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
);
169 memcg_free_cache_params(s
);
175 * kmem_cache_create - Create a cache.
176 * @name: A string which is used in /proc/slabinfo to identify this cache.
177 * @size: The size of objects to be created in this cache.
178 * @align: The required alignment for the objects.
180 * @ctor: A constructor for the objects.
182 * Returns a ptr to the cache on success, NULL on failure.
183 * Cannot be called within a interrupt, but can be interrupted.
184 * The @ctor is run when new pages are allocated by the cache.
188 * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5)
189 * to catch references to uninitialised memory.
191 * %SLAB_RED_ZONE - Insert `Red' zones around the allocated memory to check
192 * for buffer overruns.
194 * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware
195 * cacheline. This can be beneficial if you're counting cycles as closely
199 kmem_cache_create(const char *name
, size_t size
, size_t align
,
200 unsigned long flags
, void (*ctor
)(void *))
202 struct kmem_cache
*s
;
209 mutex_lock(&slab_mutex
);
211 err
= kmem_cache_sanity_check(name
, size
);
216 * Some allocators will constraint the set of valid flags to a subset
217 * of all flags. We expect them to define CACHE_CREATE_MASK in this
218 * case, and we'll just provide them with a sanitized version of the
221 flags
&= CACHE_CREATE_MASK
;
223 s
= __kmem_cache_alias(name
, size
, align
, flags
, ctor
);
227 cache_name
= kstrdup(name
, GFP_KERNEL
);
233 s
= do_kmem_cache_create(cache_name
, size
, size
,
234 calculate_alignment(flags
, align
, size
),
235 flags
, ctor
, NULL
, NULL
);
242 mutex_unlock(&slab_mutex
);
248 if (flags
& SLAB_PANIC
)
249 panic("kmem_cache_create: Failed to create slab '%s'. Error %d\n",
252 printk(KERN_WARNING
"kmem_cache_create(%s) failed with error %d",
260 EXPORT_SYMBOL(kmem_cache_create
);
262 #ifdef CONFIG_MEMCG_KMEM
264 * memcg_create_kmem_cache - Create a cache for a memory cgroup.
265 * @memcg: The memory cgroup the new cache is for.
266 * @root_cache: The parent of the new cache.
267 * @memcg_name: The name of the memory cgroup (used for naming the new cache).
269 * This function attempts to create a kmem cache that will serve allocation
270 * requests going from @memcg to @root_cache. The new cache inherits properties
273 struct kmem_cache
*memcg_create_kmem_cache(struct mem_cgroup
*memcg
,
274 struct kmem_cache
*root_cache
,
275 const char *memcg_name
)
277 struct kmem_cache
*s
= NULL
;
283 mutex_lock(&slab_mutex
);
285 cache_name
= kasprintf(GFP_KERNEL
, "%s(%d:%s)", root_cache
->name
,
286 memcg_cache_id(memcg
), memcg_name
);
290 s
= do_kmem_cache_create(cache_name
, root_cache
->object_size
,
291 root_cache
->size
, root_cache
->align
,
292 root_cache
->flags
, root_cache
->ctor
,
300 mutex_unlock(&slab_mutex
);
308 static int memcg_cleanup_cache_params(struct kmem_cache
*s
)
312 if (!s
->memcg_params
||
313 !s
->memcg_params
->is_root_cache
)
316 mutex_unlock(&slab_mutex
);
317 rc
= __memcg_cleanup_cache_params(s
);
318 mutex_lock(&slab_mutex
);
323 static int memcg_cleanup_cache_params(struct kmem_cache
*s
)
327 #endif /* CONFIG_MEMCG_KMEM */
329 void slab_kmem_cache_release(struct kmem_cache
*s
)
332 kmem_cache_free(kmem_cache
, s
);
335 void kmem_cache_destroy(struct kmem_cache
*s
)
340 mutex_lock(&slab_mutex
);
346 if (memcg_cleanup_cache_params(s
) != 0)
349 if (__kmem_cache_shutdown(s
) != 0) {
350 printk(KERN_ERR
"kmem_cache_destroy %s: "
351 "Slab cache still has objects\n", s
->name
);
358 mutex_unlock(&slab_mutex
);
359 if (s
->flags
& SLAB_DESTROY_BY_RCU
)
362 memcg_free_cache_params(s
);
363 #ifdef SLAB_SUPPORTS_SYSFS
364 sysfs_slab_remove(s
);
366 slab_kmem_cache_release(s
);
371 mutex_unlock(&slab_mutex
);
376 EXPORT_SYMBOL(kmem_cache_destroy
);
379 * kmem_cache_shrink - Shrink a cache.
380 * @cachep: The cache to shrink.
382 * Releases as many slabs as possible for a cache.
383 * To help debugging, a zero exit status indicates all slabs were released.
385 int kmem_cache_shrink(struct kmem_cache
*cachep
)
391 ret
= __kmem_cache_shrink(cachep
);
396 EXPORT_SYMBOL(kmem_cache_shrink
);
398 int slab_is_available(void)
400 return slab_state
>= UP
;
404 /* Create a cache during boot when no slab services are available yet */
405 void __init
create_boot_cache(struct kmem_cache
*s
, const char *name
, size_t size
,
411 s
->size
= s
->object_size
= size
;
412 s
->align
= calculate_alignment(flags
, ARCH_KMALLOC_MINALIGN
, size
);
413 err
= __kmem_cache_create(s
, flags
);
416 panic("Creation of kmalloc slab %s size=%zu failed. Reason %d\n",
419 s
->refcount
= -1; /* Exempt from merging for now */
422 struct kmem_cache
*__init
create_kmalloc_cache(const char *name
, size_t size
,
425 struct kmem_cache
*s
= kmem_cache_zalloc(kmem_cache
, GFP_NOWAIT
);
428 panic("Out of memory when creating slab %s\n", name
);
430 create_boot_cache(s
, name
, size
, flags
);
431 list_add(&s
->list
, &slab_caches
);
436 struct kmem_cache
*kmalloc_caches
[KMALLOC_SHIFT_HIGH
+ 1];
437 EXPORT_SYMBOL(kmalloc_caches
);
439 #ifdef CONFIG_ZONE_DMA
440 struct kmem_cache
*kmalloc_dma_caches
[KMALLOC_SHIFT_HIGH
+ 1];
441 EXPORT_SYMBOL(kmalloc_dma_caches
);
445 * Conversion table for small slabs sizes / 8 to the index in the
446 * kmalloc array. This is necessary for slabs < 192 since we have non power
447 * of two cache sizes there. The size of larger slabs can be determined using
450 static s8 size_index
[24] = {
477 static inline int size_index_elem(size_t bytes
)
479 return (bytes
- 1) / 8;
483 * Find the kmem_cache structure that serves a given size of
486 struct kmem_cache
*kmalloc_slab(size_t size
, gfp_t flags
)
490 if (unlikely(size
> KMALLOC_MAX_SIZE
)) {
491 WARN_ON_ONCE(!(flags
& __GFP_NOWARN
));
497 return ZERO_SIZE_PTR
;
499 index
= size_index
[size_index_elem(size
)];
501 index
= fls(size
- 1);
503 #ifdef CONFIG_ZONE_DMA
504 if (unlikely((flags
& GFP_DMA
)))
505 return kmalloc_dma_caches
[index
];
508 return kmalloc_caches
[index
];
512 * Create the kmalloc array. Some of the regular kmalloc arrays
513 * may already have been created because they were needed to
514 * enable allocations for slab creation.
516 void __init
create_kmalloc_caches(unsigned long flags
)
521 * Patch up the size_index table if we have strange large alignment
522 * requirements for the kmalloc array. This is only the case for
523 * MIPS it seems. The standard arches will not generate any code here.
525 * Largest permitted alignment is 256 bytes due to the way we
526 * handle the index determination for the smaller caches.
528 * Make sure that nothing crazy happens if someone starts tinkering
529 * around with ARCH_KMALLOC_MINALIGN
531 BUILD_BUG_ON(KMALLOC_MIN_SIZE
> 256 ||
532 (KMALLOC_MIN_SIZE
& (KMALLOC_MIN_SIZE
- 1)));
534 for (i
= 8; i
< KMALLOC_MIN_SIZE
; i
+= 8) {
535 int elem
= size_index_elem(i
);
537 if (elem
>= ARRAY_SIZE(size_index
))
539 size_index
[elem
] = KMALLOC_SHIFT_LOW
;
542 if (KMALLOC_MIN_SIZE
>= 64) {
544 * The 96 byte size cache is not used if the alignment
547 for (i
= 64 + 8; i
<= 96; i
+= 8)
548 size_index
[size_index_elem(i
)] = 7;
552 if (KMALLOC_MIN_SIZE
>= 128) {
554 * The 192 byte sized cache is not used if the alignment
555 * is 128 byte. Redirect kmalloc to use the 256 byte cache
558 for (i
= 128 + 8; i
<= 192; i
+= 8)
559 size_index
[size_index_elem(i
)] = 8;
561 for (i
= KMALLOC_SHIFT_LOW
; i
<= KMALLOC_SHIFT_HIGH
; i
++) {
562 if (!kmalloc_caches
[i
]) {
563 kmalloc_caches
[i
] = create_kmalloc_cache(NULL
,
568 * Caches that are not of the two-to-the-power-of size.
569 * These have to be created immediately after the
570 * earlier power of two caches
572 if (KMALLOC_MIN_SIZE
<= 32 && !kmalloc_caches
[1] && i
== 6)
573 kmalloc_caches
[1] = create_kmalloc_cache(NULL
, 96, flags
);
575 if (KMALLOC_MIN_SIZE
<= 64 && !kmalloc_caches
[2] && i
== 7)
576 kmalloc_caches
[2] = create_kmalloc_cache(NULL
, 192, flags
);
579 /* Kmalloc array is now usable */
582 for (i
= 0; i
<= KMALLOC_SHIFT_HIGH
; i
++) {
583 struct kmem_cache
*s
= kmalloc_caches
[i
];
587 n
= kasprintf(GFP_NOWAIT
, "kmalloc-%d", kmalloc_size(i
));
594 #ifdef CONFIG_ZONE_DMA
595 for (i
= 0; i
<= KMALLOC_SHIFT_HIGH
; i
++) {
596 struct kmem_cache
*s
= kmalloc_caches
[i
];
599 int size
= kmalloc_size(i
);
600 char *n
= kasprintf(GFP_NOWAIT
,
601 "dma-kmalloc-%d", size
);
604 kmalloc_dma_caches
[i
] = create_kmalloc_cache(n
,
605 size
, SLAB_CACHE_DMA
| flags
);
610 #endif /* !CONFIG_SLOB */
613 * To avoid unnecessary overhead, we pass through large allocation requests
614 * directly to the page allocator. We use __GFP_COMP, because we will need to
615 * know the allocation order to free the pages properly in kfree.
617 void *kmalloc_order(size_t size
, gfp_t flags
, unsigned int order
)
623 page
= alloc_kmem_pages(flags
, order
);
624 ret
= page
? page_address(page
) : NULL
;
625 kmemleak_alloc(ret
, size
, 1, flags
);
628 EXPORT_SYMBOL(kmalloc_order
);
630 #ifdef CONFIG_TRACING
631 void *kmalloc_order_trace(size_t size
, gfp_t flags
, unsigned int order
)
633 void *ret
= kmalloc_order(size
, flags
, order
);
634 trace_kmalloc(_RET_IP_
, ret
, size
, PAGE_SIZE
<< order
, flags
);
637 EXPORT_SYMBOL(kmalloc_order_trace
);
640 #ifdef CONFIG_SLABINFO
643 #define SLABINFO_RIGHTS (S_IWUSR | S_IRUSR)
645 #define SLABINFO_RIGHTS S_IRUSR
648 void print_slabinfo_header(struct seq_file
*m
)
651 * Output format version, so at least we can change it
652 * without _too_ many complaints.
654 #ifdef CONFIG_DEBUG_SLAB
655 seq_puts(m
, "slabinfo - version: 2.1 (statistics)\n");
657 seq_puts(m
, "slabinfo - version: 2.1\n");
659 seq_puts(m
, "# name <active_objs> <num_objs> <objsize> "
660 "<objperslab> <pagesperslab>");
661 seq_puts(m
, " : tunables <limit> <batchcount> <sharedfactor>");
662 seq_puts(m
, " : slabdata <active_slabs> <num_slabs> <sharedavail>");
663 #ifdef CONFIG_DEBUG_SLAB
664 seq_puts(m
, " : globalstat <listallocs> <maxobjs> <grown> <reaped> "
665 "<error> <maxfreeable> <nodeallocs> <remotefrees> <alienoverflow>");
666 seq_puts(m
, " : cpustat <allochit> <allocmiss> <freehit> <freemiss>");
671 static void *s_start(struct seq_file
*m
, loff_t
*pos
)
675 mutex_lock(&slab_mutex
);
677 print_slabinfo_header(m
);
679 return seq_list_start(&slab_caches
, *pos
);
682 void *slab_next(struct seq_file
*m
, void *p
, loff_t
*pos
)
684 return seq_list_next(p
, &slab_caches
, pos
);
687 void slab_stop(struct seq_file
*m
, void *p
)
689 mutex_unlock(&slab_mutex
);
693 memcg_accumulate_slabinfo(struct kmem_cache
*s
, struct slabinfo
*info
)
695 struct kmem_cache
*c
;
696 struct slabinfo sinfo
;
699 if (!is_root_cache(s
))
702 for_each_memcg_cache_index(i
) {
703 c
= cache_from_memcg_idx(s
, i
);
707 memset(&sinfo
, 0, sizeof(sinfo
));
708 get_slabinfo(c
, &sinfo
);
710 info
->active_slabs
+= sinfo
.active_slabs
;
711 info
->num_slabs
+= sinfo
.num_slabs
;
712 info
->shared_avail
+= sinfo
.shared_avail
;
713 info
->active_objs
+= sinfo
.active_objs
;
714 info
->num_objs
+= sinfo
.num_objs
;
718 int cache_show(struct kmem_cache
*s
, struct seq_file
*m
)
720 struct slabinfo sinfo
;
722 memset(&sinfo
, 0, sizeof(sinfo
));
723 get_slabinfo(s
, &sinfo
);
725 memcg_accumulate_slabinfo(s
, &sinfo
);
727 seq_printf(m
, "%-17s %6lu %6lu %6u %4u %4d",
728 cache_name(s
), sinfo
.active_objs
, sinfo
.num_objs
, s
->size
,
729 sinfo
.objects_per_slab
, (1 << sinfo
.cache_order
));
731 seq_printf(m
, " : tunables %4u %4u %4u",
732 sinfo
.limit
, sinfo
.batchcount
, sinfo
.shared
);
733 seq_printf(m
, " : slabdata %6lu %6lu %6lu",
734 sinfo
.active_slabs
, sinfo
.num_slabs
, sinfo
.shared_avail
);
735 slabinfo_show_stats(m
, s
);
740 static int s_show(struct seq_file
*m
, void *p
)
742 struct kmem_cache
*s
= list_entry(p
, struct kmem_cache
, list
);
744 if (!is_root_cache(s
))
746 return cache_show(s
, m
);
750 * slabinfo_op - iterator that generates /proc/slabinfo
760 * + further values on SMP and with statistics enabled
762 static const struct seq_operations slabinfo_op
= {
769 static int slabinfo_open(struct inode
*inode
, struct file
*file
)
771 return seq_open(file
, &slabinfo_op
);
774 static const struct file_operations proc_slabinfo_operations
= {
775 .open
= slabinfo_open
,
777 .write
= slabinfo_write
,
779 .release
= seq_release
,
782 static int __init
slab_proc_init(void)
784 proc_create("slabinfo", SLABINFO_RIGHTS
, NULL
,
785 &proc_slabinfo_operations
);
788 module_init(slab_proc_init
);
789 #endif /* CONFIG_SLABINFO */