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_DESTROY_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.
50 * (Could be removed. This was introduced to pacify the merge skeptics.)
52 static int slab_nomerge;
54 static int __init setup_slab_nomerge(char *str)
61 __setup_param("slub_nomerge", slub_nomerge, setup_slab_nomerge, 0);
64 __setup("slab_nomerge", setup_slab_nomerge);
67 * Determine the size of a slab object
69 unsigned int kmem_cache_size(struct kmem_cache *s)
71 return s->object_size;
73 EXPORT_SYMBOL(kmem_cache_size);
75 #ifdef CONFIG_DEBUG_VM
76 static int kmem_cache_sanity_check(const char *name, size_t size)
78 struct kmem_cache *s = NULL;
80 if (!name || in_interrupt() || size < sizeof(void *) ||
81 size > KMALLOC_MAX_SIZE) {
82 pr_err("kmem_cache_create(%s) integrity check failed\n", name);
86 list_for_each_entry(s, &slab_caches, list) {
91 * This happens when the module gets unloaded and doesn't
92 * destroy its slab cache and no-one else reuses the vmalloc
93 * area of the module. Print a warning.
95 res = probe_kernel_address(s->name, tmp);
97 pr_err("Slab cache with size %d has lost its name\n",
103 WARN_ON(strchr(name, ' ')); /* It confuses parsers */
107 static inline int kmem_cache_sanity_check(const char *name, size_t size)
113 void __kmem_cache_free_bulk(struct kmem_cache *s, size_t nr, void **p)
117 for (i = 0; i < nr; i++) {
119 kmem_cache_free(s, p[i]);
125 int __kmem_cache_alloc_bulk(struct kmem_cache *s, gfp_t flags, size_t nr,
130 for (i = 0; i < nr; i++) {
131 void *x = p[i] = kmem_cache_alloc(s, flags);
133 __kmem_cache_free_bulk(s, i, p);
140 #if defined(CONFIG_MEMCG) && !defined(CONFIG_SLOB)
141 void slab_init_memcg_params(struct kmem_cache *s)
143 s->memcg_params.is_root_cache = true;
144 INIT_LIST_HEAD(&s->memcg_params.list);
145 RCU_INIT_POINTER(s->memcg_params.memcg_caches, NULL);
148 static int init_memcg_params(struct kmem_cache *s,
149 struct mem_cgroup *memcg, struct kmem_cache *root_cache)
151 struct memcg_cache_array *arr;
154 s->memcg_params.is_root_cache = false;
155 s->memcg_params.memcg = memcg;
156 s->memcg_params.root_cache = root_cache;
160 slab_init_memcg_params(s);
162 if (!memcg_nr_cache_ids)
165 arr = kzalloc(sizeof(struct memcg_cache_array) +
166 memcg_nr_cache_ids * sizeof(void *),
171 RCU_INIT_POINTER(s->memcg_params.memcg_caches, arr);
175 static void destroy_memcg_params(struct kmem_cache *s)
177 if (is_root_cache(s))
178 kfree(rcu_access_pointer(s->memcg_params.memcg_caches));
181 static int update_memcg_params(struct kmem_cache *s, int new_array_size)
183 struct memcg_cache_array *old, *new;
185 if (!is_root_cache(s))
188 new = kzalloc(sizeof(struct memcg_cache_array) +
189 new_array_size * sizeof(void *), GFP_KERNEL);
193 old = rcu_dereference_protected(s->memcg_params.memcg_caches,
194 lockdep_is_held(&slab_mutex));
196 memcpy(new->entries, old->entries,
197 memcg_nr_cache_ids * sizeof(void *));
199 rcu_assign_pointer(s->memcg_params.memcg_caches, new);
205 int memcg_update_all_caches(int num_memcgs)
207 struct kmem_cache *s;
210 mutex_lock(&slab_mutex);
211 list_for_each_entry(s, &slab_caches, list) {
212 ret = update_memcg_params(s, num_memcgs);
214 * Instead of freeing the memory, we'll just leave the caches
215 * up to this point in an updated state.
220 mutex_unlock(&slab_mutex);
224 static void unlink_memcg_cache(struct kmem_cache *s)
226 list_del(&s->memcg_params.list);
229 static inline int init_memcg_params(struct kmem_cache *s,
230 struct mem_cgroup *memcg, struct kmem_cache *root_cache)
235 static inline void destroy_memcg_params(struct kmem_cache *s)
239 static inline void unlink_memcg_cache(struct kmem_cache *s)
242 #endif /* CONFIG_MEMCG && !CONFIG_SLOB */
245 * Find a mergeable slab cache
247 int slab_unmergeable(struct kmem_cache *s)
249 if (slab_nomerge || (s->flags & SLAB_NEVER_MERGE))
252 if (!is_root_cache(s))
259 * We may have set a slab to be unmergeable during bootstrap.
267 struct kmem_cache *find_mergeable(size_t size, size_t align,
268 unsigned long flags, const char *name, void (*ctor)(void *))
270 struct kmem_cache *s;
278 size = ALIGN(size, sizeof(void *));
279 align = calculate_alignment(flags, align, size);
280 size = ALIGN(size, align);
281 flags = kmem_cache_flags(size, flags, name, NULL);
283 if (flags & SLAB_NEVER_MERGE)
286 list_for_each_entry_reverse(s, &slab_caches, list) {
287 if (slab_unmergeable(s))
293 if ((flags & SLAB_MERGE_SAME) != (s->flags & SLAB_MERGE_SAME))
296 * Check if alignment is compatible.
297 * Courtesy of Adrian Drzewiecki
299 if ((s->size & ~(align - 1)) != s->size)
302 if (s->size - size >= sizeof(void *))
305 if (IS_ENABLED(CONFIG_SLAB) && align &&
306 (align > s->align || s->align % align))
315 * Figure out what the alignment of the objects will be given a set of
316 * flags, a user specified alignment and the size of the objects.
318 unsigned long calculate_alignment(unsigned long flags,
319 unsigned long align, unsigned long size)
322 * If the user wants hardware cache aligned objects then follow that
323 * suggestion if the object is sufficiently large.
325 * The hardware cache alignment cannot override the specified
326 * alignment though. If that is greater then use it.
328 if (flags & SLAB_HWCACHE_ALIGN) {
329 unsigned long ralign = cache_line_size();
330 while (size <= ralign / 2)
332 align = max(align, ralign);
335 if (align < ARCH_SLAB_MINALIGN)
336 align = ARCH_SLAB_MINALIGN;
338 return ALIGN(align, sizeof(void *));
341 static struct kmem_cache *create_cache(const char *name,
342 size_t object_size, size_t size, size_t align,
343 unsigned long flags, void (*ctor)(void *),
344 struct mem_cgroup *memcg, struct kmem_cache *root_cache)
346 struct kmem_cache *s;
350 s = kmem_cache_zalloc(kmem_cache, GFP_KERNEL);
355 s->object_size = object_size;
360 err = init_memcg_params(s, memcg, root_cache);
364 err = __kmem_cache_create(s, flags);
369 list_add(&s->list, &slab_caches);
376 destroy_memcg_params(s);
377 kmem_cache_free(kmem_cache, s);
382 * kmem_cache_create - Create a cache.
383 * @name: A string which is used in /proc/slabinfo to identify this cache.
384 * @size: The size of objects to be created in this cache.
385 * @align: The required alignment for the objects.
387 * @ctor: A constructor for the objects.
389 * Returns a ptr to the cache on success, NULL on failure.
390 * Cannot be called within a interrupt, but can be interrupted.
391 * The @ctor is run when new pages are allocated by the cache.
395 * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5)
396 * to catch references to uninitialised memory.
398 * %SLAB_RED_ZONE - Insert `Red' zones around the allocated memory to check
399 * for buffer overruns.
401 * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware
402 * cacheline. This can be beneficial if you're counting cycles as closely
406 kmem_cache_create(const char *name, size_t size, size_t align,
407 unsigned long flags, void (*ctor)(void *))
409 struct kmem_cache *s = NULL;
410 const char *cache_name;
415 memcg_get_cache_ids();
417 mutex_lock(&slab_mutex);
419 err = kmem_cache_sanity_check(name, size);
424 /* Refuse requests with allocator specific flags */
425 if (flags & ~SLAB_FLAGS_PERMITTED) {
431 * Some allocators will constraint the set of valid flags to a subset
432 * of all flags. We expect them to define CACHE_CREATE_MASK in this
433 * case, and we'll just provide them with a sanitized version of the
436 flags &= CACHE_CREATE_MASK;
438 s = __kmem_cache_alias(name, size, align, flags, ctor);
442 cache_name = kstrdup_const(name, GFP_KERNEL);
448 s = create_cache(cache_name, size, size,
449 calculate_alignment(flags, align, size),
450 flags, ctor, NULL, NULL);
453 kfree_const(cache_name);
457 mutex_unlock(&slab_mutex);
459 memcg_put_cache_ids();
464 if (flags & SLAB_PANIC)
465 panic("kmem_cache_create: Failed to create slab '%s'. Error %d\n",
468 pr_warn("kmem_cache_create(%s) failed with error %d\n",
476 EXPORT_SYMBOL(kmem_cache_create);
478 static void slab_caches_to_rcu_destroy_workfn(struct work_struct *work)
480 LIST_HEAD(to_destroy);
481 struct kmem_cache *s, *s2;
484 * On destruction, SLAB_DESTROY_BY_RCU kmem_caches are put on the
485 * @slab_caches_to_rcu_destroy list. The slab pages are freed
486 * through RCU and and the associated kmem_cache are dereferenced
487 * while freeing the pages, so the kmem_caches should be freed only
488 * after the pending RCU operations are finished. As rcu_barrier()
489 * is a pretty slow operation, we batch all pending destructions
492 mutex_lock(&slab_mutex);
493 list_splice_init(&slab_caches_to_rcu_destroy, &to_destroy);
494 mutex_unlock(&slab_mutex);
496 if (list_empty(&to_destroy))
501 list_for_each_entry_safe(s, s2, &to_destroy, list) {
502 #ifdef SLAB_SUPPORTS_SYSFS
503 sysfs_slab_release(s);
505 slab_kmem_cache_release(s);
510 static int shutdown_cache(struct kmem_cache *s)
512 if (__kmem_cache_shutdown(s) != 0)
516 if (!is_root_cache(s))
517 unlink_memcg_cache(s);
519 if (s->flags & SLAB_DESTROY_BY_RCU) {
520 list_add_tail(&s->list, &slab_caches_to_rcu_destroy);
521 schedule_work(&slab_caches_to_rcu_destroy_work);
523 #ifdef SLAB_SUPPORTS_SYSFS
524 sysfs_slab_release(s);
526 slab_kmem_cache_release(s);
533 #if defined(CONFIG_MEMCG) && !defined(CONFIG_SLOB)
535 * memcg_create_kmem_cache - Create a cache for a memory cgroup.
536 * @memcg: The memory cgroup the new cache is for.
537 * @root_cache: The parent of the new cache.
539 * This function attempts to create a kmem cache that will serve allocation
540 * requests going from @memcg to @root_cache. The new cache inherits properties
543 void memcg_create_kmem_cache(struct mem_cgroup *memcg,
544 struct kmem_cache *root_cache)
546 static char memcg_name_buf[NAME_MAX + 1]; /* protected by slab_mutex */
547 struct cgroup_subsys_state *css = &memcg->css;
548 struct memcg_cache_array *arr;
549 struct kmem_cache *s = NULL;
556 mutex_lock(&slab_mutex);
559 * The memory cgroup could have been offlined while the cache
560 * creation work was pending.
562 if (memcg->kmem_state != KMEM_ONLINE)
565 idx = memcg_cache_id(memcg);
566 arr = rcu_dereference_protected(root_cache->memcg_params.memcg_caches,
567 lockdep_is_held(&slab_mutex));
570 * Since per-memcg caches are created asynchronously on first
571 * allocation (see memcg_kmem_get_cache()), several threads can try to
572 * create the same cache, but only one of them may succeed.
574 if (arr->entries[idx])
577 cgroup_name(css->cgroup, memcg_name_buf, sizeof(memcg_name_buf));
578 cache_name = kasprintf(GFP_KERNEL, "%s(%llu:%s)", root_cache->name,
579 css->serial_nr, memcg_name_buf);
583 s = create_cache(cache_name, root_cache->object_size,
584 root_cache->size, root_cache->align,
585 root_cache->flags & CACHE_CREATE_MASK,
586 root_cache->ctor, memcg, root_cache);
588 * If we could not create a memcg cache, do not complain, because
589 * that's not critical at all as we can always proceed with the root
597 list_add(&s->memcg_params.list, &root_cache->memcg_params.list);
600 * Since readers won't lock (see cache_from_memcg_idx()), we need a
601 * barrier here to ensure nobody will see the kmem_cache partially
605 arr->entries[idx] = s;
608 mutex_unlock(&slab_mutex);
614 void memcg_deactivate_kmem_caches(struct mem_cgroup *memcg)
617 struct memcg_cache_array *arr;
618 struct kmem_cache *s, *c;
620 idx = memcg_cache_id(memcg);
625 mutex_lock(&slab_mutex);
626 list_for_each_entry(s, &slab_caches, list) {
627 if (!is_root_cache(s))
630 arr = rcu_dereference_protected(s->memcg_params.memcg_caches,
631 lockdep_is_held(&slab_mutex));
632 c = arr->entries[idx];
636 __kmem_cache_shrink(c, true);
637 arr->entries[idx] = NULL;
639 mutex_unlock(&slab_mutex);
645 void memcg_destroy_kmem_caches(struct mem_cgroup *memcg)
647 struct kmem_cache *s, *s2;
652 mutex_lock(&slab_mutex);
653 list_for_each_entry_safe(s, s2, &slab_caches, list) {
654 if (is_root_cache(s) || s->memcg_params.memcg != memcg)
657 * The cgroup is about to be freed and therefore has no charges
658 * left. Hence, all its caches must be empty by now.
660 BUG_ON(shutdown_cache(s));
662 mutex_unlock(&slab_mutex);
668 static int shutdown_memcg_caches(struct kmem_cache *s)
670 struct memcg_cache_array *arr;
671 struct kmem_cache *c, *c2;
675 BUG_ON(!is_root_cache(s));
678 * First, shutdown active caches, i.e. caches that belong to online
681 arr = rcu_dereference_protected(s->memcg_params.memcg_caches,
682 lockdep_is_held(&slab_mutex));
683 for_each_memcg_cache_index(i) {
687 if (shutdown_cache(c))
689 * The cache still has objects. Move it to a temporary
690 * list so as not to try to destroy it for a second
691 * time while iterating over inactive caches below.
693 list_move(&c->memcg_params.list, &busy);
696 * The cache is empty and will be destroyed soon. Clear
697 * the pointer to it in the memcg_caches array so that
698 * it will never be accessed even if the root cache
701 arr->entries[i] = NULL;
705 * Second, shutdown all caches left from memory cgroups that are now
708 list_for_each_entry_safe(c, c2, &s->memcg_params.list,
712 list_splice(&busy, &s->memcg_params.list);
715 * A cache being destroyed must be empty. In particular, this means
716 * that all per memcg caches attached to it must be empty too.
718 if (!list_empty(&s->memcg_params.list))
723 static inline int shutdown_memcg_caches(struct kmem_cache *s)
727 #endif /* CONFIG_MEMCG && !CONFIG_SLOB */
729 void slab_kmem_cache_release(struct kmem_cache *s)
731 __kmem_cache_release(s);
732 destroy_memcg_params(s);
733 kfree_const(s->name);
734 kmem_cache_free(kmem_cache, s);
737 void kmem_cache_destroy(struct kmem_cache *s)
747 kasan_cache_destroy(s);
748 mutex_lock(&slab_mutex);
754 err = shutdown_memcg_caches(s);
756 err = shutdown_cache(s);
759 pr_err("kmem_cache_destroy %s: Slab cache still has objects\n",
764 mutex_unlock(&slab_mutex);
769 EXPORT_SYMBOL(kmem_cache_destroy);
772 * kmem_cache_shrink - Shrink a cache.
773 * @cachep: The cache to shrink.
775 * Releases as many slabs as possible for a cache.
776 * To help debugging, a zero exit status indicates all slabs were released.
778 int kmem_cache_shrink(struct kmem_cache *cachep)
784 kasan_cache_shrink(cachep);
785 ret = __kmem_cache_shrink(cachep, false);
790 EXPORT_SYMBOL(kmem_cache_shrink);
792 bool slab_is_available(void)
794 return slab_state >= UP;
798 /* Create a cache during boot when no slab services are available yet */
799 void __init create_boot_cache(struct kmem_cache *s, const char *name, size_t size,
805 s->size = s->object_size = size;
806 s->align = calculate_alignment(flags, ARCH_KMALLOC_MINALIGN, size);
808 slab_init_memcg_params(s);
810 err = __kmem_cache_create(s, flags);
813 panic("Creation of kmalloc slab %s size=%zu failed. Reason %d\n",
816 s->refcount = -1; /* Exempt from merging for now */
819 struct kmem_cache *__init create_kmalloc_cache(const char *name, size_t size,
822 struct kmem_cache *s = kmem_cache_zalloc(kmem_cache, GFP_NOWAIT);
825 panic("Out of memory when creating slab %s\n", name);
827 create_boot_cache(s, name, size, flags);
828 list_add(&s->list, &slab_caches);
833 struct kmem_cache *kmalloc_caches[KMALLOC_SHIFT_HIGH + 1];
834 EXPORT_SYMBOL(kmalloc_caches);
836 #ifdef CONFIG_ZONE_DMA
837 struct kmem_cache *kmalloc_dma_caches[KMALLOC_SHIFT_HIGH + 1];
838 EXPORT_SYMBOL(kmalloc_dma_caches);
842 * Conversion table for small slabs sizes / 8 to the index in the
843 * kmalloc array. This is necessary for slabs < 192 since we have non power
844 * of two cache sizes there. The size of larger slabs can be determined using
847 static s8 size_index[24] = {
874 static inline int size_index_elem(size_t bytes)
876 return (bytes - 1) / 8;
880 * Find the kmem_cache structure that serves a given size of
883 struct kmem_cache *kmalloc_slab(size_t size, gfp_t flags)
887 if (unlikely(size > KMALLOC_MAX_SIZE)) {
888 WARN_ON_ONCE(!(flags & __GFP_NOWARN));
894 return ZERO_SIZE_PTR;
896 index = size_index[size_index_elem(size)];
898 index = fls(size - 1);
900 #ifdef CONFIG_ZONE_DMA
901 if (unlikely((flags & GFP_DMA)))
902 return kmalloc_dma_caches[index];
905 return kmalloc_caches[index];
909 * kmalloc_info[] is to make slub_debug=,kmalloc-xx option work at boot time.
910 * kmalloc_index() supports up to 2^26=64MB, so the final entry of the table is
913 const struct kmalloc_info_struct kmalloc_info[] __initconst = {
914 {NULL, 0}, {"kmalloc-96", 96},
915 {"kmalloc-192", 192}, {"kmalloc-8", 8},
916 {"kmalloc-16", 16}, {"kmalloc-32", 32},
917 {"kmalloc-64", 64}, {"kmalloc-128", 128},
918 {"kmalloc-256", 256}, {"kmalloc-512", 512},
919 {"kmalloc-1024", 1024}, {"kmalloc-2048", 2048},
920 {"kmalloc-4096", 4096}, {"kmalloc-8192", 8192},
921 {"kmalloc-16384", 16384}, {"kmalloc-32768", 32768},
922 {"kmalloc-65536", 65536}, {"kmalloc-131072", 131072},
923 {"kmalloc-262144", 262144}, {"kmalloc-524288", 524288},
924 {"kmalloc-1048576", 1048576}, {"kmalloc-2097152", 2097152},
925 {"kmalloc-4194304", 4194304}, {"kmalloc-8388608", 8388608},
926 {"kmalloc-16777216", 16777216}, {"kmalloc-33554432", 33554432},
927 {"kmalloc-67108864", 67108864}
931 * Patch up the size_index table if we have strange large alignment
932 * requirements for the kmalloc array. This is only the case for
933 * MIPS it seems. The standard arches will not generate any code here.
935 * Largest permitted alignment is 256 bytes due to the way we
936 * handle the index determination for the smaller caches.
938 * Make sure that nothing crazy happens if someone starts tinkering
939 * around with ARCH_KMALLOC_MINALIGN
941 void __init setup_kmalloc_cache_index_table(void)
945 BUILD_BUG_ON(KMALLOC_MIN_SIZE > 256 ||
946 (KMALLOC_MIN_SIZE & (KMALLOC_MIN_SIZE - 1)));
948 for (i = 8; i < KMALLOC_MIN_SIZE; i += 8) {
949 int elem = size_index_elem(i);
951 if (elem >= ARRAY_SIZE(size_index))
953 size_index[elem] = KMALLOC_SHIFT_LOW;
956 if (KMALLOC_MIN_SIZE >= 64) {
958 * The 96 byte size cache is not used if the alignment
961 for (i = 64 + 8; i <= 96; i += 8)
962 size_index[size_index_elem(i)] = 7;
966 if (KMALLOC_MIN_SIZE >= 128) {
968 * The 192 byte sized cache is not used if the alignment
969 * is 128 byte. Redirect kmalloc to use the 256 byte cache
972 for (i = 128 + 8; i <= 192; i += 8)
973 size_index[size_index_elem(i)] = 8;
977 static void __init new_kmalloc_cache(int idx, unsigned long flags)
979 kmalloc_caches[idx] = create_kmalloc_cache(kmalloc_info[idx].name,
980 kmalloc_info[idx].size, flags);
984 * Create the kmalloc array. Some of the regular kmalloc arrays
985 * may already have been created because they were needed to
986 * enable allocations for slab creation.
988 void __init create_kmalloc_caches(unsigned long flags)
992 for (i = KMALLOC_SHIFT_LOW; i <= KMALLOC_SHIFT_HIGH; i++) {
993 if (!kmalloc_caches[i])
994 new_kmalloc_cache(i, flags);
997 * Caches that are not of the two-to-the-power-of size.
998 * These have to be created immediately after the
999 * earlier power of two caches
1001 if (KMALLOC_MIN_SIZE <= 32 && !kmalloc_caches[1] && i == 6)
1002 new_kmalloc_cache(1, flags);
1003 if (KMALLOC_MIN_SIZE <= 64 && !kmalloc_caches[2] && i == 7)
1004 new_kmalloc_cache(2, flags);
1007 /* Kmalloc array is now usable */
1010 #ifdef CONFIG_ZONE_DMA
1011 for (i = 0; i <= KMALLOC_SHIFT_HIGH; i++) {
1012 struct kmem_cache *s = kmalloc_caches[i];
1015 int size = kmalloc_size(i);
1016 char *n = kasprintf(GFP_NOWAIT,
1017 "dma-kmalloc-%d", size);
1020 kmalloc_dma_caches[i] = create_kmalloc_cache(n,
1021 size, SLAB_CACHE_DMA | flags);
1026 #endif /* !CONFIG_SLOB */
1029 * To avoid unnecessary overhead, we pass through large allocation requests
1030 * directly to the page allocator. We use __GFP_COMP, because we will need to
1031 * know the allocation order to free the pages properly in kfree.
1033 void *kmalloc_order(size_t size, gfp_t flags, unsigned int order)
1038 flags |= __GFP_COMP;
1039 page = alloc_pages(flags, order);
1040 ret = page ? page_address(page) : NULL;
1041 kmemleak_alloc(ret, size, 1, flags);
1042 kasan_kmalloc_large(ret, size, flags);
1045 EXPORT_SYMBOL(kmalloc_order);
1047 #ifdef CONFIG_TRACING
1048 void *kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order)
1050 void *ret = kmalloc_order(size, flags, order);
1051 trace_kmalloc(_RET_IP_, ret, size, PAGE_SIZE << order, flags);
1054 EXPORT_SYMBOL(kmalloc_order_trace);
1057 #ifdef CONFIG_SLAB_FREELIST_RANDOM
1058 /* Randomize a generic freelist */
1059 static void freelist_randomize(struct rnd_state *state, unsigned int *list,
1065 for (i = 0; i < count; i++)
1068 /* Fisher-Yates shuffle */
1069 for (i = count - 1; i > 0; i--) {
1070 rand = prandom_u32_state(state);
1072 swap(list[i], list[rand]);
1076 /* Create a random sequence per cache */
1077 int cache_random_seq_create(struct kmem_cache *cachep, unsigned int count,
1080 struct rnd_state state;
1082 if (count < 2 || cachep->random_seq)
1085 cachep->random_seq = kcalloc(count, sizeof(unsigned int), gfp);
1086 if (!cachep->random_seq)
1089 /* Get best entropy at this stage of boot */
1090 prandom_seed_state(&state, get_random_long());
1092 freelist_randomize(&state, cachep->random_seq, count);
1096 /* Destroy the per-cache random freelist sequence */
1097 void cache_random_seq_destroy(struct kmem_cache *cachep)
1099 kfree(cachep->random_seq);
1100 cachep->random_seq = NULL;
1102 #endif /* CONFIG_SLAB_FREELIST_RANDOM */
1104 #ifdef CONFIG_SLABINFO
1107 #define SLABINFO_RIGHTS (S_IWUSR | S_IRUSR)
1109 #define SLABINFO_RIGHTS S_IRUSR
1112 static void print_slabinfo_header(struct seq_file *m)
1115 * Output format version, so at least we can change it
1116 * without _too_ many complaints.
1118 #ifdef CONFIG_DEBUG_SLAB
1119 seq_puts(m, "slabinfo - version: 2.1 (statistics)\n");
1121 seq_puts(m, "slabinfo - version: 2.1\n");
1123 seq_puts(m, "# name <active_objs> <num_objs> <objsize> <objperslab> <pagesperslab>");
1124 seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>");
1125 seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>");
1126 #ifdef CONFIG_DEBUG_SLAB
1127 seq_puts(m, " : globalstat <listallocs> <maxobjs> <grown> <reaped> <error> <maxfreeable> <nodeallocs> <remotefrees> <alienoverflow>");
1128 seq_puts(m, " : cpustat <allochit> <allocmiss> <freehit> <freemiss>");
1133 void *slab_start(struct seq_file *m, loff_t *pos)
1135 mutex_lock(&slab_mutex);
1136 return seq_list_start(&slab_caches, *pos);
1139 void *slab_next(struct seq_file *m, void *p, loff_t *pos)
1141 return seq_list_next(p, &slab_caches, pos);
1144 void slab_stop(struct seq_file *m, void *p)
1146 mutex_unlock(&slab_mutex);
1150 memcg_accumulate_slabinfo(struct kmem_cache *s, struct slabinfo *info)
1152 struct kmem_cache *c;
1153 struct slabinfo sinfo;
1155 if (!is_root_cache(s))
1158 for_each_memcg_cache(c, s) {
1159 memset(&sinfo, 0, sizeof(sinfo));
1160 get_slabinfo(c, &sinfo);
1162 info->active_slabs += sinfo.active_slabs;
1163 info->num_slabs += sinfo.num_slabs;
1164 info->shared_avail += sinfo.shared_avail;
1165 info->active_objs += sinfo.active_objs;
1166 info->num_objs += sinfo.num_objs;
1170 static void cache_show(struct kmem_cache *s, struct seq_file *m)
1172 struct slabinfo sinfo;
1174 memset(&sinfo, 0, sizeof(sinfo));
1175 get_slabinfo(s, &sinfo);
1177 memcg_accumulate_slabinfo(s, &sinfo);
1179 seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d",
1180 cache_name(s), sinfo.active_objs, sinfo.num_objs, s->size,
1181 sinfo.objects_per_slab, (1 << sinfo.cache_order));
1183 seq_printf(m, " : tunables %4u %4u %4u",
1184 sinfo.limit, sinfo.batchcount, sinfo.shared);
1185 seq_printf(m, " : slabdata %6lu %6lu %6lu",
1186 sinfo.active_slabs, sinfo.num_slabs, sinfo.shared_avail);
1187 slabinfo_show_stats(m, s);
1191 static int slab_show(struct seq_file *m, void *p)
1193 struct kmem_cache *s = list_entry(p, struct kmem_cache, list);
1195 if (p == slab_caches.next)
1196 print_slabinfo_header(m);
1197 if (is_root_cache(s))
1202 #if defined(CONFIG_MEMCG) && !defined(CONFIG_SLOB)
1203 int memcg_slab_show(struct seq_file *m, void *p)
1205 struct kmem_cache *s = list_entry(p, struct kmem_cache, list);
1206 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
1208 if (p == slab_caches.next)
1209 print_slabinfo_header(m);
1210 if (!is_root_cache(s) && s->memcg_params.memcg == memcg)
1217 * slabinfo_op - iterator that generates /proc/slabinfo
1226 * num-pages-per-slab
1227 * + further values on SMP and with statistics enabled
1229 static const struct seq_operations slabinfo_op = {
1230 .start = slab_start,
1236 static int slabinfo_open(struct inode *inode, struct file *file)
1238 return seq_open(file, &slabinfo_op);
1241 static const struct file_operations proc_slabinfo_operations = {
1242 .open = slabinfo_open,
1244 .write = slabinfo_write,
1245 .llseek = seq_lseek,
1246 .release = seq_release,
1249 static int __init slab_proc_init(void)
1251 proc_create("slabinfo", SLABINFO_RIGHTS, NULL,
1252 &proc_slabinfo_operations);
1255 module_init(slab_proc_init);
1256 #endif /* CONFIG_SLABINFO */
1258 static __always_inline void *__do_krealloc(const void *p, size_t new_size,
1267 if (ks >= new_size) {
1268 kasan_krealloc((void *)p, new_size, flags);
1272 ret = kmalloc_track_caller(new_size, flags);
1280 * __krealloc - like krealloc() but don't free @p.
1281 * @p: object to reallocate memory for.
1282 * @new_size: how many bytes of memory are required.
1283 * @flags: the type of memory to allocate.
1285 * This function is like krealloc() except it never frees the originally
1286 * allocated buffer. Use this if you don't want to free the buffer immediately
1287 * like, for example, with RCU.
1289 void *__krealloc(const void *p, size_t new_size, gfp_t flags)
1291 if (unlikely(!new_size))
1292 return ZERO_SIZE_PTR;
1294 return __do_krealloc(p, new_size, flags);
1297 EXPORT_SYMBOL(__krealloc);
1300 * krealloc - reallocate memory. The contents will remain unchanged.
1301 * @p: object to reallocate memory for.
1302 * @new_size: how many bytes of memory are required.
1303 * @flags: the type of memory to allocate.
1305 * The contents of the object pointed to are preserved up to the
1306 * lesser of the new and old sizes. If @p is %NULL, krealloc()
1307 * behaves exactly like kmalloc(). If @new_size is 0 and @p is not a
1308 * %NULL pointer, the object pointed to is freed.
1310 void *krealloc(const void *p, size_t new_size, gfp_t flags)
1314 if (unlikely(!new_size)) {
1316 return ZERO_SIZE_PTR;
1319 ret = __do_krealloc(p, new_size, flags);
1320 if (ret && p != ret)
1325 EXPORT_SYMBOL(krealloc);
1328 * kzfree - like kfree but zero memory
1329 * @p: object to free memory of
1331 * The memory of the object @p points to is zeroed before freed.
1332 * If @p is %NULL, kzfree() does nothing.
1334 * Note: this function zeroes the whole allocated buffer which can be a good
1335 * deal bigger than the requested buffer size passed to kmalloc(). So be
1336 * careful when using this function in performance sensitive code.
1338 void kzfree(const void *p)
1341 void *mem = (void *)p;
1343 if (unlikely(ZERO_OR_NULL_PTR(mem)))
1349 EXPORT_SYMBOL(kzfree);
1351 /* Tracepoints definitions. */
1352 EXPORT_TRACEPOINT_SYMBOL(kmalloc);
1353 EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc);
1354 EXPORT_TRACEPOINT_SYMBOL(kmalloc_node);
1355 EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc_node);
1356 EXPORT_TRACEPOINT_SYMBOL(kfree);
1357 EXPORT_TRACEPOINT_SYMBOL(kmem_cache_free);