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>
23 #define CREATE_TRACE_POINTS
24 #include <trace/events/kmem.h>
28 enum slab_state slab_state
;
29 LIST_HEAD(slab_caches
);
30 DEFINE_MUTEX(slab_mutex
);
31 struct kmem_cache
*kmem_cache
;
33 static LIST_HEAD(slab_caches_to_rcu_destroy
);
34 static void slab_caches_to_rcu_destroy_workfn(struct work_struct
*work
);
35 static DECLARE_WORK(slab_caches_to_rcu_destroy_work
,
36 slab_caches_to_rcu_destroy_workfn
);
39 * Set of flags that will prevent slab merging
41 #define SLAB_NEVER_MERGE (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
42 SLAB_TRACE | SLAB_TYPESAFE_BY_RCU | SLAB_NOLEAKTRACE | \
43 SLAB_FAILSLAB | SLAB_KASAN)
45 #define SLAB_MERGE_SAME (SLAB_RECLAIM_ACCOUNT | SLAB_CACHE_DMA | \
46 SLAB_NOTRACK | SLAB_ACCOUNT)
49 * Merge control. If this is set then no merging of slab caches will occur.
51 static bool slab_nomerge
= !IS_ENABLED(CONFIG_SLAB_MERGE_DEFAULT
);
53 static int __init
setup_slab_nomerge(char *str
)
60 __setup_param("slub_nomerge", slub_nomerge
, setup_slab_nomerge
, 0);
63 __setup("slab_nomerge", setup_slab_nomerge
);
66 * Determine the size of a slab object
68 unsigned int kmem_cache_size(struct kmem_cache
*s
)
70 return s
->object_size
;
72 EXPORT_SYMBOL(kmem_cache_size
);
74 #ifdef CONFIG_DEBUG_VM
75 static int kmem_cache_sanity_check(const char *name
, size_t size
)
77 struct kmem_cache
*s
= NULL
;
79 if (!name
|| in_interrupt() || size
< sizeof(void *) ||
80 size
> KMALLOC_MAX_SIZE
) {
81 pr_err("kmem_cache_create(%s) integrity check failed\n", name
);
85 list_for_each_entry(s
, &slab_caches
, list
) {
90 * This happens when the module gets unloaded and doesn't
91 * destroy its slab cache and no-one else reuses the vmalloc
92 * area of the module. Print a warning.
94 res
= probe_kernel_address(s
->name
, tmp
);
96 pr_err("Slab cache with size %d has lost its name\n",
102 WARN_ON(strchr(name
, ' ')); /* It confuses parsers */
106 static inline int kmem_cache_sanity_check(const char *name
, size_t size
)
112 void __kmem_cache_free_bulk(struct kmem_cache
*s
, size_t nr
, void **p
)
116 for (i
= 0; i
< nr
; i
++) {
118 kmem_cache_free(s
, p
[i
]);
124 int __kmem_cache_alloc_bulk(struct kmem_cache
*s
, gfp_t flags
, size_t nr
,
129 for (i
= 0; i
< nr
; i
++) {
130 void *x
= p
[i
] = kmem_cache_alloc(s
, flags
);
132 __kmem_cache_free_bulk(s
, i
, p
);
139 #if defined(CONFIG_MEMCG) && !defined(CONFIG_SLOB)
141 LIST_HEAD(slab_root_caches
);
143 void slab_init_memcg_params(struct kmem_cache
*s
)
145 s
->memcg_params
.root_cache
= NULL
;
146 RCU_INIT_POINTER(s
->memcg_params
.memcg_caches
, NULL
);
147 INIT_LIST_HEAD(&s
->memcg_params
.children
);
150 static int init_memcg_params(struct kmem_cache
*s
,
151 struct mem_cgroup
*memcg
, struct kmem_cache
*root_cache
)
153 struct memcg_cache_array
*arr
;
156 s
->memcg_params
.root_cache
= root_cache
;
157 s
->memcg_params
.memcg
= memcg
;
158 INIT_LIST_HEAD(&s
->memcg_params
.children_node
);
159 INIT_LIST_HEAD(&s
->memcg_params
.kmem_caches_node
);
163 slab_init_memcg_params(s
);
165 if (!memcg_nr_cache_ids
)
168 arr
= kvzalloc(sizeof(struct memcg_cache_array
) +
169 memcg_nr_cache_ids
* sizeof(void *),
174 RCU_INIT_POINTER(s
->memcg_params
.memcg_caches
, arr
);
178 static void destroy_memcg_params(struct kmem_cache
*s
)
180 if (is_root_cache(s
))
181 kvfree(rcu_access_pointer(s
->memcg_params
.memcg_caches
));
184 static void free_memcg_params(struct rcu_head
*rcu
)
186 struct memcg_cache_array
*old
;
188 old
= container_of(rcu
, struct memcg_cache_array
, rcu
);
192 static int update_memcg_params(struct kmem_cache
*s
, int new_array_size
)
194 struct memcg_cache_array
*old
, *new;
196 new = kvzalloc(sizeof(struct memcg_cache_array
) +
197 new_array_size
* sizeof(void *), GFP_KERNEL
);
201 old
= rcu_dereference_protected(s
->memcg_params
.memcg_caches
,
202 lockdep_is_held(&slab_mutex
));
204 memcpy(new->entries
, old
->entries
,
205 memcg_nr_cache_ids
* sizeof(void *));
207 rcu_assign_pointer(s
->memcg_params
.memcg_caches
, new);
209 call_rcu(&old
->rcu
, free_memcg_params
);
213 int memcg_update_all_caches(int num_memcgs
)
215 struct kmem_cache
*s
;
218 mutex_lock(&slab_mutex
);
219 list_for_each_entry(s
, &slab_root_caches
, root_caches_node
) {
220 ret
= update_memcg_params(s
, num_memcgs
);
222 * Instead of freeing the memory, we'll just leave the caches
223 * up to this point in an updated state.
228 mutex_unlock(&slab_mutex
);
232 void memcg_link_cache(struct kmem_cache
*s
)
234 if (is_root_cache(s
)) {
235 list_add(&s
->root_caches_node
, &slab_root_caches
);
237 list_add(&s
->memcg_params
.children_node
,
238 &s
->memcg_params
.root_cache
->memcg_params
.children
);
239 list_add(&s
->memcg_params
.kmem_caches_node
,
240 &s
->memcg_params
.memcg
->kmem_caches
);
244 static void memcg_unlink_cache(struct kmem_cache
*s
)
246 if (is_root_cache(s
)) {
247 list_del(&s
->root_caches_node
);
249 list_del(&s
->memcg_params
.children_node
);
250 list_del(&s
->memcg_params
.kmem_caches_node
);
254 static inline int init_memcg_params(struct kmem_cache
*s
,
255 struct mem_cgroup
*memcg
, struct kmem_cache
*root_cache
)
260 static inline void destroy_memcg_params(struct kmem_cache
*s
)
264 static inline void memcg_unlink_cache(struct kmem_cache
*s
)
267 #endif /* CONFIG_MEMCG && !CONFIG_SLOB */
270 * Find a mergeable slab cache
272 int slab_unmergeable(struct kmem_cache
*s
)
274 if (slab_nomerge
|| (s
->flags
& SLAB_NEVER_MERGE
))
277 if (!is_root_cache(s
))
284 * We may have set a slab to be unmergeable during bootstrap.
292 struct kmem_cache
*find_mergeable(size_t size
, size_t align
,
293 unsigned long flags
, const char *name
, void (*ctor
)(void *))
295 struct kmem_cache
*s
;
303 size
= ALIGN(size
, sizeof(void *));
304 align
= calculate_alignment(flags
, align
, size
);
305 size
= ALIGN(size
, align
);
306 flags
= kmem_cache_flags(size
, flags
, name
, NULL
);
308 if (flags
& SLAB_NEVER_MERGE
)
311 list_for_each_entry_reverse(s
, &slab_root_caches
, root_caches_node
) {
312 if (slab_unmergeable(s
))
318 if ((flags
& SLAB_MERGE_SAME
) != (s
->flags
& SLAB_MERGE_SAME
))
321 * Check if alignment is compatible.
322 * Courtesy of Adrian Drzewiecki
324 if ((s
->size
& ~(align
- 1)) != s
->size
)
327 if (s
->size
- size
>= sizeof(void *))
330 if (IS_ENABLED(CONFIG_SLAB
) && align
&&
331 (align
> s
->align
|| s
->align
% align
))
340 * Figure out what the alignment of the objects will be given a set of
341 * flags, a user specified alignment and the size of the objects.
343 unsigned long calculate_alignment(unsigned long flags
,
344 unsigned long align
, unsigned long size
)
347 * If the user wants hardware cache aligned objects then follow that
348 * suggestion if the object is sufficiently large.
350 * The hardware cache alignment cannot override the specified
351 * alignment though. If that is greater then use it.
353 if (flags
& SLAB_HWCACHE_ALIGN
) {
354 unsigned long ralign
= cache_line_size();
355 while (size
<= ralign
/ 2)
357 align
= max(align
, ralign
);
360 if (align
< ARCH_SLAB_MINALIGN
)
361 align
= ARCH_SLAB_MINALIGN
;
363 return ALIGN(align
, sizeof(void *));
366 static struct kmem_cache
*create_cache(const char *name
,
367 size_t object_size
, size_t size
, size_t align
,
368 unsigned long flags
, void (*ctor
)(void *),
369 struct mem_cgroup
*memcg
, struct kmem_cache
*root_cache
)
371 struct kmem_cache
*s
;
375 s
= kmem_cache_zalloc(kmem_cache
, GFP_KERNEL
);
380 s
->object_size
= object_size
;
385 err
= init_memcg_params(s
, memcg
, root_cache
);
389 err
= __kmem_cache_create(s
, flags
);
394 list_add(&s
->list
, &slab_caches
);
402 destroy_memcg_params(s
);
403 kmem_cache_free(kmem_cache
, s
);
408 * kmem_cache_create - Create a cache.
409 * @name: A string which is used in /proc/slabinfo to identify this cache.
410 * @size: The size of objects to be created in this cache.
411 * @align: The required alignment for the objects.
413 * @ctor: A constructor for the objects.
415 * Returns a ptr to the cache on success, NULL on failure.
416 * Cannot be called within a interrupt, but can be interrupted.
417 * The @ctor is run when new pages are allocated by the cache.
421 * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5)
422 * to catch references to uninitialised memory.
424 * %SLAB_RED_ZONE - Insert `Red' zones around the allocated memory to check
425 * for buffer overruns.
427 * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware
428 * cacheline. This can be beneficial if you're counting cycles as closely
432 kmem_cache_create(const char *name
, size_t size
, size_t align
,
433 unsigned long flags
, void (*ctor
)(void *))
435 struct kmem_cache
*s
= NULL
;
436 const char *cache_name
;
441 memcg_get_cache_ids();
443 mutex_lock(&slab_mutex
);
445 err
= kmem_cache_sanity_check(name
, size
);
450 /* Refuse requests with allocator specific flags */
451 if (flags
& ~SLAB_FLAGS_PERMITTED
) {
457 * Some allocators will constraint the set of valid flags to a subset
458 * of all flags. We expect them to define CACHE_CREATE_MASK in this
459 * case, and we'll just provide them with a sanitized version of the
462 flags
&= CACHE_CREATE_MASK
;
464 s
= __kmem_cache_alias(name
, size
, align
, flags
, ctor
);
468 cache_name
= kstrdup_const(name
, GFP_KERNEL
);
474 s
= create_cache(cache_name
, size
, size
,
475 calculate_alignment(flags
, align
, size
),
476 flags
, ctor
, NULL
, NULL
);
479 kfree_const(cache_name
);
483 mutex_unlock(&slab_mutex
);
485 memcg_put_cache_ids();
490 if (flags
& SLAB_PANIC
)
491 panic("kmem_cache_create: Failed to create slab '%s'. Error %d\n",
494 pr_warn("kmem_cache_create(%s) failed with error %d\n",
502 EXPORT_SYMBOL(kmem_cache_create
);
504 static void slab_caches_to_rcu_destroy_workfn(struct work_struct
*work
)
506 LIST_HEAD(to_destroy
);
507 struct kmem_cache
*s
, *s2
;
510 * On destruction, SLAB_TYPESAFE_BY_RCU kmem_caches are put on the
511 * @slab_caches_to_rcu_destroy list. The slab pages are freed
512 * through RCU and and the associated kmem_cache are dereferenced
513 * while freeing the pages, so the kmem_caches should be freed only
514 * after the pending RCU operations are finished. As rcu_barrier()
515 * is a pretty slow operation, we batch all pending destructions
518 mutex_lock(&slab_mutex
);
519 list_splice_init(&slab_caches_to_rcu_destroy
, &to_destroy
);
520 mutex_unlock(&slab_mutex
);
522 if (list_empty(&to_destroy
))
527 list_for_each_entry_safe(s
, s2
, &to_destroy
, list
) {
528 #ifdef SLAB_SUPPORTS_SYSFS
529 sysfs_slab_release(s
);
531 slab_kmem_cache_release(s
);
536 static int shutdown_cache(struct kmem_cache
*s
)
538 /* free asan quarantined objects */
539 kasan_cache_shutdown(s
);
541 if (__kmem_cache_shutdown(s
) != 0)
544 memcg_unlink_cache(s
);
547 if (s
->flags
& SLAB_TYPESAFE_BY_RCU
) {
548 list_add_tail(&s
->list
, &slab_caches_to_rcu_destroy
);
549 schedule_work(&slab_caches_to_rcu_destroy_work
);
551 #ifdef SLAB_SUPPORTS_SYSFS
552 sysfs_slab_release(s
);
554 slab_kmem_cache_release(s
);
561 #if defined(CONFIG_MEMCG) && !defined(CONFIG_SLOB)
563 * memcg_create_kmem_cache - Create a cache for a memory cgroup.
564 * @memcg: The memory cgroup the new cache is for.
565 * @root_cache: The parent of the new cache.
567 * This function attempts to create a kmem cache that will serve allocation
568 * requests going from @memcg to @root_cache. The new cache inherits properties
571 void memcg_create_kmem_cache(struct mem_cgroup
*memcg
,
572 struct kmem_cache
*root_cache
)
574 static char memcg_name_buf
[NAME_MAX
+ 1]; /* protected by slab_mutex */
575 struct cgroup_subsys_state
*css
= &memcg
->css
;
576 struct memcg_cache_array
*arr
;
577 struct kmem_cache
*s
= NULL
;
584 mutex_lock(&slab_mutex
);
587 * The memory cgroup could have been offlined while the cache
588 * creation work was pending.
590 if (memcg
->kmem_state
!= KMEM_ONLINE
)
593 idx
= memcg_cache_id(memcg
);
594 arr
= rcu_dereference_protected(root_cache
->memcg_params
.memcg_caches
,
595 lockdep_is_held(&slab_mutex
));
598 * Since per-memcg caches are created asynchronously on first
599 * allocation (see memcg_kmem_get_cache()), several threads can try to
600 * create the same cache, but only one of them may succeed.
602 if (arr
->entries
[idx
])
605 cgroup_name(css
->cgroup
, memcg_name_buf
, sizeof(memcg_name_buf
));
606 cache_name
= kasprintf(GFP_KERNEL
, "%s(%llu:%s)", root_cache
->name
,
607 css
->serial_nr
, memcg_name_buf
);
611 s
= create_cache(cache_name
, root_cache
->object_size
,
612 root_cache
->size
, root_cache
->align
,
613 root_cache
->flags
& CACHE_CREATE_MASK
,
614 root_cache
->ctor
, memcg
, root_cache
);
616 * If we could not create a memcg cache, do not complain, because
617 * that's not critical at all as we can always proceed with the root
626 * Since readers won't lock (see cache_from_memcg_idx()), we need a
627 * barrier here to ensure nobody will see the kmem_cache partially
631 arr
->entries
[idx
] = s
;
634 mutex_unlock(&slab_mutex
);
640 static void kmemcg_deactivate_workfn(struct work_struct
*work
)
642 struct kmem_cache
*s
= container_of(work
, struct kmem_cache
,
643 memcg_params
.deact_work
);
648 mutex_lock(&slab_mutex
);
650 s
->memcg_params
.deact_fn(s
);
652 mutex_unlock(&slab_mutex
);
657 /* done, put the ref from slab_deactivate_memcg_cache_rcu_sched() */
658 css_put(&s
->memcg_params
.memcg
->css
);
661 static void kmemcg_deactivate_rcufn(struct rcu_head
*head
)
663 struct kmem_cache
*s
= container_of(head
, struct kmem_cache
,
664 memcg_params
.deact_rcu_head
);
667 * We need to grab blocking locks. Bounce to ->deact_work. The
668 * work item shares the space with the RCU head and can't be
669 * initialized eariler.
671 INIT_WORK(&s
->memcg_params
.deact_work
, kmemcg_deactivate_workfn
);
672 queue_work(memcg_kmem_cache_wq
, &s
->memcg_params
.deact_work
);
676 * slab_deactivate_memcg_cache_rcu_sched - schedule deactivation after a
677 * sched RCU grace period
678 * @s: target kmem_cache
679 * @deact_fn: deactivation function to call
681 * Schedule @deact_fn to be invoked with online cpus, mems and slab_mutex
682 * held after a sched RCU grace period. The slab is guaranteed to stay
683 * alive until @deact_fn is finished. This is to be used from
684 * __kmemcg_cache_deactivate().
686 void slab_deactivate_memcg_cache_rcu_sched(struct kmem_cache
*s
,
687 void (*deact_fn
)(struct kmem_cache
*))
689 if (WARN_ON_ONCE(is_root_cache(s
)) ||
690 WARN_ON_ONCE(s
->memcg_params
.deact_fn
))
693 /* pin memcg so that @s doesn't get destroyed in the middle */
694 css_get(&s
->memcg_params
.memcg
->css
);
696 s
->memcg_params
.deact_fn
= deact_fn
;
697 call_rcu_sched(&s
->memcg_params
.deact_rcu_head
, kmemcg_deactivate_rcufn
);
700 void memcg_deactivate_kmem_caches(struct mem_cgroup
*memcg
)
703 struct memcg_cache_array
*arr
;
704 struct kmem_cache
*s
, *c
;
706 idx
= memcg_cache_id(memcg
);
711 mutex_lock(&slab_mutex
);
712 list_for_each_entry(s
, &slab_root_caches
, root_caches_node
) {
713 arr
= rcu_dereference_protected(s
->memcg_params
.memcg_caches
,
714 lockdep_is_held(&slab_mutex
));
715 c
= arr
->entries
[idx
];
719 __kmemcg_cache_deactivate(c
);
720 arr
->entries
[idx
] = NULL
;
722 mutex_unlock(&slab_mutex
);
728 void memcg_destroy_kmem_caches(struct mem_cgroup
*memcg
)
730 struct kmem_cache
*s
, *s2
;
735 mutex_lock(&slab_mutex
);
736 list_for_each_entry_safe(s
, s2
, &memcg
->kmem_caches
,
737 memcg_params
.kmem_caches_node
) {
739 * The cgroup is about to be freed and therefore has no charges
740 * left. Hence, all its caches must be empty by now.
742 BUG_ON(shutdown_cache(s
));
744 mutex_unlock(&slab_mutex
);
750 static int shutdown_memcg_caches(struct kmem_cache
*s
)
752 struct memcg_cache_array
*arr
;
753 struct kmem_cache
*c
, *c2
;
757 BUG_ON(!is_root_cache(s
));
760 * First, shutdown active caches, i.e. caches that belong to online
763 arr
= rcu_dereference_protected(s
->memcg_params
.memcg_caches
,
764 lockdep_is_held(&slab_mutex
));
765 for_each_memcg_cache_index(i
) {
769 if (shutdown_cache(c
))
771 * The cache still has objects. Move it to a temporary
772 * list so as not to try to destroy it for a second
773 * time while iterating over inactive caches below.
775 list_move(&c
->memcg_params
.children_node
, &busy
);
778 * The cache is empty and will be destroyed soon. Clear
779 * the pointer to it in the memcg_caches array so that
780 * it will never be accessed even if the root cache
783 arr
->entries
[i
] = NULL
;
787 * Second, shutdown all caches left from memory cgroups that are now
790 list_for_each_entry_safe(c
, c2
, &s
->memcg_params
.children
,
791 memcg_params
.children_node
)
794 list_splice(&busy
, &s
->memcg_params
.children
);
797 * A cache being destroyed must be empty. In particular, this means
798 * that all per memcg caches attached to it must be empty too.
800 if (!list_empty(&s
->memcg_params
.children
))
805 static inline int shutdown_memcg_caches(struct kmem_cache
*s
)
809 #endif /* CONFIG_MEMCG && !CONFIG_SLOB */
811 void slab_kmem_cache_release(struct kmem_cache
*s
)
813 __kmem_cache_release(s
);
814 destroy_memcg_params(s
);
815 kfree_const(s
->name
);
816 kmem_cache_free(kmem_cache
, s
);
819 void kmem_cache_destroy(struct kmem_cache
*s
)
829 mutex_lock(&slab_mutex
);
835 err
= shutdown_memcg_caches(s
);
837 err
= shutdown_cache(s
);
840 pr_err("kmem_cache_destroy %s: Slab cache still has objects\n",
845 mutex_unlock(&slab_mutex
);
850 EXPORT_SYMBOL(kmem_cache_destroy
);
853 * kmem_cache_shrink - Shrink a cache.
854 * @cachep: The cache to shrink.
856 * Releases as many slabs as possible for a cache.
857 * To help debugging, a zero exit status indicates all slabs were released.
859 int kmem_cache_shrink(struct kmem_cache
*cachep
)
865 kasan_cache_shrink(cachep
);
866 ret
= __kmem_cache_shrink(cachep
);
871 EXPORT_SYMBOL(kmem_cache_shrink
);
873 bool slab_is_available(void)
875 return slab_state
>= UP
;
879 /* Create a cache during boot when no slab services are available yet */
880 void __init
create_boot_cache(struct kmem_cache
*s
, const char *name
, size_t size
,
886 s
->size
= s
->object_size
= size
;
887 s
->align
= calculate_alignment(flags
, ARCH_KMALLOC_MINALIGN
, size
);
889 slab_init_memcg_params(s
);
891 err
= __kmem_cache_create(s
, flags
);
894 panic("Creation of kmalloc slab %s size=%zu failed. Reason %d\n",
897 s
->refcount
= -1; /* Exempt from merging for now */
900 struct kmem_cache
*__init
create_kmalloc_cache(const char *name
, size_t size
,
903 struct kmem_cache
*s
= kmem_cache_zalloc(kmem_cache
, GFP_NOWAIT
);
906 panic("Out of memory when creating slab %s\n", name
);
908 create_boot_cache(s
, name
, size
, flags
);
909 list_add(&s
->list
, &slab_caches
);
915 struct kmem_cache
*kmalloc_caches
[KMALLOC_SHIFT_HIGH
+ 1];
916 EXPORT_SYMBOL(kmalloc_caches
);
918 #ifdef CONFIG_ZONE_DMA
919 struct kmem_cache
*kmalloc_dma_caches
[KMALLOC_SHIFT_HIGH
+ 1];
920 EXPORT_SYMBOL(kmalloc_dma_caches
);
924 * Conversion table for small slabs sizes / 8 to the index in the
925 * kmalloc array. This is necessary for slabs < 192 since we have non power
926 * of two cache sizes there. The size of larger slabs can be determined using
929 static s8 size_index
[24] = {
956 static inline int size_index_elem(size_t bytes
)
958 return (bytes
- 1) / 8;
962 * Find the kmem_cache structure that serves a given size of
965 struct kmem_cache
*kmalloc_slab(size_t size
, gfp_t flags
)
969 if (unlikely(size
> KMALLOC_MAX_SIZE
)) {
970 WARN_ON_ONCE(!(flags
& __GFP_NOWARN
));
976 return ZERO_SIZE_PTR
;
978 index
= size_index
[size_index_elem(size
)];
980 index
= fls(size
- 1);
982 #ifdef CONFIG_ZONE_DMA
983 if (unlikely((flags
& GFP_DMA
)))
984 return kmalloc_dma_caches
[index
];
987 return kmalloc_caches
[index
];
991 * kmalloc_info[] is to make slub_debug=,kmalloc-xx option work at boot time.
992 * kmalloc_index() supports up to 2^26=64MB, so the final entry of the table is
995 const struct kmalloc_info_struct kmalloc_info
[] __initconst
= {
996 {NULL
, 0}, {"kmalloc-96", 96},
997 {"kmalloc-192", 192}, {"kmalloc-8", 8},
998 {"kmalloc-16", 16}, {"kmalloc-32", 32},
999 {"kmalloc-64", 64}, {"kmalloc-128", 128},
1000 {"kmalloc-256", 256}, {"kmalloc-512", 512},
1001 {"kmalloc-1024", 1024}, {"kmalloc-2048", 2048},
1002 {"kmalloc-4096", 4096}, {"kmalloc-8192", 8192},
1003 {"kmalloc-16384", 16384}, {"kmalloc-32768", 32768},
1004 {"kmalloc-65536", 65536}, {"kmalloc-131072", 131072},
1005 {"kmalloc-262144", 262144}, {"kmalloc-524288", 524288},
1006 {"kmalloc-1048576", 1048576}, {"kmalloc-2097152", 2097152},
1007 {"kmalloc-4194304", 4194304}, {"kmalloc-8388608", 8388608},
1008 {"kmalloc-16777216", 16777216}, {"kmalloc-33554432", 33554432},
1009 {"kmalloc-67108864", 67108864}
1013 * Patch up the size_index table if we have strange large alignment
1014 * requirements for the kmalloc array. This is only the case for
1015 * MIPS it seems. The standard arches will not generate any code here.
1017 * Largest permitted alignment is 256 bytes due to the way we
1018 * handle the index determination for the smaller caches.
1020 * Make sure that nothing crazy happens if someone starts tinkering
1021 * around with ARCH_KMALLOC_MINALIGN
1023 void __init
setup_kmalloc_cache_index_table(void)
1027 BUILD_BUG_ON(KMALLOC_MIN_SIZE
> 256 ||
1028 (KMALLOC_MIN_SIZE
& (KMALLOC_MIN_SIZE
- 1)));
1030 for (i
= 8; i
< KMALLOC_MIN_SIZE
; i
+= 8) {
1031 int elem
= size_index_elem(i
);
1033 if (elem
>= ARRAY_SIZE(size_index
))
1035 size_index
[elem
] = KMALLOC_SHIFT_LOW
;
1038 if (KMALLOC_MIN_SIZE
>= 64) {
1040 * The 96 byte size cache is not used if the alignment
1043 for (i
= 64 + 8; i
<= 96; i
+= 8)
1044 size_index
[size_index_elem(i
)] = 7;
1048 if (KMALLOC_MIN_SIZE
>= 128) {
1050 * The 192 byte sized cache is not used if the alignment
1051 * is 128 byte. Redirect kmalloc to use the 256 byte cache
1054 for (i
= 128 + 8; i
<= 192; i
+= 8)
1055 size_index
[size_index_elem(i
)] = 8;
1059 static void __init
new_kmalloc_cache(int idx
, unsigned long flags
)
1061 kmalloc_caches
[idx
] = create_kmalloc_cache(kmalloc_info
[idx
].name
,
1062 kmalloc_info
[idx
].size
, flags
);
1066 * Create the kmalloc array. Some of the regular kmalloc arrays
1067 * may already have been created because they were needed to
1068 * enable allocations for slab creation.
1070 void __init
create_kmalloc_caches(unsigned long flags
)
1074 for (i
= KMALLOC_SHIFT_LOW
; i
<= KMALLOC_SHIFT_HIGH
; i
++) {
1075 if (!kmalloc_caches
[i
])
1076 new_kmalloc_cache(i
, flags
);
1079 * Caches that are not of the two-to-the-power-of size.
1080 * These have to be created immediately after the
1081 * earlier power of two caches
1083 if (KMALLOC_MIN_SIZE
<= 32 && !kmalloc_caches
[1] && i
== 6)
1084 new_kmalloc_cache(1, flags
);
1085 if (KMALLOC_MIN_SIZE
<= 64 && !kmalloc_caches
[2] && i
== 7)
1086 new_kmalloc_cache(2, flags
);
1089 /* Kmalloc array is now usable */
1092 #ifdef CONFIG_ZONE_DMA
1093 for (i
= 0; i
<= KMALLOC_SHIFT_HIGH
; i
++) {
1094 struct kmem_cache
*s
= kmalloc_caches
[i
];
1097 int size
= kmalloc_size(i
);
1098 char *n
= kasprintf(GFP_NOWAIT
,
1099 "dma-kmalloc-%d", size
);
1102 kmalloc_dma_caches
[i
] = create_kmalloc_cache(n
,
1103 size
, SLAB_CACHE_DMA
| flags
);
1108 #endif /* !CONFIG_SLOB */
1111 * To avoid unnecessary overhead, we pass through large allocation requests
1112 * directly to the page allocator. We use __GFP_COMP, because we will need to
1113 * know the allocation order to free the pages properly in kfree.
1115 void *kmalloc_order(size_t size
, gfp_t flags
, unsigned int order
)
1120 flags
|= __GFP_COMP
;
1121 page
= alloc_pages(flags
, order
);
1122 ret
= page
? page_address(page
) : NULL
;
1123 kmemleak_alloc(ret
, size
, 1, flags
);
1124 kasan_kmalloc_large(ret
, size
, flags
);
1127 EXPORT_SYMBOL(kmalloc_order
);
1129 #ifdef CONFIG_TRACING
1130 void *kmalloc_order_trace(size_t size
, gfp_t flags
, unsigned int order
)
1132 void *ret
= kmalloc_order(size
, flags
, order
);
1133 trace_kmalloc(_RET_IP_
, ret
, size
, PAGE_SIZE
<< order
, flags
);
1136 EXPORT_SYMBOL(kmalloc_order_trace
);
1139 #ifdef CONFIG_SLAB_FREELIST_RANDOM
1140 /* Randomize a generic freelist */
1141 static void freelist_randomize(struct rnd_state
*state
, unsigned int *list
,
1147 for (i
= 0; i
< count
; i
++)
1150 /* Fisher-Yates shuffle */
1151 for (i
= count
- 1; i
> 0; i
--) {
1152 rand
= prandom_u32_state(state
);
1154 swap(list
[i
], list
[rand
]);
1158 /* Create a random sequence per cache */
1159 int cache_random_seq_create(struct kmem_cache
*cachep
, unsigned int count
,
1162 struct rnd_state state
;
1164 if (count
< 2 || cachep
->random_seq
)
1167 cachep
->random_seq
= kcalloc(count
, sizeof(unsigned int), gfp
);
1168 if (!cachep
->random_seq
)
1171 /* Get best entropy at this stage of boot */
1172 prandom_seed_state(&state
, get_random_long());
1174 freelist_randomize(&state
, cachep
->random_seq
, count
);
1178 /* Destroy the per-cache random freelist sequence */
1179 void cache_random_seq_destroy(struct kmem_cache
*cachep
)
1181 kfree(cachep
->random_seq
);
1182 cachep
->random_seq
= NULL
;
1184 #endif /* CONFIG_SLAB_FREELIST_RANDOM */
1186 #ifdef CONFIG_SLABINFO
1189 #define SLABINFO_RIGHTS (S_IWUSR | S_IRUSR)
1191 #define SLABINFO_RIGHTS S_IRUSR
1194 static void print_slabinfo_header(struct seq_file
*m
)
1197 * Output format version, so at least we can change it
1198 * without _too_ many complaints.
1200 #ifdef CONFIG_DEBUG_SLAB
1201 seq_puts(m
, "slabinfo - version: 2.1 (statistics)\n");
1203 seq_puts(m
, "slabinfo - version: 2.1\n");
1205 seq_puts(m
, "# name <active_objs> <num_objs> <objsize> <objperslab> <pagesperslab>");
1206 seq_puts(m
, " : tunables <limit> <batchcount> <sharedfactor>");
1207 seq_puts(m
, " : slabdata <active_slabs> <num_slabs> <sharedavail>");
1208 #ifdef CONFIG_DEBUG_SLAB
1209 seq_puts(m
, " : globalstat <listallocs> <maxobjs> <grown> <reaped> <error> <maxfreeable> <nodeallocs> <remotefrees> <alienoverflow>");
1210 seq_puts(m
, " : cpustat <allochit> <allocmiss> <freehit> <freemiss>");
1215 void *slab_start(struct seq_file
*m
, loff_t
*pos
)
1217 mutex_lock(&slab_mutex
);
1218 return seq_list_start(&slab_root_caches
, *pos
);
1221 void *slab_next(struct seq_file
*m
, void *p
, loff_t
*pos
)
1223 return seq_list_next(p
, &slab_root_caches
, pos
);
1226 void slab_stop(struct seq_file
*m
, void *p
)
1228 mutex_unlock(&slab_mutex
);
1232 memcg_accumulate_slabinfo(struct kmem_cache
*s
, struct slabinfo
*info
)
1234 struct kmem_cache
*c
;
1235 struct slabinfo sinfo
;
1237 if (!is_root_cache(s
))
1240 for_each_memcg_cache(c
, s
) {
1241 memset(&sinfo
, 0, sizeof(sinfo
));
1242 get_slabinfo(c
, &sinfo
);
1244 info
->active_slabs
+= sinfo
.active_slabs
;
1245 info
->num_slabs
+= sinfo
.num_slabs
;
1246 info
->shared_avail
+= sinfo
.shared_avail
;
1247 info
->active_objs
+= sinfo
.active_objs
;
1248 info
->num_objs
+= sinfo
.num_objs
;
1252 static void cache_show(struct kmem_cache
*s
, struct seq_file
*m
)
1254 struct slabinfo sinfo
;
1256 memset(&sinfo
, 0, sizeof(sinfo
));
1257 get_slabinfo(s
, &sinfo
);
1259 memcg_accumulate_slabinfo(s
, &sinfo
);
1261 seq_printf(m
, "%-17s %6lu %6lu %6u %4u %4d",
1262 cache_name(s
), sinfo
.active_objs
, sinfo
.num_objs
, s
->size
,
1263 sinfo
.objects_per_slab
, (1 << sinfo
.cache_order
));
1265 seq_printf(m
, " : tunables %4u %4u %4u",
1266 sinfo
.limit
, sinfo
.batchcount
, sinfo
.shared
);
1267 seq_printf(m
, " : slabdata %6lu %6lu %6lu",
1268 sinfo
.active_slabs
, sinfo
.num_slabs
, sinfo
.shared_avail
);
1269 slabinfo_show_stats(m
, s
);
1273 static int slab_show(struct seq_file
*m
, void *p
)
1275 struct kmem_cache
*s
= list_entry(p
, struct kmem_cache
, root_caches_node
);
1277 if (p
== slab_root_caches
.next
)
1278 print_slabinfo_header(m
);
1283 #if defined(CONFIG_MEMCG) && !defined(CONFIG_SLOB)
1284 void *memcg_slab_start(struct seq_file
*m
, loff_t
*pos
)
1286 struct mem_cgroup
*memcg
= mem_cgroup_from_css(seq_css(m
));
1288 mutex_lock(&slab_mutex
);
1289 return seq_list_start(&memcg
->kmem_caches
, *pos
);
1292 void *memcg_slab_next(struct seq_file
*m
, void *p
, loff_t
*pos
)
1294 struct mem_cgroup
*memcg
= mem_cgroup_from_css(seq_css(m
));
1296 return seq_list_next(p
, &memcg
->kmem_caches
, pos
);
1299 void memcg_slab_stop(struct seq_file
*m
, void *p
)
1301 mutex_unlock(&slab_mutex
);
1304 int memcg_slab_show(struct seq_file
*m
, void *p
)
1306 struct kmem_cache
*s
= list_entry(p
, struct kmem_cache
,
1307 memcg_params
.kmem_caches_node
);
1308 struct mem_cgroup
*memcg
= mem_cgroup_from_css(seq_css(m
));
1310 if (p
== memcg
->kmem_caches
.next
)
1311 print_slabinfo_header(m
);
1318 * slabinfo_op - iterator that generates /proc/slabinfo
1327 * num-pages-per-slab
1328 * + further values on SMP and with statistics enabled
1330 static const struct seq_operations slabinfo_op
= {
1331 .start
= slab_start
,
1337 static int slabinfo_open(struct inode
*inode
, struct file
*file
)
1339 return seq_open(file
, &slabinfo_op
);
1342 static const struct file_operations proc_slabinfo_operations
= {
1343 .open
= slabinfo_open
,
1345 .write
= slabinfo_write
,
1346 .llseek
= seq_lseek
,
1347 .release
= seq_release
,
1350 static int __init
slab_proc_init(void)
1352 proc_create("slabinfo", SLABINFO_RIGHTS
, NULL
,
1353 &proc_slabinfo_operations
);
1356 module_init(slab_proc_init
);
1357 #endif /* CONFIG_SLABINFO */
1359 static __always_inline
void *__do_krealloc(const void *p
, size_t new_size
,
1368 if (ks
>= new_size
) {
1369 kasan_krealloc((void *)p
, new_size
, flags
);
1373 ret
= kmalloc_track_caller(new_size
, flags
);
1381 * __krealloc - like krealloc() but don't free @p.
1382 * @p: object to reallocate memory for.
1383 * @new_size: how many bytes of memory are required.
1384 * @flags: the type of memory to allocate.
1386 * This function is like krealloc() except it never frees the originally
1387 * allocated buffer. Use this if you don't want to free the buffer immediately
1388 * like, for example, with RCU.
1390 void *__krealloc(const void *p
, size_t new_size
, gfp_t flags
)
1392 if (unlikely(!new_size
))
1393 return ZERO_SIZE_PTR
;
1395 return __do_krealloc(p
, new_size
, flags
);
1398 EXPORT_SYMBOL(__krealloc
);
1401 * krealloc - reallocate memory. The contents will remain unchanged.
1402 * @p: object to reallocate memory for.
1403 * @new_size: how many bytes of memory are required.
1404 * @flags: the type of memory to allocate.
1406 * The contents of the object pointed to are preserved up to the
1407 * lesser of the new and old sizes. If @p is %NULL, krealloc()
1408 * behaves exactly like kmalloc(). If @new_size is 0 and @p is not a
1409 * %NULL pointer, the object pointed to is freed.
1411 void *krealloc(const void *p
, size_t new_size
, gfp_t flags
)
1415 if (unlikely(!new_size
)) {
1417 return ZERO_SIZE_PTR
;
1420 ret
= __do_krealloc(p
, new_size
, flags
);
1421 if (ret
&& p
!= ret
)
1426 EXPORT_SYMBOL(krealloc
);
1429 * kzfree - like kfree but zero memory
1430 * @p: object to free memory of
1432 * The memory of the object @p points to is zeroed before freed.
1433 * If @p is %NULL, kzfree() does nothing.
1435 * Note: this function zeroes the whole allocated buffer which can be a good
1436 * deal bigger than the requested buffer size passed to kmalloc(). So be
1437 * careful when using this function in performance sensitive code.
1439 void kzfree(const void *p
)
1442 void *mem
= (void *)p
;
1444 if (unlikely(ZERO_OR_NULL_PTR(mem
)))
1450 EXPORT_SYMBOL(kzfree
);
1452 /* Tracepoints definitions. */
1453 EXPORT_TRACEPOINT_SYMBOL(kmalloc
);
1454 EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc
);
1455 EXPORT_TRACEPOINT_SYMBOL(kmalloc_node
);
1456 EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc_node
);
1457 EXPORT_TRACEPOINT_SYMBOL(kfree
);
1458 EXPORT_TRACEPOINT_SYMBOL(kmem_cache_free
);