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;
34 * Set of flags that will prevent slab merging
36 #define SLAB_NEVER_MERGE (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
37 SLAB_TRACE | SLAB_DESTROY_BY_RCU | SLAB_NOLEAKTRACE | \
38 SLAB_FAILSLAB | SLAB_KASAN)
40 #define SLAB_MERGE_SAME (SLAB_RECLAIM_ACCOUNT | SLAB_CACHE_DMA | \
41 SLAB_NOTRACK | SLAB_ACCOUNT)
44 * Merge control. If this is set then no merging of slab caches will occur.
45 * (Could be removed. This was introduced to pacify the merge skeptics.)
47 static int slab_nomerge;
49 static int __init setup_slab_nomerge(char *str)
56 __setup_param("slub_nomerge", slub_nomerge, setup_slab_nomerge, 0);
59 __setup("slab_nomerge", setup_slab_nomerge);
62 * Determine the size of a slab object
64 unsigned int kmem_cache_size(struct kmem_cache *s)
66 return s->object_size;
68 EXPORT_SYMBOL(kmem_cache_size);
70 #ifdef CONFIG_DEBUG_VM
71 static int kmem_cache_sanity_check(const char *name, size_t size)
73 struct kmem_cache *s = NULL;
75 if (!name || in_interrupt() || size < sizeof(void *) ||
76 size > KMALLOC_MAX_SIZE) {
77 pr_err("kmem_cache_create(%s) integrity check failed\n", name);
81 list_for_each_entry(s, &slab_caches, list) {
86 * This happens when the module gets unloaded and doesn't
87 * destroy its slab cache and no-one else reuses the vmalloc
88 * area of the module. Print a warning.
90 res = probe_kernel_address(s->name, tmp);
92 pr_err("Slab cache with size %d has lost its name\n",
98 WARN_ON(strchr(name, ' ')); /* It confuses parsers */
102 static inline int kmem_cache_sanity_check(const char *name, size_t size)
108 void __kmem_cache_free_bulk(struct kmem_cache *s, size_t nr, void **p)
112 for (i = 0; i < nr; i++) {
114 kmem_cache_free(s, p[i]);
120 int __kmem_cache_alloc_bulk(struct kmem_cache *s, gfp_t flags, size_t nr,
125 for (i = 0; i < nr; i++) {
126 void *x = p[i] = kmem_cache_alloc(s, flags);
128 __kmem_cache_free_bulk(s, i, p);
135 #if defined(CONFIG_MEMCG) && !defined(CONFIG_SLOB)
136 void slab_init_memcg_params(struct kmem_cache *s)
138 s->memcg_params.is_root_cache = true;
139 INIT_LIST_HEAD(&s->memcg_params.list);
140 RCU_INIT_POINTER(s->memcg_params.memcg_caches, NULL);
143 static int init_memcg_params(struct kmem_cache *s,
144 struct mem_cgroup *memcg, struct kmem_cache *root_cache)
146 struct memcg_cache_array *arr;
149 s->memcg_params.is_root_cache = false;
150 s->memcg_params.memcg = memcg;
151 s->memcg_params.root_cache = root_cache;
155 slab_init_memcg_params(s);
157 if (!memcg_nr_cache_ids)
160 arr = kzalloc(sizeof(struct memcg_cache_array) +
161 memcg_nr_cache_ids * sizeof(void *),
166 RCU_INIT_POINTER(s->memcg_params.memcg_caches, arr);
170 static void destroy_memcg_params(struct kmem_cache *s)
172 if (is_root_cache(s))
173 kfree(rcu_access_pointer(s->memcg_params.memcg_caches));
176 static int update_memcg_params(struct kmem_cache *s, int new_array_size)
178 struct memcg_cache_array *old, *new;
180 if (!is_root_cache(s))
183 new = kzalloc(sizeof(struct memcg_cache_array) +
184 new_array_size * sizeof(void *), GFP_KERNEL);
188 old = rcu_dereference_protected(s->memcg_params.memcg_caches,
189 lockdep_is_held(&slab_mutex));
191 memcpy(new->entries, old->entries,
192 memcg_nr_cache_ids * sizeof(void *));
194 rcu_assign_pointer(s->memcg_params.memcg_caches, new);
200 int memcg_update_all_caches(int num_memcgs)
202 struct kmem_cache *s;
205 mutex_lock(&slab_mutex);
206 list_for_each_entry(s, &slab_caches, list) {
207 ret = update_memcg_params(s, num_memcgs);
209 * Instead of freeing the memory, we'll just leave the caches
210 * up to this point in an updated state.
215 mutex_unlock(&slab_mutex);
219 static inline int init_memcg_params(struct kmem_cache *s,
220 struct mem_cgroup *memcg, struct kmem_cache *root_cache)
225 static inline void destroy_memcg_params(struct kmem_cache *s)
228 #endif /* CONFIG_MEMCG && !CONFIG_SLOB */
231 * Find a mergeable slab cache
233 int slab_unmergeable(struct kmem_cache *s)
235 if (slab_nomerge || (s->flags & SLAB_NEVER_MERGE))
238 if (!is_root_cache(s))
245 * We may have set a slab to be unmergeable during bootstrap.
253 struct kmem_cache *find_mergeable(size_t size, size_t align,
254 unsigned long flags, const char *name, void (*ctor)(void *))
256 struct kmem_cache *s;
258 if (slab_nomerge || (flags & SLAB_NEVER_MERGE))
264 size = ALIGN(size, sizeof(void *));
265 align = calculate_alignment(flags, align, size);
266 size = ALIGN(size, align);
267 flags = kmem_cache_flags(size, flags, name, NULL);
269 list_for_each_entry_reverse(s, &slab_caches, list) {
270 if (slab_unmergeable(s))
276 if ((flags & SLAB_MERGE_SAME) != (s->flags & SLAB_MERGE_SAME))
279 * Check if alignment is compatible.
280 * Courtesy of Adrian Drzewiecki
282 if ((s->size & ~(align - 1)) != s->size)
285 if (s->size - size >= sizeof(void *))
288 if (IS_ENABLED(CONFIG_SLAB) && align &&
289 (align > s->align || s->align % align))
298 * Figure out what the alignment of the objects will be given a set of
299 * flags, a user specified alignment and the size of the objects.
301 unsigned long calculate_alignment(unsigned long flags,
302 unsigned long align, unsigned long size)
305 * If the user wants hardware cache aligned objects then follow that
306 * suggestion if the object is sufficiently large.
308 * The hardware cache alignment cannot override the specified
309 * alignment though. If that is greater then use it.
311 if (flags & SLAB_HWCACHE_ALIGN) {
312 unsigned long ralign = cache_line_size();
313 while (size <= ralign / 2)
315 align = max(align, ralign);
318 if (align < ARCH_SLAB_MINALIGN)
319 align = ARCH_SLAB_MINALIGN;
321 return ALIGN(align, sizeof(void *));
324 static struct kmem_cache *create_cache(const char *name,
325 size_t object_size, size_t size, size_t align,
326 unsigned long flags, void (*ctor)(void *),
327 struct mem_cgroup *memcg, struct kmem_cache *root_cache)
329 struct kmem_cache *s;
333 s = kmem_cache_zalloc(kmem_cache, GFP_KERNEL);
338 s->object_size = object_size;
343 err = init_memcg_params(s, memcg, root_cache);
347 err = __kmem_cache_create(s, flags);
352 list_add(&s->list, &slab_caches);
359 destroy_memcg_params(s);
360 kmem_cache_free(kmem_cache, s);
365 * kmem_cache_create - Create a cache.
366 * @name: A string which is used in /proc/slabinfo to identify this cache.
367 * @size: The size of objects to be created in this cache.
368 * @align: The required alignment for the objects.
370 * @ctor: A constructor for the objects.
372 * Returns a ptr to the cache on success, NULL on failure.
373 * Cannot be called within a interrupt, but can be interrupted.
374 * The @ctor is run when new pages are allocated by the cache.
378 * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5)
379 * to catch references to uninitialised memory.
381 * %SLAB_RED_ZONE - Insert `Red' zones around the allocated memory to check
382 * for buffer overruns.
384 * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware
385 * cacheline. This can be beneficial if you're counting cycles as closely
389 kmem_cache_create(const char *name, size_t size, size_t align,
390 unsigned long flags, void (*ctor)(void *))
392 struct kmem_cache *s = NULL;
393 const char *cache_name;
398 memcg_get_cache_ids();
400 mutex_lock(&slab_mutex);
402 err = kmem_cache_sanity_check(name, size);
408 * Some allocators will constraint the set of valid flags to a subset
409 * of all flags. We expect them to define CACHE_CREATE_MASK in this
410 * case, and we'll just provide them with a sanitized version of the
413 flags &= CACHE_CREATE_MASK;
415 s = __kmem_cache_alias(name, size, align, flags, ctor);
419 cache_name = kstrdup_const(name, GFP_KERNEL);
425 s = create_cache(cache_name, size, size,
426 calculate_alignment(flags, align, size),
427 flags, ctor, NULL, NULL);
430 kfree_const(cache_name);
434 mutex_unlock(&slab_mutex);
436 memcg_put_cache_ids();
441 if (flags & SLAB_PANIC)
442 panic("kmem_cache_create: Failed to create slab '%s'. Error %d\n",
445 pr_warn("kmem_cache_create(%s) failed with error %d\n",
453 EXPORT_SYMBOL(kmem_cache_create);
455 static int shutdown_cache(struct kmem_cache *s,
456 struct list_head *release, bool *need_rcu_barrier)
458 if (__kmem_cache_shutdown(s) != 0)
461 if (s->flags & SLAB_DESTROY_BY_RCU)
462 *need_rcu_barrier = true;
464 list_move(&s->list, release);
468 static void release_caches(struct list_head *release, bool need_rcu_barrier)
470 struct kmem_cache *s, *s2;
472 if (need_rcu_barrier)
475 list_for_each_entry_safe(s, s2, release, list) {
476 #ifdef SLAB_SUPPORTS_SYSFS
477 sysfs_slab_remove(s);
479 slab_kmem_cache_release(s);
484 #if defined(CONFIG_MEMCG) && !defined(CONFIG_SLOB)
486 * memcg_create_kmem_cache - Create a cache for a memory cgroup.
487 * @memcg: The memory cgroup the new cache is for.
488 * @root_cache: The parent of the new cache.
490 * This function attempts to create a kmem cache that will serve allocation
491 * requests going from @memcg to @root_cache. The new cache inherits properties
494 void memcg_create_kmem_cache(struct mem_cgroup *memcg,
495 struct kmem_cache *root_cache)
497 static char memcg_name_buf[NAME_MAX + 1]; /* protected by slab_mutex */
498 struct cgroup_subsys_state *css = &memcg->css;
499 struct memcg_cache_array *arr;
500 struct kmem_cache *s = NULL;
507 mutex_lock(&slab_mutex);
510 * The memory cgroup could have been offlined while the cache
511 * creation work was pending.
513 if (memcg->kmem_state != KMEM_ONLINE)
516 idx = memcg_cache_id(memcg);
517 arr = rcu_dereference_protected(root_cache->memcg_params.memcg_caches,
518 lockdep_is_held(&slab_mutex));
521 * Since per-memcg caches are created asynchronously on first
522 * allocation (see memcg_kmem_get_cache()), several threads can try to
523 * create the same cache, but only one of them may succeed.
525 if (arr->entries[idx])
528 cgroup_name(css->cgroup, memcg_name_buf, sizeof(memcg_name_buf));
529 cache_name = kasprintf(GFP_KERNEL, "%s(%llu:%s)", root_cache->name,
530 css->serial_nr, memcg_name_buf);
534 s = create_cache(cache_name, root_cache->object_size,
535 root_cache->size, root_cache->align,
536 root_cache->flags, root_cache->ctor,
539 * If we could not create a memcg cache, do not complain, because
540 * that's not critical at all as we can always proceed with the root
548 list_add(&s->memcg_params.list, &root_cache->memcg_params.list);
551 * Since readers won't lock (see cache_from_memcg_idx()), we need a
552 * barrier here to ensure nobody will see the kmem_cache partially
556 arr->entries[idx] = s;
559 mutex_unlock(&slab_mutex);
565 void memcg_deactivate_kmem_caches(struct mem_cgroup *memcg)
568 struct memcg_cache_array *arr;
569 struct kmem_cache *s, *c;
571 idx = memcg_cache_id(memcg);
576 mutex_lock(&slab_mutex);
577 list_for_each_entry(s, &slab_caches, list) {
578 if (!is_root_cache(s))
581 arr = rcu_dereference_protected(s->memcg_params.memcg_caches,
582 lockdep_is_held(&slab_mutex));
583 c = arr->entries[idx];
587 __kmem_cache_shrink(c, true);
588 arr->entries[idx] = NULL;
590 mutex_unlock(&slab_mutex);
596 static int __shutdown_memcg_cache(struct kmem_cache *s,
597 struct list_head *release, bool *need_rcu_barrier)
599 BUG_ON(is_root_cache(s));
601 if (shutdown_cache(s, release, need_rcu_barrier))
604 list_del(&s->memcg_params.list);
608 void memcg_destroy_kmem_caches(struct mem_cgroup *memcg)
611 bool need_rcu_barrier = false;
612 struct kmem_cache *s, *s2;
617 mutex_lock(&slab_mutex);
618 list_for_each_entry_safe(s, s2, &slab_caches, list) {
619 if (is_root_cache(s) || s->memcg_params.memcg != memcg)
622 * The cgroup is about to be freed and therefore has no charges
623 * left. Hence, all its caches must be empty by now.
625 BUG_ON(__shutdown_memcg_cache(s, &release, &need_rcu_barrier));
627 mutex_unlock(&slab_mutex);
632 release_caches(&release, need_rcu_barrier);
635 static int shutdown_memcg_caches(struct kmem_cache *s,
636 struct list_head *release, bool *need_rcu_barrier)
638 struct memcg_cache_array *arr;
639 struct kmem_cache *c, *c2;
643 BUG_ON(!is_root_cache(s));
646 * First, shutdown active caches, i.e. caches that belong to online
649 arr = rcu_dereference_protected(s->memcg_params.memcg_caches,
650 lockdep_is_held(&slab_mutex));
651 for_each_memcg_cache_index(i) {
655 if (__shutdown_memcg_cache(c, release, need_rcu_barrier))
657 * The cache still has objects. Move it to a temporary
658 * list so as not to try to destroy it for a second
659 * time while iterating over inactive caches below.
661 list_move(&c->memcg_params.list, &busy);
664 * The cache is empty and will be destroyed soon. Clear
665 * the pointer to it in the memcg_caches array so that
666 * it will never be accessed even if the root cache
669 arr->entries[i] = NULL;
673 * Second, shutdown all caches left from memory cgroups that are now
676 list_for_each_entry_safe(c, c2, &s->memcg_params.list,
678 __shutdown_memcg_cache(c, release, need_rcu_barrier);
680 list_splice(&busy, &s->memcg_params.list);
683 * A cache being destroyed must be empty. In particular, this means
684 * that all per memcg caches attached to it must be empty too.
686 if (!list_empty(&s->memcg_params.list))
691 static inline int shutdown_memcg_caches(struct kmem_cache *s,
692 struct list_head *release, bool *need_rcu_barrier)
696 #endif /* CONFIG_MEMCG && !CONFIG_SLOB */
698 void slab_kmem_cache_release(struct kmem_cache *s)
700 __kmem_cache_release(s);
701 destroy_memcg_params(s);
702 kfree_const(s->name);
703 kmem_cache_free(kmem_cache, s);
706 void kmem_cache_destroy(struct kmem_cache *s)
709 bool need_rcu_barrier = false;
718 kasan_cache_destroy(s);
719 mutex_lock(&slab_mutex);
725 err = shutdown_memcg_caches(s, &release, &need_rcu_barrier);
727 err = shutdown_cache(s, &release, &need_rcu_barrier);
730 pr_err("kmem_cache_destroy %s: Slab cache still has objects\n",
735 mutex_unlock(&slab_mutex);
740 release_caches(&release, need_rcu_barrier);
742 EXPORT_SYMBOL(kmem_cache_destroy);
745 * kmem_cache_shrink - Shrink a cache.
746 * @cachep: The cache to shrink.
748 * Releases as many slabs as possible for a cache.
749 * To help debugging, a zero exit status indicates all slabs were released.
751 int kmem_cache_shrink(struct kmem_cache *cachep)
757 kasan_cache_shrink(cachep);
758 ret = __kmem_cache_shrink(cachep, false);
763 EXPORT_SYMBOL(kmem_cache_shrink);
765 bool slab_is_available(void)
767 return slab_state >= UP;
771 /* Create a cache during boot when no slab services are available yet */
772 void __init create_boot_cache(struct kmem_cache *s, const char *name, size_t size,
778 s->size = s->object_size = size;
779 s->align = calculate_alignment(flags, ARCH_KMALLOC_MINALIGN, size);
781 slab_init_memcg_params(s);
783 err = __kmem_cache_create(s, flags);
786 panic("Creation of kmalloc slab %s size=%zu failed. Reason %d\n",
789 s->refcount = -1; /* Exempt from merging for now */
792 struct kmem_cache *__init create_kmalloc_cache(const char *name, size_t size,
795 struct kmem_cache *s = kmem_cache_zalloc(kmem_cache, GFP_NOWAIT);
798 panic("Out of memory when creating slab %s\n", name);
800 create_boot_cache(s, name, size, flags);
801 list_add(&s->list, &slab_caches);
806 struct kmem_cache *kmalloc_caches[KMALLOC_SHIFT_HIGH + 1];
807 EXPORT_SYMBOL(kmalloc_caches);
809 #ifdef CONFIG_ZONE_DMA
810 struct kmem_cache *kmalloc_dma_caches[KMALLOC_SHIFT_HIGH + 1];
811 EXPORT_SYMBOL(kmalloc_dma_caches);
815 * Conversion table for small slabs sizes / 8 to the index in the
816 * kmalloc array. This is necessary for slabs < 192 since we have non power
817 * of two cache sizes there. The size of larger slabs can be determined using
820 static s8 size_index[24] = {
847 static inline int size_index_elem(size_t bytes)
849 return (bytes - 1) / 8;
853 * Find the kmem_cache structure that serves a given size of
856 struct kmem_cache *kmalloc_slab(size_t size, gfp_t flags)
860 if (unlikely(size > KMALLOC_MAX_SIZE)) {
861 WARN_ON_ONCE(!(flags & __GFP_NOWARN));
867 return ZERO_SIZE_PTR;
869 index = size_index[size_index_elem(size)];
871 index = fls(size - 1);
873 #ifdef CONFIG_ZONE_DMA
874 if (unlikely((flags & GFP_DMA)))
875 return kmalloc_dma_caches[index];
878 return kmalloc_caches[index];
882 * kmalloc_info[] is to make slub_debug=,kmalloc-xx option work at boot time.
883 * kmalloc_index() supports up to 2^26=64MB, so the final entry of the table is
889 } const kmalloc_info[] __initconst = {
890 {NULL, 0}, {"kmalloc-96", 96},
891 {"kmalloc-192", 192}, {"kmalloc-8", 8},
892 {"kmalloc-16", 16}, {"kmalloc-32", 32},
893 {"kmalloc-64", 64}, {"kmalloc-128", 128},
894 {"kmalloc-256", 256}, {"kmalloc-512", 512},
895 {"kmalloc-1024", 1024}, {"kmalloc-2048", 2048},
896 {"kmalloc-4096", 4096}, {"kmalloc-8192", 8192},
897 {"kmalloc-16384", 16384}, {"kmalloc-32768", 32768},
898 {"kmalloc-65536", 65536}, {"kmalloc-131072", 131072},
899 {"kmalloc-262144", 262144}, {"kmalloc-524288", 524288},
900 {"kmalloc-1048576", 1048576}, {"kmalloc-2097152", 2097152},
901 {"kmalloc-4194304", 4194304}, {"kmalloc-8388608", 8388608},
902 {"kmalloc-16777216", 16777216}, {"kmalloc-33554432", 33554432},
903 {"kmalloc-67108864", 67108864}
907 * Patch up the size_index table if we have strange large alignment
908 * requirements for the kmalloc array. This is only the case for
909 * MIPS it seems. The standard arches will not generate any code here.
911 * Largest permitted alignment is 256 bytes due to the way we
912 * handle the index determination for the smaller caches.
914 * Make sure that nothing crazy happens if someone starts tinkering
915 * around with ARCH_KMALLOC_MINALIGN
917 void __init setup_kmalloc_cache_index_table(void)
921 BUILD_BUG_ON(KMALLOC_MIN_SIZE > 256 ||
922 (KMALLOC_MIN_SIZE & (KMALLOC_MIN_SIZE - 1)));
924 for (i = 8; i < KMALLOC_MIN_SIZE; i += 8) {
925 int elem = size_index_elem(i);
927 if (elem >= ARRAY_SIZE(size_index))
929 size_index[elem] = KMALLOC_SHIFT_LOW;
932 if (KMALLOC_MIN_SIZE >= 64) {
934 * The 96 byte size cache is not used if the alignment
937 for (i = 64 + 8; i <= 96; i += 8)
938 size_index[size_index_elem(i)] = 7;
942 if (KMALLOC_MIN_SIZE >= 128) {
944 * The 192 byte sized cache is not used if the alignment
945 * is 128 byte. Redirect kmalloc to use the 256 byte cache
948 for (i = 128 + 8; i <= 192; i += 8)
949 size_index[size_index_elem(i)] = 8;
953 static void __init new_kmalloc_cache(int idx, unsigned long flags)
955 kmalloc_caches[idx] = create_kmalloc_cache(kmalloc_info[idx].name,
956 kmalloc_info[idx].size, flags);
960 * Create the kmalloc array. Some of the regular kmalloc arrays
961 * may already have been created because they were needed to
962 * enable allocations for slab creation.
964 void __init create_kmalloc_caches(unsigned long flags)
968 for (i = KMALLOC_SHIFT_LOW; i <= KMALLOC_SHIFT_HIGH; i++) {
969 if (!kmalloc_caches[i])
970 new_kmalloc_cache(i, flags);
973 * Caches that are not of the two-to-the-power-of size.
974 * These have to be created immediately after the
975 * earlier power of two caches
977 if (KMALLOC_MIN_SIZE <= 32 && !kmalloc_caches[1] && i == 6)
978 new_kmalloc_cache(1, flags);
979 if (KMALLOC_MIN_SIZE <= 64 && !kmalloc_caches[2] && i == 7)
980 new_kmalloc_cache(2, flags);
983 /* Kmalloc array is now usable */
986 #ifdef CONFIG_ZONE_DMA
987 for (i = 0; i <= KMALLOC_SHIFT_HIGH; i++) {
988 struct kmem_cache *s = kmalloc_caches[i];
991 int size = kmalloc_size(i);
992 char *n = kasprintf(GFP_NOWAIT,
993 "dma-kmalloc-%d", size);
996 kmalloc_dma_caches[i] = create_kmalloc_cache(n,
997 size, SLAB_CACHE_DMA | flags);
1002 #endif /* !CONFIG_SLOB */
1005 * To avoid unnecessary overhead, we pass through large allocation requests
1006 * directly to the page allocator. We use __GFP_COMP, because we will need to
1007 * know the allocation order to free the pages properly in kfree.
1009 void *kmalloc_order(size_t size, gfp_t flags, unsigned int order)
1014 flags |= __GFP_COMP;
1015 page = alloc_pages(flags, order);
1016 ret = page ? page_address(page) : NULL;
1017 kmemleak_alloc(ret, size, 1, flags);
1018 kasan_kmalloc_large(ret, size, flags);
1021 EXPORT_SYMBOL(kmalloc_order);
1023 #ifdef CONFIG_TRACING
1024 void *kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order)
1026 void *ret = kmalloc_order(size, flags, order);
1027 trace_kmalloc(_RET_IP_, ret, size, PAGE_SIZE << order, flags);
1030 EXPORT_SYMBOL(kmalloc_order_trace);
1033 #ifdef CONFIG_SLAB_FREELIST_RANDOM
1034 /* Randomize a generic freelist */
1035 static void freelist_randomize(struct rnd_state *state, unsigned int *list,
1041 for (i = 0; i < count; i++)
1044 /* Fisher-Yates shuffle */
1045 for (i = count - 1; i > 0; i--) {
1046 rand = prandom_u32_state(state);
1048 swap(list[i], list[rand]);
1052 /* Create a random sequence per cache */
1053 int cache_random_seq_create(struct kmem_cache *cachep, unsigned int count,
1056 struct rnd_state state;
1058 if (count < 2 || cachep->random_seq)
1061 cachep->random_seq = kcalloc(count, sizeof(unsigned int), gfp);
1062 if (!cachep->random_seq)
1065 /* Get best entropy at this stage of boot */
1066 prandom_seed_state(&state, get_random_long());
1068 freelist_randomize(&state, cachep->random_seq, count);
1072 /* Destroy the per-cache random freelist sequence */
1073 void cache_random_seq_destroy(struct kmem_cache *cachep)
1075 kfree(cachep->random_seq);
1076 cachep->random_seq = NULL;
1078 #endif /* CONFIG_SLAB_FREELIST_RANDOM */
1080 #ifdef CONFIG_SLABINFO
1083 #define SLABINFO_RIGHTS (S_IWUSR | S_IRUSR)
1085 #define SLABINFO_RIGHTS S_IRUSR
1088 static void print_slabinfo_header(struct seq_file *m)
1091 * Output format version, so at least we can change it
1092 * without _too_ many complaints.
1094 #ifdef CONFIG_DEBUG_SLAB
1095 seq_puts(m, "slabinfo - version: 2.1 (statistics)\n");
1097 seq_puts(m, "slabinfo - version: 2.1\n");
1099 seq_puts(m, "# name <active_objs> <num_objs> <objsize> <objperslab> <pagesperslab>");
1100 seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>");
1101 seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>");
1102 #ifdef CONFIG_DEBUG_SLAB
1103 seq_puts(m, " : globalstat <listallocs> <maxobjs> <grown> <reaped> <error> <maxfreeable> <nodeallocs> <remotefrees> <alienoverflow>");
1104 seq_puts(m, " : cpustat <allochit> <allocmiss> <freehit> <freemiss>");
1109 void *slab_start(struct seq_file *m, loff_t *pos)
1111 mutex_lock(&slab_mutex);
1112 return seq_list_start(&slab_caches, *pos);
1115 void *slab_next(struct seq_file *m, void *p, loff_t *pos)
1117 return seq_list_next(p, &slab_caches, pos);
1120 void slab_stop(struct seq_file *m, void *p)
1122 mutex_unlock(&slab_mutex);
1126 memcg_accumulate_slabinfo(struct kmem_cache *s, struct slabinfo *info)
1128 struct kmem_cache *c;
1129 struct slabinfo sinfo;
1131 if (!is_root_cache(s))
1134 for_each_memcg_cache(c, s) {
1135 memset(&sinfo, 0, sizeof(sinfo));
1136 get_slabinfo(c, &sinfo);
1138 info->active_slabs += sinfo.active_slabs;
1139 info->num_slabs += sinfo.num_slabs;
1140 info->shared_avail += sinfo.shared_avail;
1141 info->active_objs += sinfo.active_objs;
1142 info->num_objs += sinfo.num_objs;
1146 static void cache_show(struct kmem_cache *s, struct seq_file *m)
1148 struct slabinfo sinfo;
1150 memset(&sinfo, 0, sizeof(sinfo));
1151 get_slabinfo(s, &sinfo);
1153 memcg_accumulate_slabinfo(s, &sinfo);
1155 seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d",
1156 cache_name(s), sinfo.active_objs, sinfo.num_objs, s->size,
1157 sinfo.objects_per_slab, (1 << sinfo.cache_order));
1159 seq_printf(m, " : tunables %4u %4u %4u",
1160 sinfo.limit, sinfo.batchcount, sinfo.shared);
1161 seq_printf(m, " : slabdata %6lu %6lu %6lu",
1162 sinfo.active_slabs, sinfo.num_slabs, sinfo.shared_avail);
1163 slabinfo_show_stats(m, s);
1167 static int slab_show(struct seq_file *m, void *p)
1169 struct kmem_cache *s = list_entry(p, struct kmem_cache, list);
1171 if (p == slab_caches.next)
1172 print_slabinfo_header(m);
1173 if (is_root_cache(s))
1178 #if defined(CONFIG_MEMCG) && !defined(CONFIG_SLOB)
1179 int memcg_slab_show(struct seq_file *m, void *p)
1181 struct kmem_cache *s = list_entry(p, struct kmem_cache, list);
1182 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
1184 if (p == slab_caches.next)
1185 print_slabinfo_header(m);
1186 if (!is_root_cache(s) && s->memcg_params.memcg == memcg)
1193 * slabinfo_op - iterator that generates /proc/slabinfo
1202 * num-pages-per-slab
1203 * + further values on SMP and with statistics enabled
1205 static const struct seq_operations slabinfo_op = {
1206 .start = slab_start,
1212 static int slabinfo_open(struct inode *inode, struct file *file)
1214 return seq_open(file, &slabinfo_op);
1217 static const struct file_operations proc_slabinfo_operations = {
1218 .open = slabinfo_open,
1220 .write = slabinfo_write,
1221 .llseek = seq_lseek,
1222 .release = seq_release,
1225 static int __init slab_proc_init(void)
1227 proc_create("slabinfo", SLABINFO_RIGHTS, NULL,
1228 &proc_slabinfo_operations);
1231 module_init(slab_proc_init);
1232 #endif /* CONFIG_SLABINFO */
1234 static __always_inline void *__do_krealloc(const void *p, size_t new_size,
1243 if (ks >= new_size) {
1244 kasan_krealloc((void *)p, new_size, flags);
1248 ret = kmalloc_track_caller(new_size, flags);
1256 * __krealloc - like krealloc() but don't free @p.
1257 * @p: object to reallocate memory for.
1258 * @new_size: how many bytes of memory are required.
1259 * @flags: the type of memory to allocate.
1261 * This function is like krealloc() except it never frees the originally
1262 * allocated buffer. Use this if you don't want to free the buffer immediately
1263 * like, for example, with RCU.
1265 void *__krealloc(const void *p, size_t new_size, gfp_t flags)
1267 if (unlikely(!new_size))
1268 return ZERO_SIZE_PTR;
1270 return __do_krealloc(p, new_size, flags);
1273 EXPORT_SYMBOL(__krealloc);
1276 * krealloc - reallocate memory. The contents will remain unchanged.
1277 * @p: object to reallocate memory for.
1278 * @new_size: how many bytes of memory are required.
1279 * @flags: the type of memory to allocate.
1281 * The contents of the object pointed to are preserved up to the
1282 * lesser of the new and old sizes. If @p is %NULL, krealloc()
1283 * behaves exactly like kmalloc(). If @new_size is 0 and @p is not a
1284 * %NULL pointer, the object pointed to is freed.
1286 void *krealloc(const void *p, size_t new_size, gfp_t flags)
1290 if (unlikely(!new_size)) {
1292 return ZERO_SIZE_PTR;
1295 ret = __do_krealloc(p, new_size, flags);
1296 if (ret && p != ret)
1301 EXPORT_SYMBOL(krealloc);
1304 * kzfree - like kfree but zero memory
1305 * @p: object to free memory of
1307 * The memory of the object @p points to is zeroed before freed.
1308 * If @p is %NULL, kzfree() does nothing.
1310 * Note: this function zeroes the whole allocated buffer which can be a good
1311 * deal bigger than the requested buffer size passed to kmalloc(). So be
1312 * careful when using this function in performance sensitive code.
1314 void kzfree(const void *p)
1317 void *mem = (void *)p;
1319 if (unlikely(ZERO_OR_NULL_PTR(mem)))
1325 EXPORT_SYMBOL(kzfree);
1327 /* Tracepoints definitions. */
1328 EXPORT_TRACEPOINT_SYMBOL(kmalloc);
1329 EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc);
1330 EXPORT_TRACEPOINT_SYMBOL(kmalloc_node);
1331 EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc_node);
1332 EXPORT_TRACEPOINT_SYMBOL(kfree);
1333 EXPORT_TRACEPOINT_SYMBOL(kmem_cache_free);