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;
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 if (flags & SLAB_NEVER_MERGE)
272 list_for_each_entry_reverse(s, &slab_caches, list) {
273 if (slab_unmergeable(s))
279 if ((flags & SLAB_MERGE_SAME) != (s->flags & SLAB_MERGE_SAME))
282 * Check if alignment is compatible.
283 * Courtesy of Adrian Drzewiecki
285 if ((s->size & ~(align - 1)) != s->size)
288 if (s->size - size >= sizeof(void *))
291 if (IS_ENABLED(CONFIG_SLAB) && align &&
292 (align > s->align || s->align % align))
301 * Figure out what the alignment of the objects will be given a set of
302 * flags, a user specified alignment and the size of the objects.
304 unsigned long calculate_alignment(unsigned long flags,
305 unsigned long align, unsigned long size)
308 * If the user wants hardware cache aligned objects then follow that
309 * suggestion if the object is sufficiently large.
311 * The hardware cache alignment cannot override the specified
312 * alignment though. If that is greater then use it.
314 if (flags & SLAB_HWCACHE_ALIGN) {
315 unsigned long ralign = cache_line_size();
316 while (size <= ralign / 2)
318 align = max(align, ralign);
321 if (align < ARCH_SLAB_MINALIGN)
322 align = ARCH_SLAB_MINALIGN;
324 return ALIGN(align, sizeof(void *));
327 static struct kmem_cache *create_cache(const char *name,
328 size_t object_size, size_t size, size_t align,
329 unsigned long flags, void (*ctor)(void *),
330 struct mem_cgroup *memcg, struct kmem_cache *root_cache)
332 struct kmem_cache *s;
336 s = kmem_cache_zalloc(kmem_cache, GFP_KERNEL);
341 s->object_size = object_size;
346 err = init_memcg_params(s, memcg, root_cache);
350 err = __kmem_cache_create(s, flags);
355 list_add(&s->list, &slab_caches);
362 destroy_memcg_params(s);
363 kmem_cache_free(kmem_cache, s);
368 * kmem_cache_create - Create a cache.
369 * @name: A string which is used in /proc/slabinfo to identify this cache.
370 * @size: The size of objects to be created in this cache.
371 * @align: The required alignment for the objects.
373 * @ctor: A constructor for the objects.
375 * Returns a ptr to the cache on success, NULL on failure.
376 * Cannot be called within a interrupt, but can be interrupted.
377 * The @ctor is run when new pages are allocated by the cache.
381 * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5)
382 * to catch references to uninitialised memory.
384 * %SLAB_RED_ZONE - Insert `Red' zones around the allocated memory to check
385 * for buffer overruns.
387 * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware
388 * cacheline. This can be beneficial if you're counting cycles as closely
392 kmem_cache_create(const char *name, size_t size, size_t align,
393 unsigned long flags, void (*ctor)(void *))
395 struct kmem_cache *s = NULL;
396 const char *cache_name;
401 memcg_get_cache_ids();
403 mutex_lock(&slab_mutex);
405 err = kmem_cache_sanity_check(name, size);
410 /* Refuse requests with allocator specific flags */
411 if (flags & ~SLAB_FLAGS_PERMITTED) {
417 * Some allocators will constraint the set of valid flags to a subset
418 * of all flags. We expect them to define CACHE_CREATE_MASK in this
419 * case, and we'll just provide them with a sanitized version of the
422 flags &= CACHE_CREATE_MASK;
424 s = __kmem_cache_alias(name, size, align, flags, ctor);
428 cache_name = kstrdup_const(name, GFP_KERNEL);
434 s = create_cache(cache_name, size, size,
435 calculate_alignment(flags, align, size),
436 flags, ctor, NULL, NULL);
439 kfree_const(cache_name);
443 mutex_unlock(&slab_mutex);
445 memcg_put_cache_ids();
450 if (flags & SLAB_PANIC)
451 panic("kmem_cache_create: Failed to create slab '%s'. Error %d\n",
454 pr_warn("kmem_cache_create(%s) failed with error %d\n",
462 EXPORT_SYMBOL(kmem_cache_create);
464 static int shutdown_cache(struct kmem_cache *s,
465 struct list_head *release, bool *need_rcu_barrier)
467 if (__kmem_cache_shutdown(s) != 0)
470 if (s->flags & SLAB_DESTROY_BY_RCU)
471 *need_rcu_barrier = true;
473 list_move(&s->list, release);
477 static void release_caches(struct list_head *release, bool need_rcu_barrier)
479 struct kmem_cache *s, *s2;
481 if (need_rcu_barrier)
484 list_for_each_entry_safe(s, s2, release, list) {
485 #ifdef SLAB_SUPPORTS_SYSFS
486 sysfs_slab_release(s);
488 slab_kmem_cache_release(s);
493 #if defined(CONFIG_MEMCG) && !defined(CONFIG_SLOB)
495 * memcg_create_kmem_cache - Create a cache for a memory cgroup.
496 * @memcg: The memory cgroup the new cache is for.
497 * @root_cache: The parent of the new cache.
499 * This function attempts to create a kmem cache that will serve allocation
500 * requests going from @memcg to @root_cache. The new cache inherits properties
503 void memcg_create_kmem_cache(struct mem_cgroup *memcg,
504 struct kmem_cache *root_cache)
506 static char memcg_name_buf[NAME_MAX + 1]; /* protected by slab_mutex */
507 struct cgroup_subsys_state *css = &memcg->css;
508 struct memcg_cache_array *arr;
509 struct kmem_cache *s = NULL;
516 mutex_lock(&slab_mutex);
519 * The memory cgroup could have been offlined while the cache
520 * creation work was pending.
522 if (memcg->kmem_state != KMEM_ONLINE)
525 idx = memcg_cache_id(memcg);
526 arr = rcu_dereference_protected(root_cache->memcg_params.memcg_caches,
527 lockdep_is_held(&slab_mutex));
530 * Since per-memcg caches are created asynchronously on first
531 * allocation (see memcg_kmem_get_cache()), several threads can try to
532 * create the same cache, but only one of them may succeed.
534 if (arr->entries[idx])
537 cgroup_name(css->cgroup, memcg_name_buf, sizeof(memcg_name_buf));
538 cache_name = kasprintf(GFP_KERNEL, "%s(%llu:%s)", root_cache->name,
539 css->serial_nr, memcg_name_buf);
543 s = create_cache(cache_name, root_cache->object_size,
544 root_cache->size, root_cache->align,
545 root_cache->flags & CACHE_CREATE_MASK,
546 root_cache->ctor, memcg, root_cache);
548 * If we could not create a memcg cache, do not complain, because
549 * that's not critical at all as we can always proceed with the root
557 list_add(&s->memcg_params.list, &root_cache->memcg_params.list);
560 * Since readers won't lock (see cache_from_memcg_idx()), we need a
561 * barrier here to ensure nobody will see the kmem_cache partially
565 arr->entries[idx] = s;
568 mutex_unlock(&slab_mutex);
574 void memcg_deactivate_kmem_caches(struct mem_cgroup *memcg)
577 struct memcg_cache_array *arr;
578 struct kmem_cache *s, *c;
580 idx = memcg_cache_id(memcg);
585 mutex_lock(&slab_mutex);
586 list_for_each_entry(s, &slab_caches, list) {
587 if (!is_root_cache(s))
590 arr = rcu_dereference_protected(s->memcg_params.memcg_caches,
591 lockdep_is_held(&slab_mutex));
592 c = arr->entries[idx];
596 __kmem_cache_shrink(c, true);
597 arr->entries[idx] = NULL;
599 mutex_unlock(&slab_mutex);
605 static int __shutdown_memcg_cache(struct kmem_cache *s,
606 struct list_head *release, bool *need_rcu_barrier)
608 BUG_ON(is_root_cache(s));
610 if (shutdown_cache(s, release, need_rcu_barrier))
613 list_del(&s->memcg_params.list);
617 void memcg_destroy_kmem_caches(struct mem_cgroup *memcg)
620 bool need_rcu_barrier = false;
621 struct kmem_cache *s, *s2;
626 mutex_lock(&slab_mutex);
627 list_for_each_entry_safe(s, s2, &slab_caches, list) {
628 if (is_root_cache(s) || s->memcg_params.memcg != memcg)
631 * The cgroup is about to be freed and therefore has no charges
632 * left. Hence, all its caches must be empty by now.
634 BUG_ON(__shutdown_memcg_cache(s, &release, &need_rcu_barrier));
636 mutex_unlock(&slab_mutex);
641 release_caches(&release, need_rcu_barrier);
644 static int shutdown_memcg_caches(struct kmem_cache *s,
645 struct list_head *release, bool *need_rcu_barrier)
647 struct memcg_cache_array *arr;
648 struct kmem_cache *c, *c2;
652 BUG_ON(!is_root_cache(s));
655 * First, shutdown active caches, i.e. caches that belong to online
658 arr = rcu_dereference_protected(s->memcg_params.memcg_caches,
659 lockdep_is_held(&slab_mutex));
660 for_each_memcg_cache_index(i) {
664 if (__shutdown_memcg_cache(c, release, need_rcu_barrier))
666 * The cache still has objects. Move it to a temporary
667 * list so as not to try to destroy it for a second
668 * time while iterating over inactive caches below.
670 list_move(&c->memcg_params.list, &busy);
673 * The cache is empty and will be destroyed soon. Clear
674 * the pointer to it in the memcg_caches array so that
675 * it will never be accessed even if the root cache
678 arr->entries[i] = NULL;
682 * Second, shutdown all caches left from memory cgroups that are now
685 list_for_each_entry_safe(c, c2, &s->memcg_params.list,
687 __shutdown_memcg_cache(c, release, need_rcu_barrier);
689 list_splice(&busy, &s->memcg_params.list);
692 * A cache being destroyed must be empty. In particular, this means
693 * that all per memcg caches attached to it must be empty too.
695 if (!list_empty(&s->memcg_params.list))
700 static inline int shutdown_memcg_caches(struct kmem_cache *s,
701 struct list_head *release, bool *need_rcu_barrier)
705 #endif /* CONFIG_MEMCG && !CONFIG_SLOB */
707 void slab_kmem_cache_release(struct kmem_cache *s)
709 __kmem_cache_release(s);
710 destroy_memcg_params(s);
711 kfree_const(s->name);
712 kmem_cache_free(kmem_cache, s);
715 void kmem_cache_destroy(struct kmem_cache *s)
718 bool need_rcu_barrier = false;
727 kasan_cache_destroy(s);
728 mutex_lock(&slab_mutex);
734 err = shutdown_memcg_caches(s, &release, &need_rcu_barrier);
736 err = shutdown_cache(s, &release, &need_rcu_barrier);
739 pr_err("kmem_cache_destroy %s: Slab cache still has objects\n",
744 mutex_unlock(&slab_mutex);
749 release_caches(&release, need_rcu_barrier);
751 EXPORT_SYMBOL(kmem_cache_destroy);
754 * kmem_cache_shrink - Shrink a cache.
755 * @cachep: The cache to shrink.
757 * Releases as many slabs as possible for a cache.
758 * To help debugging, a zero exit status indicates all slabs were released.
760 int kmem_cache_shrink(struct kmem_cache *cachep)
766 kasan_cache_shrink(cachep);
767 ret = __kmem_cache_shrink(cachep, false);
772 EXPORT_SYMBOL(kmem_cache_shrink);
774 bool slab_is_available(void)
776 return slab_state >= UP;
780 /* Create a cache during boot when no slab services are available yet */
781 void __init create_boot_cache(struct kmem_cache *s, const char *name, size_t size,
787 s->size = s->object_size = size;
788 s->align = calculate_alignment(flags, ARCH_KMALLOC_MINALIGN, size);
790 slab_init_memcg_params(s);
792 err = __kmem_cache_create(s, flags);
795 panic("Creation of kmalloc slab %s size=%zu failed. Reason %d\n",
798 s->refcount = -1; /* Exempt from merging for now */
801 struct kmem_cache *__init create_kmalloc_cache(const char *name, size_t size,
804 struct kmem_cache *s = kmem_cache_zalloc(kmem_cache, GFP_NOWAIT);
807 panic("Out of memory when creating slab %s\n", name);
809 create_boot_cache(s, name, size, flags);
810 list_add(&s->list, &slab_caches);
815 struct kmem_cache *kmalloc_caches[KMALLOC_SHIFT_HIGH + 1];
816 EXPORT_SYMBOL(kmalloc_caches);
818 #ifdef CONFIG_ZONE_DMA
819 struct kmem_cache *kmalloc_dma_caches[KMALLOC_SHIFT_HIGH + 1];
820 EXPORT_SYMBOL(kmalloc_dma_caches);
824 * Conversion table for small slabs sizes / 8 to the index in the
825 * kmalloc array. This is necessary for slabs < 192 since we have non power
826 * of two cache sizes there. The size of larger slabs can be determined using
829 static s8 size_index[24] = {
856 static inline int size_index_elem(size_t bytes)
858 return (bytes - 1) / 8;
862 * Find the kmem_cache structure that serves a given size of
865 struct kmem_cache *kmalloc_slab(size_t size, gfp_t flags)
869 if (unlikely(size > KMALLOC_MAX_SIZE)) {
870 WARN_ON_ONCE(!(flags & __GFP_NOWARN));
876 return ZERO_SIZE_PTR;
878 index = size_index[size_index_elem(size)];
880 index = fls(size - 1);
882 #ifdef CONFIG_ZONE_DMA
883 if (unlikely((flags & GFP_DMA)))
884 return kmalloc_dma_caches[index];
887 return kmalloc_caches[index];
891 * kmalloc_info[] is to make slub_debug=,kmalloc-xx option work at boot time.
892 * kmalloc_index() supports up to 2^26=64MB, so the final entry of the table is
895 const struct kmalloc_info_struct kmalloc_info[] __initconst = {
896 {NULL, 0}, {"kmalloc-96", 96},
897 {"kmalloc-192", 192}, {"kmalloc-8", 8},
898 {"kmalloc-16", 16}, {"kmalloc-32", 32},
899 {"kmalloc-64", 64}, {"kmalloc-128", 128},
900 {"kmalloc-256", 256}, {"kmalloc-512", 512},
901 {"kmalloc-1024", 1024}, {"kmalloc-2048", 2048},
902 {"kmalloc-4096", 4096}, {"kmalloc-8192", 8192},
903 {"kmalloc-16384", 16384}, {"kmalloc-32768", 32768},
904 {"kmalloc-65536", 65536}, {"kmalloc-131072", 131072},
905 {"kmalloc-262144", 262144}, {"kmalloc-524288", 524288},
906 {"kmalloc-1048576", 1048576}, {"kmalloc-2097152", 2097152},
907 {"kmalloc-4194304", 4194304}, {"kmalloc-8388608", 8388608},
908 {"kmalloc-16777216", 16777216}, {"kmalloc-33554432", 33554432},
909 {"kmalloc-67108864", 67108864}
913 * Patch up the size_index table if we have strange large alignment
914 * requirements for the kmalloc array. This is only the case for
915 * MIPS it seems. The standard arches will not generate any code here.
917 * Largest permitted alignment is 256 bytes due to the way we
918 * handle the index determination for the smaller caches.
920 * Make sure that nothing crazy happens if someone starts tinkering
921 * around with ARCH_KMALLOC_MINALIGN
923 void __init setup_kmalloc_cache_index_table(void)
927 BUILD_BUG_ON(KMALLOC_MIN_SIZE > 256 ||
928 (KMALLOC_MIN_SIZE & (KMALLOC_MIN_SIZE - 1)));
930 for (i = 8; i < KMALLOC_MIN_SIZE; i += 8) {
931 int elem = size_index_elem(i);
933 if (elem >= ARRAY_SIZE(size_index))
935 size_index[elem] = KMALLOC_SHIFT_LOW;
938 if (KMALLOC_MIN_SIZE >= 64) {
940 * The 96 byte size cache is not used if the alignment
943 for (i = 64 + 8; i <= 96; i += 8)
944 size_index[size_index_elem(i)] = 7;
948 if (KMALLOC_MIN_SIZE >= 128) {
950 * The 192 byte sized cache is not used if the alignment
951 * is 128 byte. Redirect kmalloc to use the 256 byte cache
954 for (i = 128 + 8; i <= 192; i += 8)
955 size_index[size_index_elem(i)] = 8;
959 static void __init new_kmalloc_cache(int idx, unsigned long flags)
961 kmalloc_caches[idx] = create_kmalloc_cache(kmalloc_info[idx].name,
962 kmalloc_info[idx].size, flags);
966 * Create the kmalloc array. Some of the regular kmalloc arrays
967 * may already have been created because they were needed to
968 * enable allocations for slab creation.
970 void __init create_kmalloc_caches(unsigned long flags)
974 for (i = KMALLOC_SHIFT_LOW; i <= KMALLOC_SHIFT_HIGH; i++) {
975 if (!kmalloc_caches[i])
976 new_kmalloc_cache(i, flags);
979 * Caches that are not of the two-to-the-power-of size.
980 * These have to be created immediately after the
981 * earlier power of two caches
983 if (KMALLOC_MIN_SIZE <= 32 && !kmalloc_caches[1] && i == 6)
984 new_kmalloc_cache(1, flags);
985 if (KMALLOC_MIN_SIZE <= 64 && !kmalloc_caches[2] && i == 7)
986 new_kmalloc_cache(2, flags);
989 /* Kmalloc array is now usable */
992 #ifdef CONFIG_ZONE_DMA
993 for (i = 0; i <= KMALLOC_SHIFT_HIGH; i++) {
994 struct kmem_cache *s = kmalloc_caches[i];
997 int size = kmalloc_size(i);
998 char *n = kasprintf(GFP_NOWAIT,
999 "dma-kmalloc-%d", size);
1002 kmalloc_dma_caches[i] = create_kmalloc_cache(n,
1003 size, SLAB_CACHE_DMA | flags);
1008 #endif /* !CONFIG_SLOB */
1011 * To avoid unnecessary overhead, we pass through large allocation requests
1012 * directly to the page allocator. We use __GFP_COMP, because we will need to
1013 * know the allocation order to free the pages properly in kfree.
1015 void *kmalloc_order(size_t size, gfp_t flags, unsigned int order)
1020 flags |= __GFP_COMP;
1021 page = alloc_pages(flags, order);
1022 ret = page ? page_address(page) : NULL;
1023 kmemleak_alloc(ret, size, 1, flags);
1024 kasan_kmalloc_large(ret, size, flags);
1027 EXPORT_SYMBOL(kmalloc_order);
1029 #ifdef CONFIG_TRACING
1030 void *kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order)
1032 void *ret = kmalloc_order(size, flags, order);
1033 trace_kmalloc(_RET_IP_, ret, size, PAGE_SIZE << order, flags);
1036 EXPORT_SYMBOL(kmalloc_order_trace);
1039 #ifdef CONFIG_SLAB_FREELIST_RANDOM
1040 /* Randomize a generic freelist */
1041 static void freelist_randomize(struct rnd_state *state, unsigned int *list,
1047 for (i = 0; i < count; i++)
1050 /* Fisher-Yates shuffle */
1051 for (i = count - 1; i > 0; i--) {
1052 rand = prandom_u32_state(state);
1054 swap(list[i], list[rand]);
1058 /* Create a random sequence per cache */
1059 int cache_random_seq_create(struct kmem_cache *cachep, unsigned int count,
1062 struct rnd_state state;
1064 if (count < 2 || cachep->random_seq)
1067 cachep->random_seq = kcalloc(count, sizeof(unsigned int), gfp);
1068 if (!cachep->random_seq)
1071 /* Get best entropy at this stage of boot */
1072 prandom_seed_state(&state, get_random_long());
1074 freelist_randomize(&state, cachep->random_seq, count);
1078 /* Destroy the per-cache random freelist sequence */
1079 void cache_random_seq_destroy(struct kmem_cache *cachep)
1081 kfree(cachep->random_seq);
1082 cachep->random_seq = NULL;
1084 #endif /* CONFIG_SLAB_FREELIST_RANDOM */
1086 #ifdef CONFIG_SLABINFO
1089 #define SLABINFO_RIGHTS (S_IWUSR | S_IRUSR)
1091 #define SLABINFO_RIGHTS S_IRUSR
1094 static void print_slabinfo_header(struct seq_file *m)
1097 * Output format version, so at least we can change it
1098 * without _too_ many complaints.
1100 #ifdef CONFIG_DEBUG_SLAB
1101 seq_puts(m, "slabinfo - version: 2.1 (statistics)\n");
1103 seq_puts(m, "slabinfo - version: 2.1\n");
1105 seq_puts(m, "# name <active_objs> <num_objs> <objsize> <objperslab> <pagesperslab>");
1106 seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>");
1107 seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>");
1108 #ifdef CONFIG_DEBUG_SLAB
1109 seq_puts(m, " : globalstat <listallocs> <maxobjs> <grown> <reaped> <error> <maxfreeable> <nodeallocs> <remotefrees> <alienoverflow>");
1110 seq_puts(m, " : cpustat <allochit> <allocmiss> <freehit> <freemiss>");
1115 void *slab_start(struct seq_file *m, loff_t *pos)
1117 mutex_lock(&slab_mutex);
1118 return seq_list_start(&slab_caches, *pos);
1121 void *slab_next(struct seq_file *m, void *p, loff_t *pos)
1123 return seq_list_next(p, &slab_caches, pos);
1126 void slab_stop(struct seq_file *m, void *p)
1128 mutex_unlock(&slab_mutex);
1132 memcg_accumulate_slabinfo(struct kmem_cache *s, struct slabinfo *info)
1134 struct kmem_cache *c;
1135 struct slabinfo sinfo;
1137 if (!is_root_cache(s))
1140 for_each_memcg_cache(c, s) {
1141 memset(&sinfo, 0, sizeof(sinfo));
1142 get_slabinfo(c, &sinfo);
1144 info->active_slabs += sinfo.active_slabs;
1145 info->num_slabs += sinfo.num_slabs;
1146 info->shared_avail += sinfo.shared_avail;
1147 info->active_objs += sinfo.active_objs;
1148 info->num_objs += sinfo.num_objs;
1152 static void cache_show(struct kmem_cache *s, struct seq_file *m)
1154 struct slabinfo sinfo;
1156 memset(&sinfo, 0, sizeof(sinfo));
1157 get_slabinfo(s, &sinfo);
1159 memcg_accumulate_slabinfo(s, &sinfo);
1161 seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d",
1162 cache_name(s), sinfo.active_objs, sinfo.num_objs, s->size,
1163 sinfo.objects_per_slab, (1 << sinfo.cache_order));
1165 seq_printf(m, " : tunables %4u %4u %4u",
1166 sinfo.limit, sinfo.batchcount, sinfo.shared);
1167 seq_printf(m, " : slabdata %6lu %6lu %6lu",
1168 sinfo.active_slabs, sinfo.num_slabs, sinfo.shared_avail);
1169 slabinfo_show_stats(m, s);
1173 static int slab_show(struct seq_file *m, void *p)
1175 struct kmem_cache *s = list_entry(p, struct kmem_cache, list);
1177 if (p == slab_caches.next)
1178 print_slabinfo_header(m);
1179 if (is_root_cache(s))
1184 #if defined(CONFIG_MEMCG) && !defined(CONFIG_SLOB)
1185 int memcg_slab_show(struct seq_file *m, void *p)
1187 struct kmem_cache *s = list_entry(p, struct kmem_cache, list);
1188 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
1190 if (p == slab_caches.next)
1191 print_slabinfo_header(m);
1192 if (!is_root_cache(s) && s->memcg_params.memcg == memcg)
1199 * slabinfo_op - iterator that generates /proc/slabinfo
1208 * num-pages-per-slab
1209 * + further values on SMP and with statistics enabled
1211 static const struct seq_operations slabinfo_op = {
1212 .start = slab_start,
1218 static int slabinfo_open(struct inode *inode, struct file *file)
1220 return seq_open(file, &slabinfo_op);
1223 static const struct file_operations proc_slabinfo_operations = {
1224 .open = slabinfo_open,
1226 .write = slabinfo_write,
1227 .llseek = seq_lseek,
1228 .release = seq_release,
1231 static int __init slab_proc_init(void)
1233 proc_create("slabinfo", SLABINFO_RIGHTS, NULL,
1234 &proc_slabinfo_operations);
1237 module_init(slab_proc_init);
1238 #endif /* CONFIG_SLABINFO */
1240 static __always_inline void *__do_krealloc(const void *p, size_t new_size,
1249 if (ks >= new_size) {
1250 kasan_krealloc((void *)p, new_size, flags);
1254 ret = kmalloc_track_caller(new_size, flags);
1262 * __krealloc - like krealloc() but don't free @p.
1263 * @p: object to reallocate memory for.
1264 * @new_size: how many bytes of memory are required.
1265 * @flags: the type of memory to allocate.
1267 * This function is like krealloc() except it never frees the originally
1268 * allocated buffer. Use this if you don't want to free the buffer immediately
1269 * like, for example, with RCU.
1271 void *__krealloc(const void *p, size_t new_size, gfp_t flags)
1273 if (unlikely(!new_size))
1274 return ZERO_SIZE_PTR;
1276 return __do_krealloc(p, new_size, flags);
1279 EXPORT_SYMBOL(__krealloc);
1282 * krealloc - reallocate memory. The contents will remain unchanged.
1283 * @p: object to reallocate memory for.
1284 * @new_size: how many bytes of memory are required.
1285 * @flags: the type of memory to allocate.
1287 * The contents of the object pointed to are preserved up to the
1288 * lesser of the new and old sizes. If @p is %NULL, krealloc()
1289 * behaves exactly like kmalloc(). If @new_size is 0 and @p is not a
1290 * %NULL pointer, the object pointed to is freed.
1292 void *krealloc(const void *p, size_t new_size, gfp_t flags)
1296 if (unlikely(!new_size)) {
1298 return ZERO_SIZE_PTR;
1301 ret = __do_krealloc(p, new_size, flags);
1302 if (ret && p != ret)
1307 EXPORT_SYMBOL(krealloc);
1310 * kzfree - like kfree but zero memory
1311 * @p: object to free memory of
1313 * The memory of the object @p points to is zeroed before freed.
1314 * If @p is %NULL, kzfree() does nothing.
1316 * Note: this function zeroes the whole allocated buffer which can be a good
1317 * deal bigger than the requested buffer size passed to kmalloc(). So be
1318 * careful when using this function in performance sensitive code.
1320 void kzfree(const void *p)
1323 void *mem = (void *)p;
1325 if (unlikely(ZERO_OR_NULL_PTR(mem)))
1331 EXPORT_SYMBOL(kzfree);
1333 /* Tracepoints definitions. */
1334 EXPORT_TRACEPOINT_SYMBOL(kmalloc);
1335 EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc);
1336 EXPORT_TRACEPOINT_SYMBOL(kmalloc_node);
1337 EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc_node);
1338 EXPORT_TRACEPOINT_SYMBOL(kfree);
1339 EXPORT_TRACEPOINT_SYMBOL(kmem_cache_free);