generates /proc/slabinfo
685 * + further values on SMP and with statistics enabled
687 static const struct seq_operations slabinfo_op
= {
694 static int slabinfo_open(struct inode
*inode
, struct file
*file
)
696 return seq_open(file
, &slabinfo_op
);
699 static const struct file_operations proc_slabinfo_operations
= {
700 .open
= slabinfo_open
,
702 .write
= slabinfo_write
,
704 .release
= seq_release
,
707 static int __init
slab_proc_init(void)
709 proc_create("slabinfo", SLABINFO_RIGHTS
, NULL
,
710 &proc_slabinfo_operations
);
713 module_init(slab_proc_init
);
714 #endif /* CONFIG_SLABINFO */